Tech Guides - Artificial Intelligence

TensorFlow is an open source machine learning framework for carrying out high-performance numerical computations. It provides excellent architecture support which allows easy deployment of computations across a variety of platforms ranging from desktops to clusters of servers, mobiles, and edge devices. Have you ever thought, why TensorFlow has become so popular in such a short span of time? What made TensorFlow so special, that we seeing a huge surge of developers and researchers opting for the TensorFlow framework? Interestingly, when it comes to artificial intelligence frameworks showdown, you will find TensorFlow emerging as a clear winner most of the time. The major credit goes to the soaring popularity and contributions across various forums such as GitHub, Stack Overflow, and Quora. The fact is, TensorFlow is being used in over 6000 open source repositories showing their roots in many real-world research and applications. How TensorFlow came to be The library was developed by a group of researchers and engineers from the Google Brain team within Google AI organization. They wanted a library that provides strong support for machine learning and deep learning and advanced numerical computations across different scientific domains. Since the time Google open sourced its machine learning framework in 2015, TensorFlow has grown in popularity with more than 1500 projects mentions on GitHub. The constant updates made to the TensorFlow ecosystem is the real cherry on the cake. This has ensured all the new challenges developers and researchers face are addressed, thus easing the complex computations and providing newer features, promises, and performance improvements with the support of high-level APIs. By open sourcing the library, the Google research team have received all the benefits from a huge set of contributors outside their existing core team. Their idea was to make TensorFlow popular by open sourcing it, thus making sure all new research ideas are implemented in TensorFlow first allowing Google to productize those ideas. Read Also: 6 reasons why Google open sourced TensorFlow What makes TensorFlow different from the rest? With more and more research and real-life use cases going mainstream, we can see a big trend among programmers, and developers flocking towards the tool called TensorFlow. The popularity for TensorFlow is quite evident, with big names adopting TensorFlow for carrying out artificial intelligence tasks. Many popular companies such as NVIDIA, Twitter, Snapchat, Uber and more are using TensorFlow for all their major operations and research areas. On one hand, someone can make a case that TensorFlow’s popularity is based on its origins/legacy. TensorFlow being developed under the house of “Google” enjoys the reputation of the household name. There’s no doubt, TensorFlow has been better marketed than some of its competitors. Source: The Data Incubator However that’s not the full story. There are many other compelling reasons why small scale to large scale companies prefer using TensorFlow over other machine learning tools TensorFlow key functionalities TensorFlow provides an accessible and readable syntax which is essential for making these programming resources easier to use. The complex syntax is the last thing developers need to know given machine learning’s advanced nature. TensorFlow provides excellent functionalities and services when compared to other popular deep learning frameworks. These high-level operations are essential for carrying out complex parallel computations and for building advanced neural network models. TensorFlow is a low-level library which provides more flexibility. Thus you can define your own functionalities or services for your models. This is a very important parameter for researchers because it allows them to change the model based on changing user requirements. TensorFlow provides more network control. Thus allowing developers and researchers to understand how operations are implemented across the network. They can always keep track of new changes done over time. Distributed training The trend of distributed deep learning began in 2017, when Facebook released a paper showing a set of methods to reduce the training time of a convolutional neural network model. The test was done on RESNET-50 model on ImageNet dataset which took one hour to train instead of two weeks. 256 GPUs spread over 32 servers were used. This revolutionary test has open the gates for many research work which have massively reduced the experimentation time by running many tasks in parallel on multiple GPUs. Google’s distributed TensorFlow has allowed all the researchers and developers to scale out complex distributed training using in-built methods and operations that optimizes distributed deep learning among servers. . Google’s distributed TensorFlow engine which is part of the regular TensorFlow repo, works exceptionally well with the existing TensorFlow’s operations and functionalities. It has allowed exploring two of the most important distributed methods: Distribute the training time of a neural network model over many servers to reduce the training time. Searching for good hyperparameters by running parallel experiments over multiple servers. Google has given distributed TensorFlow engine the required power to steal the share of the market acquired by other distributed projects such as Microsoft’s CNTK, AMPLab's SparkNet, and CaffeOnSpark. Even though the competition is tough, Google has still managed to become more popular when compared to the other alternatives in the market. From research to production Google has, in some ways, democratized deep learning., The key reason is TensorFlow’s high-level APIs making deep learning accessible to everyone. TensorFlow provides pre-built functions and advanced operations to ease the task of building different neural network models. It provides the required infrastructure and hardware which makes them one of the leading libraries used extensively by researchers and students in the deep learning domain. In addition to research tools, TensorFlow extends the services by bringing the model in production using TensorFlow Serving. It is specifically designed for production environments, which provides a flexible, high-performance serving system for machine learning models. It provides all the functionalities and operations which makes it easy to deploy new algorithms and experiments as per changing requirements and preferences. It provides an excellent feature of out-of-the-box integration with TensorFlow models which can be easily extended to serve other types of models and data. TensorFlow’s API is a complete package which is easier to use and read, plus provides helpful operators, debugging and monitoring tools, and deployment features. This has led to growing use of TensorFlow library as a complete package within the ecosystem by the emerging body of students, researchers, developers, production engineers from various fields who are gravitating towards artificial intelligence. There is a TensorFlow for web, mobile, edge, embedded and more TensorFlow provides a range of services and modules within their existing ecosystem making them as one of the ground-breaking end-to-end tools to provide state-of-the-art deep learning. TensorFlow.js for machine learning on the web JavaScript library for training and deploying machine learning models in the browser. This library provides flexible and intuitive APIs to build and train new and pre-existing models from scratch right in the browser or under Node.js. TensorFlow Lite for mobile and embedded ML It is a TensorFlow lightweight solution used for mobile and embedded devices. It is fast since it enables on-device machine learning inference with low latency. It supports hardware acceleration with the Android Neural Networks API. The future releases of TensorFlow Lite will bring more built-in operators, performance improvements, and will support more models to simplify the developer’s experience of bringing machine learning services within mobile devices. TensorFlow Hub for reusable machine learning A library which is used extensively to reuse machine learning models. Thus you can transfer learning by reusing parts of machine learning models. TensorBoard for visual debugging While training a complex neural network model, the computations you use in TensorFlow can be very confusing. TensorBoard makes it very easy to understand and debug your TensorFlow programs in the form of visualizations. It allows you to easily inspect and understand your TensorFlow runs and graphs. Sonnet Sonnet is a DeepMind library which is built on top of TensorFlow extensively used to build complex neural network models. All of this factors have made the TensorFlow library immensely appealing for building a wide spectrum of machine learning and deep learning projects. This tool has become a preferred choice for everyone from space research giant NASA and other confidential government agencies, to an impressive roster of private sector giants. Road Ahead for TensorFlow TensorFlow no doubt is better marketed compared to the other deep learning frameworks. The community appears to be moving very fast. In any given hour, there are approximately 10 people around the world contributing or improving the TensorFlow project on GitHub. TensorFlow dominates the field with the largest active community. It will be interesting to see what new advances TensorFlow and other utilities make possible for the future of our digital world. Continuing the recent trend of rapid updates, the TensorFlow team is making sure they address all the current and active challenges faced by the contributors and the developers while building machine learning and deep learning models. TensorFlow 2.0 will be a major update, we can expect the release candidate by next year early March. The preview version of this major milestone is expected to hit later this year. The major focus will be on ease of use, additional support for more platforms and languages, and eager execution will be the central feature of TensorFlow 2.0. This breakthrough version will add more functionalities and operations to handle current research areas such as reinforcement learning, GANs, building advanced neural network models more efficiently. Google will continue to invest and upgrade their existing TensorFlow ecosystem. According to Google’s CEO, Sundar Pichai “artificial intelligence is more important than electricity or fire.” TensorFlow is the solution they have come up with to bring artificial intelligence into reality and provide a stepping stone to revolutionize humankind. Read more The 5 biggest announcements from TensorFlow Developer Summit 2018 The Deep Learning Framework Showdown: TensorFlow vs CNTK Tensor Processing Unit (TPU) 3.0: Google’s answer to cloud-ready Artificial Intelligence

Transfer learning is a method of reusing a model or knowledge for another related task. Transfer learning is sometimes also considered as an extension of existing ML algorithms. Extensive research and work is being done in the context of transfer learning and on understanding how knowledge can be transferred among tasks. However, the Neural Information Processing Systems (NIPS) 1995 workshop Learning to Learn: Knowledge Consolidation and Transfer in Inductive Systems is believed to have provided the initial motivations for research in this field. The literature on transfer learning has gone through a lot of iterations, and the terms associated with it have been used loosely and often interchangeably. Hence, it is sometimes confusing to differentiate between transfer learning, domain adaptation, and multitask learning. Rest assured, these are all related and try to solve similar problems. In this article, we will look into the five types of deep transfer learning to get more clarity on how these differ from each other. [box type="shadow" align="" class="" width=""]This article is an excerpt from a book written by Dipanjan Sarkar, Raghav Bali, and Tamoghna Ghosh titled Hands-On Transfer Learning with Python. This book covers deep learning and transfer learning in detail. It also focuses on real-world examples and research problems using TensorFlow, Keras, and the Python ecosystem with hands-on examples.[/box] #1 Domain adaptation Domain adaptation is usually referred to in scenarios where the marginal probabilities between the source and target domains are different, such as P (Xs) ≠ P (Xt). There is an inherent shift or drift in the data distribution of the source and target domains that requires tweaks to transfer the learning. For instance, a corpus of movie reviews labeled as positive or negative would be different from a corpus of product-review sentiments. A classifier trained on movie-review sentiment would see a different distribution if utilized to classify product reviews. Thus, domain adaptation techniques are utilized in transfer learning in these scenarios. #2 Domain confusion Different layers in a deep learning network capture different sets of features. We can utilize this fact to learn domain-invariant features and improve their transferability across domains. Instead of allowing the model to learn any representation, we nudge the representations of both domains to be as similar as possible. This can be achieved by applying certain preprocessing steps directly to the representations themselves. Some of these have been discussed by Baochen Sun, Jiashi Feng, and Kate Saenko in their paper Return of Frustratingly Easy Domain Adaptation. This nudge toward the similarity of representation has also been presented by Ganin et. al. in their paper, Domain-Adversarial Training of Neural Networks. The basic idea behind this technique is to add another objective to the source model to encourage similarity by confusing the domain itself, hence domain confusion. #3 Multitask learning Multitask learning is a slightly different flavor of the transfer learning world. In the case of multitask learning, several tasks are learned simultaneously without distinction between the source and targets. In this case, the learner receives information about multiple tasks at once, as compared to transfer learning, where the learner initially has no idea about the target task. This is depicted in the following diagram: Multitask learning: Learner receives information from all tasks simultaneously #4 One-shot learning Deep learning systems are data hungry by nature, such that they need many training examples to learn the weights. This is one of the limiting aspects of deep neural networks, though such is not the case with human learning. For instance, once a child is shown what an apple looks like, they can easily identify a different variety of apple (with one or a few training examples); this is not the case with ML and deep learning algorithms. One-shot learning is a variant of transfer learning where we try to infer the required output based on just one or a few training examples. This is essentially helpful in real-world scenarios where it is not possible to have labeled data for every possible class (if it is a classification task) and in scenarios where new classes can be added often. The landmark paper by Fei-Fei and their co-authors, One Shot Learning of Object Categories, is supposedly what coined the term one-shot learning and the research in this subfield. This paper presented a variation on a Bayesian framework for representation learning for object categorization. This approach has since been improved upon, and applied using deep learning systems. #5 Zero-shot learning Zero-shot learning is another extreme variant of transfer learning, which relies on no labeled examples to learn a task. This might sound unbelievable, especially when learning using examples is what most supervised learning algorithms are about. Zero-data learning, or zero-short learning, methods make clever adjustments during the training stage itself to exploit additional information to understand unseen data. In their book on Deep Learning, Goodfellow and their co-authors present zero-shot learning as a scenario where three variables are learned, such as the traditional input variable, x, the traditional output variable, y, and the additional random variable that describes the task, T. The model is thus trained to learn the conditional probability distribution of P(y | x, T). Zero-shot learning comes in handy in scenarios such as machine translation, where we may not even have labels in the target language. In this article we learned about the five types of deep transfer learning types: Domain adaptation, domain confusion, multitask learning, one-shot learning, and zero-shot learning. If you found this post useful, do check out the book, Hands-On Transfer Learning with Python, which covers deep learning and transfer learning in detail. It also focuses on real-world examples and research problems using TensorFlow, Keras, and the Python ecosystem with hands-on examples. CMU students propose a competitive reinforcement learning approach based on A3C using visual transfer between Atari games What is Meta Learning? Is the machine learning process similar to how humans learn?

Okay, you’re probably here because you’ve got just a few months to graduate and the projects section of your resume is blank. Or you’re just an inquisitive little nerd scraping the WWW for ways to crack that dream job. Either way, you’re not alone and there are ten thousand others trying to build a great Data Science portfolio to land them a good job. Look no further, we’ll try our best to help you on how to make a portfolio that catches the recruiter’s eye! David “Trent” Salazar‘s portfolio is a great example of a wholesome one and Sajal Sharma’s, is a good example of how one can display their Data Science Portfolios on a platform like Github. Companies are on the lookout for employees who can add value to the business. To showcase this on your resume effectively, the first step is to understand the different ways in which you can add value. 4 things you need to show in a data science portfolio Data science can be broken down into 4 broad areas: Obtaining insights from data and presenting them to the business leaders Designing an application that directly benefits the customer Designing an application or system that directly benefits other teams in the organisation Sharing expertise on data science with other teams You’ll need to ensure that your portfolio portrays all or at least most of the above, in order to easily make it through a job selection. So let’s see what we can do to make a great portfolio. Demonstrate that you know what you're doing So the idea is to show the recruiter that you’re capable of performing the critical aspects of Data Science, i.e. import a data set, clean the data, extract useful information from the data using various techniques, and finally visualise the findings and communicate them. Apart from the technical skills, there are a few soft skills that are expected as well. For instance, the ability to communicate and collaborate with others, the ability to reason and take the initiative when required. If your project is actually able to communicate these things, you’re in! Stay focused and be specific You might know a lot, but rather than throwing all your skills, projects and knowledge in the employer’s face, it’s always better to be focused on doing something and doing it right. Just as you’d do in your resume, keeping things short and sweet, you can implement this while building your portfolio too. Always remember, the interviewer is looking for specific skills. Research the data science job market Find 5-6 jobs, probably from Linkedin or Indeed, that interest you and go through their descriptions thoroughly. Understand what kind of skills the employer is looking for. For example, it could be classification, machine learning, statistical modeling or regression. Pick up the tools that are required for the job - for example, Python, R, TensorFlow, Hadoop, or whatever might get the job done. If you don’t know how to use that tool, you’ll want to skill-up as you work your way through the projects. Also, identify the kind of data that they would like you to be working on, like text or numerical, etc. Now, once you have this information at hand, start building your project around these skills and tools. Be a problem solver Working on projects that are not actual ‘problems’ that you’re solving, won’t stand out in your portfolio. The closer your projects are to the real-world, the easier it will be for the recruiter to make their decision to choose you. This will also showcase your analytical skills and how you’ve applied data science to solve a prevailing problem. Put at least 3 diverse projects in your data science portfolio A nice way to create a portfolio is to list 3 good projects that are diverse in nature. Here are some interesting projects to get you started on your portfolio: Data Cleaning and wrangling Data Cleaning is one of the most critical tasks that a data scientist performs. By taking a group of diverse data sets, consolidating and making sense of them, you’re giving the recruiter confidence that you know how to prep them for analysis. For example, you can take Twitter or Whatsapp data and clean it for analysis. The process is pretty simple; you first find a “dirty” data set, then spot an interesting angle to approach the data from, clean it up and perform analysis on it, and finally present your findings. Data storytelling Storytelling showcases not only your ability to draw insight from raw data, but it also reveals how well you’re able to convey the insights to others and persuade them. For example, you can use data from the bus system in your country and gather insights to identify which stops incur the most delays. This could be fixed by changing their route. Make sure your analysis is descriptive and your code and logic can be followed. Here’s what you do; first you find a good dataset, then you explore the data and spot correlations in the data. Then you visualize it before you start writing up your narrative. Tackle the data from various angles and pick up the most interesting one. If it’s interesting to you, it will most probably be interesting to anyone else who’s reviewing it. Break down and explain each step in detail, each code snippet, as if you were describing it to a friend. The idea is to teach the reviewer something new as you run through the analysis. End to end data science If you’re more into Machine Learning, or algorithm writing, you should do an end-to-end data science project. The project should be capable of taking in data, processing it and finally learning from it, every step of the way. For example, you can pick up fuel pricing data for your city or maybe stock market data. The data needs to be dynamic and updated regularly. The trick for this one is to keep the code simple so that it’s easy to set up and run. You first need to identify a good topic. Understand here that we will not be working with a single dataset, rather you will need to import and parse all the data and bring it under a single dataset yourself. Next, get the training and test data ready to make predictions. Document your code and other findings and you’re good to go. Prove you have the data science skill set If you want to get that job, you’ve got to have the appropriate tools to get the job done. Here’s a list of some of the most popular tools with a link to the right material for you to skill-up: Data science languages There's a number of key languages in data science that are essential. It might seem obvious, but making sure they're on your resume and demonstrated in your portfolio is incredibly important. Include things like: Python R Java Scala SQL Big Data tools If you're applying for big data roles, demonstrating your experience with the key technologies is a must. It not only proves you have the skills, but also shows that you have an awareness of what tools can be used to build a big data solution or project. You'll need: Hadoop, Spark Hive Machine learning frameworks With machine learning so in demand, if you can prove you've used a number of machine learning frameworks, you've already done a lot to impress. Remember, many organizations won't actually know as much about machine learning as you think. In fact, they might even be hiring you with a view to building out this capability. Remember to include: TensorFlow Caffe2 Keras PyTorch Data visualisation tools Data visualization is a crucial component of any data science project. If you can visualize and communicate data effectively, you're immediately demonstrating you're able to collaborate with others and make your insights accessible and useful to the wider business. Include tools like these in your resume and portfolio:  D3.js Excel chart  Tableau  ggplot2 So there you have it. You know what to do to build a decent data science portfolio. It’s really worth attending competitions and challenges. It will not only help you keep up to data and well oiled with your skills, but also give you a broader picture of what people are actually working on and with what tools they’re able to solve problems.

Artificial Intelligence is one of the hottest technologies currently. From work colleagues to your boss, chances are that most (yourself included) wish to create the next big AI project. Artificial Intelligence is a vast field and with thousands of languages to choose from, it can get a bit difficult to pick the language that will bring the most value to your project. For anyone wanting to dive in the AI space, the initial stage of choosing the right language can really decelerate the development process. Moreover, making a right choice about the language for the Artificial Intelligence development depends on your skills and needs. Following are the top 5 programming languages for Artificial Intelligence development: 1.Python Python, hands down, is the number one programming language when it comes to Artificial Intelligence development. Not only is it one of the most popular languages in the field of data science, machine learning, and Artificial Intelligence in general, it is also popular among game developers, web developers, cybersecurity professionals and others. It offers a ton of libraries and frameworks in Machine Learning and Deep Learning that are extremely powerful and essential for AI development such as TensorFlow, Theano, Keras, Scikit Learn, etc. Python is the go-to language for AI development for most people, novices and experts alike. Pros It’s quite easy to learn due to its simple syntax. This helps in implementing the AI algorithms in a quick and easy manner. Development is faster in Python as compared to Java, C++ or Ruby. It is a multi-paradigm programming language and supports object-oriented, functional and procedure-oriented programming languages. Python has a ton of libraries and tools to offer. Python libraries such as Scikit-learn, Numpy, CNTK, etc are quite trending. It is a portable language and can be used on multiple operating systems namely Windows, Mac OS, Linux, and Unix. Cons Integration of the AI systems with non-Python infrastructure. For e.g. for an infrastructure built around Java, it would be advisable to build deep learning models using Java rather than Python. If you are a data scientist, a machine learning developer or just a domain expert like a bioinformatician who hasn’t yet learned a programming language, Python is your best bet. It is easy to learn, translate equations and logic well in few lines of code and has a rich development ecosystem. 2.  C++ C++  comes second on the list when it comes to top 5 programming languages for Artificial Intelligence development. There are cases where C++ supersedes Python even though it is not the most common language when talking about AI development. For instance, when working with an embedded environment where you don’t want a lot of overhead due to Java Virtual Machine or Python Interpreter; C++ is a perfect choice. C++ also consists of some popular libraries and frameworks in AI, machine learning and deep learning namely, Mlpack, shark, OpenNN, Caffe, Dlib, etc. Pros Execution in C++ is very fast which is why it can be the go-to language when it comes to AI projects that are time-sensitive. It offers substantial use of algorithms. It uses statistical AI techniques quite effectively. Data hiding and inheritance make it possible to reuse the existing code during the development process. It is also suitable for machine learning and Neural Networks. Cons It follows a bottom-up approach and this makes it very complex for large-scale projects. If you are a game developer, you’ve already dabbled with C++ in some form or the other. Given the popularity of C++ among developers, it goes without saying, that if you choose C++, it can definitely kickstart your AI development process to build smarter, more interactive games. 3. Java Java is a close contender to C++. From Machine Learning to Natural language processing, Java comes with a plethora of libraries for all aspects of Artificial Intelligence development. Java has all the infrastructure that you need to create your next big AI project. Some popular Java libraries and frameworks are Deeplearning4j, Weka, Java-ML, etc. Pros Java follows the once Written Read/Run Anywhere (WORA) principle. It is a time-efficient language as it can be run on any platform without the need for re-compilation every time because of Virtual Machine Technology. Java works well for search algorithms, neural networks, and NLP. It is a multi-paradigm language i.e. it supports object-oriented, procedure-oriented and functional programming languages. It is easy to debug. Cons As mentioned, Java has a complex and verbose code structure which can be a bit time-consuming as it increases the development time. If you are into development of software, web, mobile or anywhere in between, you’ve worked with Java at some point, probably you still are. Most commercial apps have Java baked in them. The familiarity and robustness that Java has to offer is a good reason to pick Java when working with AI development. This is especially relevant if you want to enter well-established domains like banking that are historically built on top of Java-based systems. 4. Scala Just like Java, Scala belongs to the JVM family. Scala is a fairly new language in the AI space but it’s finding quite a bit of recognition recently in many corporations and startups. It has a lot to offer in terms of convenience which is why developers enjoy working with it. Also, ScalaNLP, DeepLearning4j, etc are all tools and libraries that make the AI development process a bit easier with Scala. Let’s have a look at the features that make it a good choice for AI development. Pros It’s good for projects that need scalability. It combines the strengths of Functional and Imperative programming models to act as a powerful tool which helps build highly concurrent applications while reaping the benefits of an OO approach at the same time. It provides good concurrency support which helps with projects involving real-time parallelized analytics. Scala has a good open source community when it comes to statistical learning, information theory and Artificial Intelligence in general. Cons Scala falls short when it comes to machine learning libraries. Scala consists of concepts such as implicits as well as type classes. These might not be familiar to programmers coming from the object-oriented world. The learning curve in Scala is steep. Even though Scala lacks in machine learning libraries, its scalability, and concurrency support makes it a good option for AI development. With more companies such as IBM and lightbend collaborating together to use Scala for building more AI applications, it’s no secret that Scala’s use for AI development is on constant demand in the present as well as for the future. 5. R R is a language that’s catching up in the race recently for AI development. Primarily used for academic research, R is written by statisticians and it provides basic data management which makes tasks really easy. It’s not as pricey as statistical software namely Matlab or SAS, which makes it a great substitute for this software and a golden child of data science. Pros R comes with plenty packages that help boost its performance. There are packages available for pre-modeling, modeling and post modeling stages in data analysis. R is very efficient in tasks such as continuous regression, model validation, and data visualization. R being a statistical language offers very robust statistical model packages for data analysis such as caret, ggplot, dplyr, lattice, etc which can help boost the AI development process. Major tasks can be done with little code developed in an interactive environment which makes it easy for the developers to try out new ideas and verify them with varied graphics functions that come with R. Cons R’s major drawback is its inconsistency due to third-party algorithms. Development speed is quite slow when it comes to R as you have to learn new ways for data modeling. You also have to make predictions every time when using a new algorithm. R is one of those skills that’s mainly demanded by recruiters in data science and machine learning. Overall, R is a very clever language. It is freely available, runs on server as well as common hardware. R can help amp up your AI development process to a great extent. Other languages worth mentioning There are three other languages that deserve a mention in this article: Go, Lisp and Prolog. Let’s have a look at what makes these a good choice for AI development. Go Go has been receiving a lot of attention recently. There might not be as many projects available in AI development using Go as for now but the language is on its path to continuous growth these days. For instance, AlphaGo, is a first computer program in Go that was able to defeat the world champion human Go player, proves how powerful the language is in terms of features that it can offer. Pros You don’t have to call out to libraries, you can make use of Go’s existing machine learning libraries. It doesn’t consist of classes. It only consists of packages which make the code cleaner and clear. It doesn’t support inheritance which makes it easy to modify the code in Go. Cons There aren’t many solid libraries for core AI development tasks. With Go, it is possible to pull off core ML and some reinforcement learning tasks as well, despite the lack of libraries. But given other versatile features of Go, the future looks bright for this language with it finding more applications in AI development. Lisp Lisp is one of the oldest languages for AI development and as such gets an honorary mention. It is a very popular language in AI academic research and is equally effective in the AI development process as well. However, it is not such a usual choice among the developers of recent times. Also, most modern libraries in machine learning, deep learning, and AI are written in popular languages such as C++, Python, etc. But I wouldn’t write off Lisp yet. It still has an immense capacity to build some really innovative AI projects, if take the time to learn it. Pros Its flexible and extendable nature enables fast prototyping, thereby, providing developers with the needed freedom to quickly test out ideas and theories. Since it was custom built for AI, its symbolic information processing capability is above par. It is suitable for machine learning and inductive learning based projects. Recompilation of functions alongside the running program is possible which saves time. Cons Since it is an old language, not a lot of developers are well-versed with it. Also, new software and hardware have to be configured to be able to accommodate using Lisp. Given the vintage nature of Lisp for the AI world, it is quite interesting to see how things work in Lisp for AI development.  The most famous example of a lisp-based AI project is DART (Dynamic Analysis and Replanning Tool), used by the U.S. military. Prolog Finally, we have Prolog, which is another old language primarily associated with AI development and symbolic computation. Pros It is a declarative language where everything is dictated by rules and facts. It supports mechanisms such as tree-based data structuring, automatic backtracking, nondeterminism and pattern matching which is helpful for AI development. This makes it quite a powerful language for AI development. Its varied features are quite helpful in creating AI projects for different fields such as medical, voice control, networking and other such Artificial development projects. It is flexible in nature and is used extensively for theorem proving, natural language processing, non-numerical programming, and AI in general. Cons High level of difficulty when it comes to learning Prolog as compared to other languages. Apart from the above-mentioned features, implementation of symbolic computation in other languages can take up to tens of pages of indigestible code. But the same algorithms implemented in Prolog results in a clear and concise program that easily fits on one page. So those are the top programming languages for Artificial Intelligence development. Choosing the right language eventually depends on the nature of your project. If you want to pick an easy to learn language go for Python but if you are working on a project where speed and performance are most critical then pick C++. If you are a creature of habit, Java is a good choice. If you are a thrill seeker who wants to learn a new and different language, choose Scala, R or Go, and if you are feeling particularly adventurous, explore the quaint old worlds of Lisp or Prolog. Why is Python so good for AI and Machine Learning? 5 Python Experts Explain Top 6 Java Machine Learning/Deep Learning frameworks you can’t miss 15 Useful Python Libraries to make your Data Science tasks Easier

Wouldn’t it be magical if we could watch old black and white movie footages and images in color? Deep learning, more precisely, GANs can help here. A recent approach by a software researcher Jason Antic tagged as ‘DeOldify’ is a deep learning based project for colorizing and restoring old images and film footages. https://twitter.com/johnbreslin/status/1127690102560448513 https://twitter.com/johnbreslin/status/1129360541955366913 In one of the sessions at the recent Facebook Developer Conference held from April 30 - May 1, 2019, Antic, along with Jeremy Howard, and Uri Manor talked about how by using GANs one can reconstruct images and videos, such as increasing their resolution or adding color to a black and white film. However, they also pointed out that GANs can be slow, and difficult and expensive to train. They demonstrated how to colorize old black & white movies and drastically increase the resolution of microscopy images using new PyTorch-based tools from fast.ai, the Salk Institute, and DeOldify that can be trained in just a few hours on a single GPU. https://twitter.com/citnaj/status/1123748626965114880 DeOldify makes use of a NoGAN training, which combines the benefits of GAN training (wonderful colorization) while eliminating the nasty side effects (like flickering objects in the video). NoGAN training is crucial while getting some images or videos stable and colorful. An example of DeOldify trying to achieve a stable video is as follows: Source: GitHub Antic said, “the video is rendered using isolated image generation without any sort of temporal modeling tacked on. The process performs 30-60 minutes of the GAN portion of "NoGAN" training, using 1% to 3% of Imagenet data once. Then, as with still image colorization, we "DeOldify" individual frames before rebuilding the video.” The three models in DeOldify DeOldify includes three models including video, stable and artistic. Each of the models has its strengths and weaknesses, and their own use cases. The Video model is for video and the other two are for images. Stable https://twitter.com/johnbreslin/status/1126733668347564034 This model achieves the best results with landscapes and portraits and produces fewer zombies (where faces or limbs stay gray rather than being colored in properly). It generally has less unusual miscolorations than artistic, but it's also less colorful in general. This model uses a resnet101 backbone on a UNet with an emphasis on width of layers on the decoder side. This model was trained with 3 critic pretrain/GAN cycle repeats via NoGAN, in addition to the initial generator/critic pretrain/GAN NoGAN training, at 192px. This adds up to a total of 7% of Imagenet data trained once (3 hours of direct GAN training). Artistic https://twitter.com/johnbreslin/status/1129364635730272256 This model achieves the highest quality results in image coloration, with respect to interesting details and vibrance. However, in order to achieve this, one has to adjust the rendering resolution or render_factor. Additionally, the model does not do as well as ‘stable’ in a few key common scenarios- nature scenes and portraits. Artistic model uses a resnet34 backbone on a UNet with an emphasis on depth of layers on the decoder side. This model was trained with 5 critic pretrain/GAN cycle repeats via NoGAN, in addition to the initial generator/critic pretrain/GAN NoGAN training, at 192px. This adds up to a total of 32% of Imagenet data trained once (12.5 hours of direct GAN training). Video https://twitter.com/citnaj/status/1124719757997907968 The Video model is optimized for smooth, consistent and flicker-free video. This would definitely be the least colorful of the three models; while being almost close to the ‘stable’ model. In terms of architecture, this model is the same as "stable"; however, differs in training. It's trained for a mere 2.2% of Imagenet data once at 192px, using only the initial generator/critic pretrain/GAN NoGAN training (1 hour of direct GAN training). DeOldify was achieved by combining certain approaches including: Self-Attention Generative Adversarial Network: Here, Antic has modified the generator, a pre-trained U-Net, to have the spectral normalization and self-attention. Two Time-Scale Update Rule: It’s just one to one generator/critic iterations and higher critic learning rate. This is modified to incorporate a "threshold" critic loss that makes sure that the critic is "caught up" before moving on to generator training. This is particularly useful for the "NoGAN" method. NoGAN doesn’t have a separate research paper. This, in fact, is a new type of GAN training developed to solve some key problems in the previous DeOldify model. NoGAN includes the benefits of GAN training while spending minimal time doing direct GAN training. Antic says, “I'm looking to make old photos and film look reeeeaaally good with GANs, and more importantly, make the project useful.” “I'll be actively updating and improving the code over the foreseeable future. I'll try to make this as user-friendly as possible, but I'm sure there's going to be hiccups along the way”, he further added. To further know about the hardware components and other details head over to Jason Antic’s GitHub page. Training Deep Convolutional GANs to generate Anime Characters [Tutorial] Sherin Thomas explains how to build a pipeline in PyTorch for deep learning workflows Using deep learning methods to detect malware in Android Applications

A formal definition of machine learning proposed by computer scientist Tom M. Mitchell states that a machine learns whenever it is able to utilize an experience such that its performance improves on similar experiences in the future. Although this definition is intuitive, it completely ignores the process of exactly how experience can be translated into future action—and of course, learning is always easier said than done! While human brains are naturally capable of learning from birth, the conditions necessary for computers to learn must be made explicit. For this reason, although it is not strictly necessary to understand the theoretical basis of learning, this foundation helps to understand, distinguish, and implement machine learning algorithms. This article is taken from the book Machine learning with R - Second Edition, written by Brett Lantz. Regardless of whether the learner is a human or machine, the basic learning process is similar. It can be divided into four interrelated components: Data storage utilizes observation, memory, and recall to provide a factual basis for further reasoning. Abstraction involves the translation of stored data into broader representations and concepts. Generalization uses abstracted data to create knowledge and inferences that drive action in new contexts. Evaluation provides a feedback mechanism to measure the utility of learned knowledge and inform potential improvements. The following figure illustrates the steps in the learning process: Keep in mind that although the learning process has been conceptualized as four distinct components, they are merely organized this way for illustrative purposes. In reality, the entire learning process is inextricably linked. In human beings, the process occurs subconsciously. We recollect, deduce, induct, and intuit with the confines of our mind's eye, and because this process is hidden, any differences from person to person are attributed to a vague notion of subjectivity. In contrast, with computers these processes are explicit, and because the entire process is transparent, the learned knowledge can be examined, transferred, and utilized for future action. Data storage for advanced reasoning All learning must begin with data. Humans and computers alike utilize data storage as a foundation for more advanced reasoning. In a human being, this consists of a brain that uses electrochemical signals in a network of biological cells to store and process observations for short- and long-term future recall. Computers have similar capabilities of short- and long-term recall using hard disk drives, flash memory, and random access memory (RAM) in combination with a central processing unit (CPU). It may seem obvious to say so, but the ability to store and retrieve data alone is not sufficient for learning. Without a higher level of understanding, knowledge is limited exclusively to recall, meaning exclusively what is seen before and nothing else. The data is merely ones and zeros on a disk. They are stored memories with no broader meaning. To better understand the nuances of this idea, it may help to think about the last time you studied for a difficult test, perhaps for a university final exam or a career certification. Did you wish for an eidetic (photographic) memory? If so, you may be disappointed to learn that perfect recall is unlikely to be of much assistance. Even if you could memorize material perfectly, your rote learning is of no use, unless you know in advance the exact questions and answers that will appear in the exam. Otherwise, you would be stuck in an attempt to memorize answers to every question that could conceivably be asked. Obviously, this is an unsustainable strategy. Instead, a better approach is to spend time selectively, memorizing a small set of representative ideas while developing strategies on how the ideas relate and how to use the stored information. In this way, large ideas can be understood without needing to memorize them by rote. Abstraction of stored data This work of assigning meaning to stored data occurs during the abstraction process, in which raw data comes to have a more abstract meaning. This type of connection, say between an object and its representation, is exemplified by the famous René Magritte painting The Treachery of Images: Source: http://collections.lacma.org/node/239578 The painting depicts a tobacco pipe with the caption Ceci n'est pas une pipe ("this is not a pipe"). The point Magritte was illustrating is that a representation of a pipe is not truly a pipe. Yet, in spite of the fact that the pipe is not real, anybody viewing the painting easily recognizes it as a pipe. This suggests that the observer's mind is able to connect the picture of a pipe to the idea of a pipe, to a memory of a physical pipe that could be held in the hand. Abstracted connections like these are the basis of knowledge representation, the formation of logical structures that assist in turning raw sensory information into a meaningful insight. During a machine's process of knowledge representation, the computer summarizes stored raw data using a model, an explicit description of the patterns within the data. Just like Magritte's pipe, the model representation takes on a life beyond the raw data. It represents an idea greater than the sum of its parts. There are many different types of models. You may be already familiar with some. Examples include: Mathematical equations Relational diagrams such as trees and graphs Logical if/else rules Groupings of data known as clusters The choice of model is typically not left up to the machine. Instead, the learning task and data on hand inform model selection. The process of fitting a model to a dataset is known as training. When the model has been trained, the data is transformed into an abstract form that summarizes the original information. It is important to note that a learned model does not itself provide new data, yet it does result in new knowledge. How can this be? The answer is that imposing an assumed structure on the underlying data gives insight into the unseen by supposing a concept about how data elements are related. Take for instance the discovery of gravity. By fitting equations to observational data, Sir Isaac Newton inferred the concept of gravity. But the force we now know as gravity was always present. It simply wasn't recognized until Newton recognized it as an abstract concept that relates some data to others—specifically, by becoming the g term in a model that explains observations of falling objects. Most models may not result in the development of theories that shake up scientific thought for centuries. Still, your model might result in the discovery of previously unseen relationships among data. A model trained on genomic data might find several genes that, when combined, are responsible for the onset of diabetes; banks might discover a seemingly innocuous type of transaction that systematically appears prior to fraudulent activity; and psychologists might identify a combination of personality characteristics indicating a new disorder. These underlying patterns were always present, but by simply presenting information in a different format, a new idea is conceptualized. Generalization for future action The learning process is not complete until the learner is able to use its abstracted knowledge for future action. However, among the countless underlying patterns that might be identified during the abstraction process and the myriad ways to model these patterns, some will be more useful than others. Unless the production of abstractions is limited, the learner will be unable to proceed. It would be stuck where it started—with a large pool of information, but no actionable insight. The term generalization describes the process of turning abstracted knowledge into a form that can be utilized for future action, on tasks that are similar, but not identical, to those it has seen before. Generalization is a somewhat vague process that is a bit difficult to describe. Traditionally, it has been imagined as a search through the entire set of models (that is, theories or inferences) that could be abstracted during training. In other words, if you can imagine a hypothetical set containing every possible theory that could be established from the data, generalization involves the reduction of this set into a manageable number of important findings. In generalization, the learner is tasked with limiting the patterns it discovers to only those that will be most relevant to its future tasks. Generally, it is not feasible to reduce the number of patterns by examining them one-by-one and ranking them by future utility. Instead, machine learning algorithms generally employ shortcuts that reduce the search space more quickly. Toward this end, the algorithm will employ heuristics, which are educated guesses about where to find the most useful inferences. Heuristics are routinely used by human beings to quickly generalize experience to new scenarios. If you have ever utilized your gut instinct to make a snap decision prior to fully evaluating your circumstances, you were intuitively using mental heuristics. The incredible human ability to make quick decisions often relies not on computer-like logic, but rather on heuristics guided by emotions. Sometimes, this can result in illogical conclusions. For example, more people express fear of airline travel versus automobile travel, despite automobiles being statistically more dangerous. This can be explained by the availability heuristic, which is the tendency of people to estimate the likelihood of an event by how easily its examples can be recalled. Accidents involving air travel are highly publicized. Being traumatic events, they are likely to be recalled very easily, whereas car accidents barely warrant a mention in the newspaper. The folly of misapplied heuristics is not limited to human beings. The heuristics employed by machine learning algorithms also sometimes result in erroneous conclusions. The algorithm is said to have a bias if the conclusions are systematically erroneous, or wrong in a predictable manner. For example, suppose that a machine learning algorithm learned to identify faces by finding two dark circles representing eyes, positioned above a straight line indicating a mouth. The algorithm might then have trouble with, or be biased against, faces that do not conform to its model. Faces with glasses, turned at an angle, looking sideways, or with various skin tones might not be detected by the algorithm. Similarly, it could be biased toward faces with certain skin tones, face shapes, or other characteristics that do not conform to its understanding of the world. In modern usage, the word bias has come to carry quite negative connotations. Various forms of media frequently claim to be free from bias, and claim to report the facts objectively, untainted by emotion. Still, consider for a moment the possibility that a little bias might be useful. Without a bit of arbitrariness, might it be a bit difficult to decide among several competing choices, each with distinct strengths and weaknesses? Indeed, some recent studies in the field of psychology have suggested that individuals born with damage to portions of the brain responsible for emotion are ineffectual in decision making, and might spend hours debating simple decisions such as what color shirt to wear or where to eat lunch. Paradoxically, bias is what blinds us from some information while also allowing us to utilize other information for action. It is how machine learning algorithms choose among the countless ways to understand a set of data. Evaluate the learner’s success Bias is a necessary evil associated with the abstraction and generalization processes inherent in any learning task. In order to drive action in the face of limitless possibility, each learner must be biased in a particular way. Consequently, each learner has its weaknesses and there is no single learning algorithm to rule them all. Therefore, the final step in the generalization process is to evaluate or measure the learner's success in spite of its biases and use this information to inform additional training if needed. Generally, evaluation occurs after a model has been trained on an initial training dataset. Then, the model is evaluated on a new test dataset in order to judge how well its characterization of the training data generalizes to new, unseen data. It's worth noting that it is exceedingly rare for a model to perfectly generalize to every unforeseen case. In parts, models fail to perfectly generalize due to the problem of noise, a term that describes unexplained or unexplainable variations in data. Noisy data is caused by seemingly random events, such as: Measurement error due to imprecise sensors that sometimes add or subtract a bit from the readings Issues with human subjects, such as survey respondents reporting random answers to survey questions, in order to finish more quickly Data quality problems, including missing, null, truncated, incorrectly coded, or corrupted values Phenomena that are so complex or so little understood that they impact the data in ways that appear to be unsystematic Trying to model noise is the basis of a problem called overfitting. Because most noisy data is unexplainable by definition, attempting to explain the noise will result in erroneous conclusions that do not generalize well to new cases. Efforts to explain the noise will also typically result in more complex models that will miss the true pattern that the learner tries to identify. A model that seems to perform well during training, but does poorly during evaluation, is said to be overfitted to the training dataset, as it does not generalize well to the test dataset. Solutions to the problem of overfitting are specific to particular machine learning approaches. For now, the important point is to be aware of the issue. How well the models are able to handle noisy data is an important source of distinction among them. We saw that machine learning process is similar to how humans learn in their daily lives.To To discover how to build machine learning algorithms, prepare data, and dig deep into data prediction techniques with R, check out this book Machine learning with R - Second edition. A Machine learning roadmap for Web Developers Why TensorFlow always tops machine learning and artificial intelligence tool surveys Intelligent Edge Analytics: 7 ways machine learning is driving edge computing adoption in 2018

Ah, Raspberry Pi, the little computer that could. On its initial release back in 2012, it quickly gained popularity among creators and hobbyists as a cheap and portable computer that could be the brain of their hardware projects. Fast forward to 2017, and Raspberry Pi is on its third generation and has been used in many more projects across various fields of study.  Tech giants are noticing this trend and have started to pay closer attention to the miniature computer. Microsoft, for example, released Windows 10 IoT Core, a variant of Windows 10 that could run on a Raspberry Pi. Recently, Google revealed that they have plans to bring artificial intelligence tools to the Pi. And not just Google's AI, more and more AI libraries and tools are being ported to the Raspberry Pi every day.  But what does it all mean? Does it have any impact on Raspberry Pi’s usage? Does it change anything in the world of Internet of Things? For starters, let's recap what the Raspberry Pi is and how it has been used so far. The Raspberry Pi, in short, is a super cheap computer (it only costs $35) and is the size of a credit card. However, despite its ability to be usedasa usual, general-purpose computer, most people useRaspberry Pias the base of their hardware projects.  These projects range from simple toy-like projects to complicated gadgets that actually do important work. They can be as simple as a media center for your TV or as complex as a house automation system. Do keep in mind that this kind of projectscan always be built using desktop computers, but it's not really practical to do so without the low price and the small size of the Raspberry Pi.  Before we go on talking about having artificial intelligence on the Raspberry Pi, we need to have the same understanding of AI. (from http://i2.cdn.turner.com/cnn/2010/TECH/innovation/07/09/face.recognition.facebook/t1larg.tech.face.recognition.courtesy.jpg)  Artificial Intelligence has a wide range of complexity. It can range from a complicated digital assistant like Siri, to a news-sorting program, to a simple face detection system that can be found in many cameras. The more complicated the AI system, the bigger the computing power required by the system. So, with the limited processing power we have on the Raspberry Pi, the types of AI that can run on that mini computer will be limited to the simple ones as well.  Also, there's another aspect of AI called machine learning. It's the kind of technology that enables an AI to play and win against humans in a match of Go. The core of machine learning is basically to make a computer improve its own algorithm by processing a large amount of data. For example, if we feed a computer thousands of cat pictures, it will be able to define a pattern for 'cat' and use that pattern to find cats in other pictures.  There are two parts in machine learning. The first one is the training part, where we let a computer find an algorithm that suits the problem. The second aspect is the application part, where we apply the new algorithm to solve the actual problem. While the application part can usually be run on a Raspberry Pi, the training part requires a much higher processing power. To make it work, the training part is done on a high-performance computer elsewhere, and the Raspberry Pi only executes the training result. So, now we know that the Raspberry Pi can run simple AI. But what's the impact of this?  Well, to put it simply, having AI will enable creators to build an entirely new class of gadgets on the Raspberry Pi. It will allow the makers to create an actually smart device based on the small computer. Without AI, the so-called smart device will only act following a limited set of rules that have been defined. For example, we can develop a device that automatically turns off lights at a specific time every day, but without AI we can't have the device detect if there's anyone in the room or not.  With artificial intelligence, our devices will be able to adapt to unscripted changes in our environment. Imagine connecting a toy car with a Raspberry Pi and a webcam and have the car be able to smartly map its path to the goal, or a device that automatically opens the garage door if it sees our car coming in. Having AI on the Raspberry Pi will enable the development of such smart devices.  There's another thing to consider. One of Raspberry Pi's strong points is its versatility. With its USB ports and GPIO pins, the computer is able to interface with various digital sensors. The addition of AI will enable the Raspberry Pi to process even more sensors like fingerprint readers or speech recognition with a microphone, further enhancing its flexibility.  All in all, artificial intelligence is a perfect addition to the Raspberry Pi. It enables the creation of even smarter devices based on the computer and unlocks the potential of the Internet of Things to every maker and tinkerer in the world. About the author RakaMahesa is a game developer at Chocoarts (http://chocoarts.com/),who is interested in digital technology in general. Outside of work hours, he likes to work on his own projects, with Corridoom VR being his latest released game. Raka also regularly tweets as @legacy99. 

Computer Vision is one of those technologies that has grown in leaps and bounds over the past few years. If you look back 10 years, it wasn’t the case, as CV was more a topic of academic interest. Now, however, computer vision is clearly a driver and benefactor of the renowned Artificial Intelligence. Through this article, we’ll understand the factors that have sparked the rise of Computer Vision. A billion $ market You heard it right! Computer Vision is a billion dollar market, thanks to the likes of Intel, Amazon, Netflix, etc investing heavily in the technology’s development. And from the way events are unfolding, the market is expected to hit a record $ 17 billion, by 2023. That’s at a cumulative growth rate of over 7% per year, from 2018 to 2023. Now this is a joint figure for both the hardware and software components related to Computer Vision. Under the spotlight Let’s talk a bit about a few companies that are already taking advantage of Computer Vision, and are benefiting from it. Intel There are several large organisations that are investing heavily in Computer Vision. Last year, we saw Intel invest $15 Billion towards acquiring Mobileye, an Israeli auto startup. Intel published its findings stating that the autonomous vehicle market itself would rise to $ 7 Trillion by 2050. The autonomous vehicle industry will be one of the largest implementers of computer vision technology. These vehicles will use Computer Vision to “see” their surroundings and communicate with other vehicles. Netflix Netflix on the other hand, is using Computer Vision for more creative purposes. With the rise of Netflix’s original content, the company is investing in Computer Vision to harvest static image frames directly from the source videos to provide a flexible source of raw artwork, which is used for digital merchandising. For example, within a single episode of Stranger Things, there are nearly 86k static video frames, that would had to have been analysed by human teams to identify the most appropriate stills to be featured. This meant first going through each of those 86k images, then understanding what worked for viewers of the previous episode and then applying the learning in the selection of future images. Need I estimate how long that would have taken to do? Now, Computer Vision performs this task seamlessly, with a much higher accuracy than that of humans. Pinterest Pinterest, the popular social networking application, sees millions of images, GIFs and other visuals shared every day. In 2017, they released an application feature callen Lens, that allows users to use their phone’s camera to search for similar looking decor, food and clothing, in the real world. Users can simply point their cameras at an image and Pinterest will show them similar styles and ideas. Recent reports reveal that Pinterest’s revenue has grown by a staggering 58%! National Surveillance in CCTV The world’s biggest AI startup, SenseTime, provides China with the world’s largest and most sophisticated CCTV network. With over 170 Mn CCTV cameras, the government authorities and police departments are able to seamlessly identify people. They perform this by wearing smart glasses, that have facial recognition capabilities. Bring this technology to Dubai and you’ve got a supercop in a supercar! The nation-wide surveillance project that’s named Skynet, began as early as 2005, although recent advances in AI have given it a boost. Reading through discussions like these is real fun. People used to quip that such “fancy” machines are only for the screen. If only they knew that such a machine would be a reality just a few years from then. Clearly, computer vision is one of the most highly valued commercial applications of machine learning and when integrated with AI, it’s an offer only a few can resist! Star Acquisitions that matter Several acquisitions have taken place in the field of Computer Vision in the past two years alone. The most notable of them being Intel’s acquisition of Movidius, to the tune of $400 Mn. Here are some of the others that have happened since 2016: Twitter acquires Magic Pony Technology for $150Mn Snap Inc acquires Obvious Engineering for $47 Mn Salesforce acquires Metamind for $32.8 Mn Google acquires Eyefluence for $21.6 Mn This shows the potential of the computer vision market and how big players are in the race to dive deep into the technology. Three little things driving computer vision I would say there are 3 clear growth factors that are contributing to the rise of Computer Vision: Deep Learning Advancements in Hardware Growth of the Datasets Deep Learning The advancements in the field of Deep Learning are bound to boost Computer Vision. Deep Learning algorithms are capable of processing tonnes of images, much more accurately than humans. Take Feature Extraction for example. The primary pain point with feature extraction is that you have to choose which features to look for in a given image. This becomes cumbersome and almost impossible when the number of classes you are trying to define, starts to grow. There are so many features, that you have to deal with a plethora of parameters, that have to be fine-tuned. Deep Learning simplifies this process for you. Advancements in Hardware With new hardware like GPUs capable of processing petabytes of data, algorithms are capable of running faster and more efficiently. This has led to the advancement in real-time processing and vision capabilities. Pioneering hardware manufacturers like NVIDIA and Intel are in a race to create more powerful and capable hardware to support deep learning capabilities for Computer Vision. Growth of the Datasets Training Deep Learning algorithms isn’t a daunting task anymore. There are plenty of open source data sets that you can choose from to train your algorithms. The more the data, the better is the training and accuracy. Here are some of the most notable data sets for computer vision. ImageNet with 15 million images, is a massive dataset Open Images has 9 million images Microsoft Common Objects in Context (COCO) has around 330K images CALTECH-101  has approximately 9,000 images Where tha money at? The job market for Computer Vision is on a rise too, with Computer Vision featuring at #3 on the list of top jobs in 2018, according to Indeed. Organisations are looking for Computer Vision Engineers who are well versed with writing efficient algorithms for handling large amounts of data. Source: Indeed.com So is it the right time to invest or perhaps learn Computer Vision? You bet it is! It’s clear that Computer Vision is a rapidly growing market and will have a sustained growth for the next few years. If you’re just planning to start out or even if you’re competent in using tools for Computer Vision, here are some resources to help you skill up with popular CV tools and techniques. Introducing Intel’s OpenVINO computer vision toolkit for edge computing Top 10 Tools for Computer Vision Computer Vision with Keras, Part 1

On November 9, 2015, a storm loomed over the SF Bay area creating major outages. At Mountain View, California, Google engineers were busy creating a storm of their own. That day, Sundar Pichai announced to the world that TensorFlow, their machine learning system, was going Open Source. He said: “...today we’re also open-sourcing TensorFlow. We hope this will let the machine learning community—everyone from academic researchers, to engineers, to hobbyists—exchange ideas much more quickly, through working code rather than just research papers.” That day the tech world may not have fully grasped the gravity of the announcement but those in the know knew it was a pivotal moment in Google’s transformational journey into an AI first world. How did TensorFlow begin? TensorFlow was part of a former Google product called DistBelief. DistBelief was responsible for a program called DeepDream. The program was built for scientists and engineers to visualise how deep neural networks process images. As fate would have it, the algorithm went viral and everyone started visualising abstract and psychedelic art in it. Although people were having fun playing with image forms, they were unaware of the technology that powered those images - neural networks and deep learning - the exact reason why TensorFlow was built for. TensorFlow is a machine learning platform that allows one to run a wide range of algorithms like the aforementioned neural networks and deep learning based projects. TensorFlow with its flexibility, high performance, portability, and production-readiness is changing the landscape of artificial intelligence and machine learning. Be it face recognition, music, and art creation or detecting clickbait headline for blogs, the use cases are immense. With Google open sourcing TensorFlow, the platform that powers Google search and other smart Google products is now accessible to everyone - researchers, scientists, machine learning experts, students, and others. So why did Google open source TensorFlow? Yes, Google made a world of difference to the machine learning community at large by open sourcing TensorFlow. But what was in it for Google? As it turns out, a whole lot. Let’s look at a few. Google is feeling the heat from rival deep learning frameworks Major deep learning frameworks like Theano, Keras, etc., were already open source. Keeping a framework proprietary was becoming a strategic disadvantage as most DL core users i.e. scientists, engineers, and academicians prefer using open source software for their work. “Pure” researchers and aspiring “Phds” are key groups that file major patents in the world of AI. By open sourcing TensorFlow, Google gave this community access to a platform it backs to power their research. This makes migrating the world’s algorithms from other deep learning tools onto TensorFlow theoretically possible. AI as a trend is clearly here to stay and Google wants a platform that leads this trend. An open source TensorFlow can better support the Google Brain project Behind all the PR, Google does not speak much about its pet project Google Brain. When Sundar Pichai talks of Google’s transformation from Search to AI, this project is doing all the work behind the scenes. Google Brain is headed by some of the best minds in the industry like Jeff Dean, Geoffery Hilton, Andrew NG, among many others. They developed TensorFlow and they might still have some state-of-the-art features up their sleeves privy only to them. After all, they have done a plethora of stunning research in areas like parallel computing, machine intelligence, natural language processing and many more. With TensorFlow now open sourced, this team can accelerate the development of the platform and also make significant inroads into areas they are currently researching on. This research can then potentially develop into future products for Google which will allow them to expand their AI and Cloud clout, especially in the enterprise market. Tapping into the collective wisdom of the academic intelligentsia Most innovations and breakthroughs come from universities before they go mainstream and become major products in enterprises. AI, still making this transition, will need a lot of investment in research. To work on difficult algorithms, researchers will need access to sophisticated ML frameworks. Selling TensorFlow to universities is an old school way to solve the problem - that’s why we no longer hear about products like LabView. Instead, by open-sourcing TensorFlow, the team at Google now has the world’s best minds working on difficult AI problems on their platform for free. As these researchers start writing papers on AI using TensorFlow, it will keep adding to the existing body of knowledge. They will have all the access to bleeding-edge algorithms that are not yet available in the market. Their engineers could simply pick and choose what they like and start developing commercially ready services. Google wants to develop TensorFlow as a platform-as-a-service for AI application development An advantage of open-sourcing a tool is that it accelerates time to build and test through collaborative app development. This means most of the basic infrastructure and modules to build a variety of TensorFlow based applications will already exist on the platform. TensorFlow developers can develop and ship interesting modular products by mixing and matching code and providing a further layer of customization or abstraction. What Amazon did for storage with AWS, Google can do for AI with TensorFlow. It won’t come as a surprise if Google came up with their own integrated AI ecosystem with TensorFlow on the Google Cloud promising you the AI resources your company would need. Suppose you want a voice based search function on your ecommerce mobile application. Instead, of completely reinventing the wheel, you could buy TensorFlow powered services provided by Google. With easy APIs, you can get voice based search and save substantial developer cost and time. Open sourcing TensorFlow will help Google to extend their talent pipeline in a competitive Silicon Valley jobs market Hiring for AI development is  competitive in the Silicon Valley as all major companies vie for attention from the same niche talent pool. With TensorFlow made freely available, Google’s HR team can quickly reach out to a talent pool specifically well versed with the technology and also save on training cost. Just look at the interest TensorFlow has generated on a forum like StackOverflow: This indicates that growing number of users are asking and inquiring about TensorFlow. Some of these users will migrate into power users who the Google HR team can tap into. A developer pool at this scale would never have been possible with a proprietary tool. Replicating the success and learning from Android Agreed, a direct comparison with Android is not possible. However, the size of the mobile market and Google’s strategic goal of mobile-first when they introduced Android bear striking similarity with the nascent AI ecosystem we have today and Google’s current AI-first rhetoric. In just a decade since its launch, Android now owns more than 85% of the smartphone mobile OS market. Piggybacking on Android’s success, Google now has control of mobile search (96.19%), services (Google Play), a strong connection with the mobile developer community and even a viable entry into the mobile hardware market. Open sourcing Android did not stop Google from making money. Google was able to monetize through other ways like mobile search, mobile advertisements, Google Play, devices like Nexus, mobile payments, etc. Google did not have all this infrastructure planned and ready before Android was open sourced - It innovated, improvised, and created along the way. In the future, we can expect Google to adopt key learnings from its Android growth story and apply to TensorFlow’s market expansion strategy. We can also see supporting infrastructures and models for commercialising TensorFlow emerge for enterprise developers. [dropcap]T[/dropcap]he road to AI world domination for Google is on the back of an open sourced TensorFlow platform. It appears not just exciting but also promises to be one full of exponential growth, crowdsourced innovation and learnings drawn from other highly successful Google products and services. The storm that started two years ago is surely morphing into a hurricane. As Professor Michael Guerzhoy of University of Toronto quotes in Business Insider “Ten years ago, it took me months to do something that for my students takes a few days with TensorFlow.”

‘Can Recurrent neural networks warp time?’ is authored by Corentin Tallec and Yann Ollivier to be presented at ICLR 2018. This paper explains that plain RNNs cannot account for warpings, leaky RNNs can account for uniform time scalings but not irregular warpings, and gated RNNs can adapt to irregular warpings. Gating mechanism of LSTMS (and GRUs) to time invariance / warping What problem is the paper trying to solve? In this paper, that authors prove that learnable gates in a recurrent model formally provide quasi-invariance to general time transformations in the input data. Further, the authors try to recover part of the LSTM architecture from a simple axiomatic approach. This leads to a new way of initializing gate biases in LSTMs and GRUs. Experimentally, this new chrono initialization is shown to greatly improve learning of long term dependencies, with minimal implementation effort. Paper summary The authors have derived the self loop feedback gating mechanism of recurrent networks from first principles via a postulate of invariance to time warpings. Gated connections appear to regulate the local time constants in recurrent models. With this in mind, the chrono initialization, a principled way of initializing gate biases in LSTMs, has been introduced. Experimentally, chrono initialization is shown to bring notable benefits when facing long term dependencies. Key takeaways In this paper, the authors show that postulating invariance to time transformations in the data (taking invariance to time warping as an axiom) necessarily leads to a gate-like mechanism in recurrent models. The paper provides precise prescriptions on how to initialize gate biases depending on the range of time dependencies to be captured. The empirical benefits of the new initialization on both synthetic and real world data have been tested. The authors also observed a substantial improvement with long-term dependencies, and slight gains or no change when short-term dependencies dominate. Reviewer comments summary Overall Score: 25/30 Average Score: 8 According to a reviewer, the core insight of the paper is the link between recurrent network design and its effect on how the network reacts to time transformations. This insight is simple, elegant and valuable, as per the reviewer. A minor complaint highlighted is that there are an unnecessarily large number of paragraph breaks, which make reading slightly jarring. Recurrent neural networks and the LSTM architecture Build a generative chatbot using recurrent neural networks (LSTM RNNs) How to recognize Patterns with Neural Networks in Java  

Last year, at Google I/O, Sundar Pichai, the CEO of Google, said: “We are moving from a mobile-first world to an AI-first world” Is it only applicable to Google? Not really. In the recent past, we have seen several advancements in Artificial Intelligence and in parallel a plethora of intelligent apps coming into the market. These advancements are enabling developers to take their apps to the next level by integrating recommendation service, image recognition, speech recognition, voice translation, and many more cool capabilities. Artificial Intelligence is becoming a potent tool for mobile developers to experiment and innovate. The Artificial Intelligence components that are integral to mobile experiences, such as voice-based assistants and location-based services, increasingly require mobile developers to have a basic understanding of Artificial Intelligence to be effective. Of course, you don’t have to be Artificial Intelligence experts to include intelligent components in your app. But, you should definitely understand something about what you’re building into your app and why. After all AI in mobile is not just limited to calling an API, isn't it? There’s more to it and in this article we will explore how Artificial Intelligence will shape the mobile developer role in the immediate future. Read also: AI on mobile: How AI is taking over the mobile devices marketspace What is changing in the mobile developer role? Focus shifting to data With Artificial Intelligence becoming more and more accessible, intelligent apps are becoming the new norm for businesses. Artificial Intelligence strengthens the relationship between brands and customers, inspiring developers to build smart apps that increase user retention. This also means that developers have to direct their focus to data. They have to understand things like how the data will be collected? How will the data be fed to machines and how often will data input be needed? When nearly 1 in 4 people abandon an app after its first use, as a mobile app developer, you need to rethink how you drive in-app personalization and engagement. Explore “humanized” way of user-app interaction With so many chatbots such as Siri and Google Assistant coming into the market, we can see that “humanizing” the interaction between the user and the app is becoming mainstream. “Humanizing” is the process where the app becomes relatable to the user, and the more effective it is conducted, the more the end user will interact with the app. Users now want easy navigation and searching system and Artificial Intelligence fits perfectly in the scenario. The advances in technologies like text-to-speech, speech-to-text, Natural Language Processing, and cloud services, in general, have contributed to the mass adoption of these types of interfaces. Companies are increasingly expecting mobile developers to be comfortable working with AI functionalities Artificial Intelligence is the future. Companies are now expecting their mobile developers to know how to handle the huge amount of data generated every day and how to use it. Here's is an example of what Google wants their engineers to do: “We're looking for engineers who bring fresh ideas from all areas, including information retrieval, distributed computing, large-scale system design, networking and data storage, security, artificial intelligence, natural language processing, UI design and mobile; the list goes on and is growing every day.” This open-ended requirement list shows that it is the right time to learn and embrace Artificial Intelligence as soon as possible. What skills do you need to build intelligent apps? Ideally, data scientists are the ones who conceptualize mathematical models and machine learning engineers are the ones who translate it into the code and train the model. But, when you are working in a resource-tight environment, for example in a start-up, you will be responsible for doing the end-to-end job. It is not as scary as it sounds, because you have several resources to get started with! Taking your first steps with machine learning as a service Learning anything starts with motivating yourself. Directly diving into the maths and coding part of machine learning might exhaust and bore you. That's why it's a good idea to know what the end goal of your entire learning process is going to be and what types of solutions are possible using machine learning. There are many products available that you can try to quickly get started such as Google Cloud AutoML (Beta), Firebase MLKit (Beta), and Fritz Mobile SDK, among others. Read also: Machine Learning as a Service (MLaaS): How Google Cloud Platform, Microsoft Azure, and AWS are democratizing Artificial Intelligence Getting your hands dirty After getting a “warm-up” the next step will involve creating and training your own model. This is where you’ll be introduced to TensorFlow Lite, which is going to be your best friend throughout your journey as a machine learning mobile developer. There are many other machine learning tools coming into the market that you can make use of. These tools make building AI in mobile easier. For instance, you can use Dialogflow, a Natural Language Understanding (NLU) platform that makes it easy for developers to design and integrate conversational user interfaces into mobile apps, web applications, devices, and bots. You can then integrate it on Alexa, Cortana, Facebook Messenger, and other platforms your users are on. Read also: 7 Artificial Intelligence tools mobile developers need to know For practicing you can leverage an amazing codelab by Google, TensorFlow For Poets. It guides you through creating and training a custom image classification model. Through this codelab you will learn the basics of data collection, model optimization, and other key components involved in creating your own model. The codelab is divided into two parts. The first part covers creating and training the model, and the second part is focused on TensorFlow Lite which is the mobile version of TensorFlow that allows you to run the same model on a mobile device. Mathematics is the foundation of machine learning Love it or hate it, machine learning and Artificial Intelligence are built on mathematical principles like calculus, linear algebra, probability, statistics, and optimization. You need to learn some essential foundational concepts and the notation used to express them. There are many reasons why learning mathematics for machine learning is important. It will help you in the process of selecting the right algorithm which includes giving considerations to accuracy, training time, model complexity, number of parameters and number of features. Maths is needed when choosing parameter settings and validation strategies, identifying underfitting and overfitting by understanding the bias-variance tradeoff. Read also: Bias-Variance tradeoff: How to choose between bias and variance for your machine learning model [Tutorial] Read also: What is Statistical Analysis and why does it matter? What are the key aspects of Artificial Intelligence for mobile to keep in mind? Understanding the problem Your number one priority should be the user problem you are trying to solve. Instead of randomly integrating a machine learning model into an application, developers should understand how the model applies to the particular application or use case. This is important because you might end up building a great machine learning model with excellent accuracy rate, but if it does not solve any problem, it will end up being redundant. You must also understand that while there are many business problems which require machine learning approaches, not all of them do. Most business problems can be solved through simple analytics or a baseline approach. Data is your best friend Machine learning is dependent on data; the data that you use, and how you use it, will define the success of your machine learning model. You can also make use of thousands of open source datasets available online. Google recently launched a tool for dataset search named, Google Dataset Search which will make it easier for you to search the right dataset for your problem. Typically, there’s no shortage of data; however, the abundant existence of data does not mean that the data is clean, reliable, or can be used as intended. Data cleanliness is a huge issue. For example, a typical company will have multiple customer records for a single individual, all of which differ slightly. If the data isn’t clean, it isn’t reliable. The bottom line is, it’s a bad practice to just grabbing the data and using it without considering its origin. Read also: Best Machine Learning Datasets for beginners Decide which model to choose A machine learning algorithm is trained and the artifact that it creates after the training process is called the machine learning model. An ML model is used to find patterns in data without the developer having to explicitly program those patterns. We cannot look through such a huge amount of data and understand the patterns. Think of the model as your helper who will look through all those terabytes of data and extract knowledge and insights from the data. You have two choices here: either you can create your own model or use a pre-built model. While there are several pre-built models available, your business-specific use cases may require specialized models to yield the desired results. These off-the-shelf model may also need some fine-tuning or modification to deliver the value the app is intended to provide. Read also: 10 machine learning algorithms every engineer needs to know Thinking about resource utilization is important Artificial Intelligence-powered apps or apps, in general, should be developed with resource utilization in mind. Though companies are working towards improving mobile hardware, currently, it is not the same as what we can accomplish with GPU clusters in the cloud. Therefore, developers need to consider how the models they intend to use would affect resources including battery power and memory usage. In terms of computational resources, inferencing or making predictions is less costly than training. Inferencing on the device means that the models need to be loaded into RAM, which also requires significant computational time on the GPU or CPU. In scenarios that involve continuous inferencing, such as audio and image data which can chew up bandwidth quickly, on-device inferencing is a good choice. Learning never stops Maintenance is important, and to do that you need to establish a feedback loop and have a process and culture of continuous evaluation and improvement. A change in consumer behavior or a market trend can make a negative impact on the model. Eventually, something will break or no longer work as intended, which is another reason why developers need to understand the basics of what it is they’re adding to an app. You need to have some knowledge of how the Artificial Intelligence component that you just put together is working or how it could be made to run faster. Wrapping up Before falling for the Artificial Intelligence and machine learning hype, it’s important to understand and analyze the problem you are trying to solve. You should examine whether applying machine learning can improve the quality of the service, and decide if this improvement justifies the effort of deploying a machine learning model. If you just want a simple API endpoint and don’t want to dedicate much time in deploying a model, cloud-based web services are the best option for you. Tools like ML Kit for Firebase looks promising and seems like a good choice for startups or developers just starting out. TensorFlow Lite and Core ML are good options if you have mobile developers on your team or if you’re willing to get your hands a little dirty. Artificial Intelligence is influencing the app development process by providing us a data-driven approach for solving user problems. It wouldn't be surprising if in the near future Artificial Intelligence becomes a forerunning factor for app developers in their expertise and creativity. 10 useful Google Cloud Artificial Intelligence services for your next machine learning project [Tutorial] How Artificial Intelligence is going to transform the Data Center How Serverless computing is making Artificial Intelligence development easier

Mobile App Development has taken over and completely re-written the healthcare industry. According to Healthcare Mobility Solutions reports, the Mobile healthcare application market is expected to be worth more than $84 million by the year 2020. These mobile applications are not just limited to use by patients but are also massively used by doctors and nurses. As technology evolves, it simultaneously opens up the possibility of being used in multiple ways. Similar has been the journey of healthcare mobile app development that has originated from the latest trends in technology and has made its way to being an industry in itself. The technological trends that have helped build mobile apps for the healthcare industry are Blockchain You probably know blockchain technology, thanks to all the cryptocurrency rage in recent years. The blockchain is basically a peer-to-peer database that keeps a verified record of all transactions, or any other information that one needs to track and have it accessible to a large community. The healthcare industry can use a technology that allows it to record the medical history of patients, and store it electronically, in an encrypted form, that cannot be altered or hacked into. Blockchain succeeds where a lot of health applications fail, in the secure retention of patient data. The Internet of Things The Internet of Things (IoT) is all about connectivity. It is a way of interconnecting electronic devices, software, applications, etc., to ensure easy access and management across platforms. The loT will assist medical professionals in gaining access to valuable patient information so that doctors can monitor the progress of their patients. This makes treatment of the patient easier, and more closely monitored, as doctors can access the patient’s current profile anywhere and suggest treatment, medicine, and dosages. Augmented Reality From the video gaming industry, Augmented Reality has made its way to the medical sector. AR refers to the creation of an interactive experience of a real-world environment through superimposition of computer-generated perceptual information. AR is increasingly used to develop mobile applications that can be used by doctors and surgeons as a training experience. It stimulates a real-world experience of diagnosis and surgery, and by doing so, enhances the knowledge and its practical application that all doctors must necessarily possess. This form of training is not limited in nature, and can, therefore, simultaneously train a large number of medical practitioners. Big Data Analytics Big Data has the potential to provide comprehensive statistical information, only accessed and processed through sophisticated software. Big Data Analytics becomes extremely useful when it comes to managing the hospital’s resources and records in an efficient manner. Aside from this, it is used in the development of mobile applications that store all patient data, thus again, eliminating the need for excessive paperwork. This allows medical professionals to focus more on attending and treating the patients, rather than managing database. These technological trends have led to the development of a diverse variety of mobile applications to be used for multiple purposes in the healthcare industry. Listed below are the benefits of the mobile apps deploying these technological trends, for the professionals and the patients alike. Telemedicine Mobile applications can potentially play a crucial role in making medical services available to the masses. An example is an on-call physician on telemedicine duty. A mobile application will allow the physician to be available for a patient consult without having to operate via  PC. This will make the doctors more accessible and will bring quality treatment to the patients quickly. Enhanced Patient Engagement There are mobile applications that place all patient data – from past medical history to performance metrics, patient feedback, changes in the treatment patterns and schedules, at the push of a button on the smartphone application for the medical professional to consider and make a decision on the go. Since all data is recorded in real-time, it makes it easy for doctors to change shifts without having to explain to the next doctor the condition of the patient in person. The mobile application has all the data the supervisors or nurses need. Easy Access to Medical Facilities There are a number of mobile applications that allow patients to search for medical professionals in their area, read their reviews and feedback by other patients, and then make an online appointment if they are satisfied with the information that they find. Apart from these, they can also download and store their medical lab reports, and order medicines online at affordable prices. Easy Payment of Bills Like in every other sector, mobile applications in healthcare have made monetary transactions extremely easy. Patients or their family members, no longer need to spend hours waiting in the line to pay the bills. They can instantly pick a payment plan and pay bills immediately or add reminders to be notified when a bill is due. Therefore, it can be safely said that the revolution that the healthcare industry is undergoing and has worked in the favor of all the parties involved – Medical Professionals, Patients, Hospital Management and the Mobile App Developers. Author's Bio Ritesh Patil is the co-founder of Mobisoft Infotech that helps startups and enterprises in mobile technology. He’s an avid blogger and writes on mobile application development. He has developed innovative mobile applications across various fields such as Finance, Insurance, Health, Entertainment, Productivity, Social Causes, Education and many more and has bagged numerous awards for the same. Social Media – Twitter, LinkedIn Healthcare Analytics: Logistic Regression to Reduce Patient Readmissions How IBM Watson is paving the road for Healthcare 3.0 7 Popular Applications of Artificial Intelligence in Healthcare

Generative Adversarial Networks (GANs) are a prominent branch of Machine learning research today. As deep neural networks require a lot of data to train on, they perform poorly if data provided is not sufficient. GANs can overcome this problem by generating new and real data, without using the tricks like data augmentation. As the application of GANs in the Machine learning industry is still at the infancy level, it is considered a highly desirable niche skill. Having an added hands-on experience raises the bar higher in the job market. It can fetch you a higher pay over your colleagues and can also be the feature that sets your resume stand apart. Source: Gartner's Hype Cycle 2017  GANs along with CNNs and RNNs are a part of the in demand deep neural network experience in the industry. Here are five reasons why you should learn GANs today and how Kuntal Ganguly’s book, Learning Generative Adversarial Networks help you do just that. Kuntal is a big data analytics engineer at Amazon Web Services. He has around 7 years of experience building large-scale, data-driven systems using big data frameworks and machine learning. He has designed, developed, and deployed several large-scale distributed applications, without any assistance. Kuntal is a seasoned author with a rich set of books ranging across the data science spectrum from machine learning, deep learning, to Generative Adversarial Networks, published under his belt.[/author] The book shows how to implement GANs in your machine learning models in a quick and easy format with plenty of real-world examples and hands-on tutorials. 1. Unsupervised Learning now a cakewalk with GANs A major challenge of unsupervised learning is the massive amount of unlabelled data one needed to work through as part of data preparation. In traditional neural networks, this labeling of data is both costly and time-consuming. A creative aspect of Deep learning is now possible using Generative Adversarial Networks. Here, the neural networks are capable of generating realistic images from the real-world datasets (such as MNIST and CIFAR). GANs provide an easy way to train the DL algorithms. This is done by slashing down the amount of data required to train the neural network models, that too, with no labeling of data required. This book uses a semi-supervised approach to solve the problem of unsupervised learning for classifying images. However, this could be easily leveraged into developer’s own problem domain. 2. GANs help you change a horse into a zebra using Image style transfer https://www.youtube.com/watch?v=9reHvktowLY Turning an apple into an orange is Magic!! GANs can do this magic, without casting a  spell. Transferring Image-to-Image style, where the styling of one image is applied to the other. What GANs can do is, they can perform image-to-image translations across various domains (such as changing apple to orange or horse to zebra) using Cycle Consistent Generative Network (Cycle GANs). Detailed examples of how to turn the image of an apple to an orange using TensorFlow, and how of turn an image of a horse into a zebra using a GAN model, are given in this book.  3. GANs inputs your text and outputs an image Generative Adversarial networks can also be utilized for text-to-image synthesis. An example of this is in generating a photo-realistic image based on a caption. To do this, a dataset of images with their associated captions are given as training data. The dataset is first encoded using a hybrid neural network called the character-level convolutional Recurrent Neural network, which creates a joint representation of both in multimodal space for both the generator and the discriminator. In this book, Kuntal showcases the technique of stacking multiple generative networks to generate realistic images from textual information using StackGANs.Further, the book goes on to explain the coupling of two generative networks, to automatically discover relationships among various domains (a relationship between shoes and handbags or actor and actress) using DiscoGANs. 4. GANs + Transfer Learning = No more model generation from scratch Source: Learning Generative Adversarial Networks Data is the basis to train any Machine learning model, scarcity of which can lead to a poorly-trained model, which can have high chances of failure. Some real-life scenarios may not have sufficient data, hardware, or resources to train bigger networks in order to achieve the desired accuracy. So, is training from scratch a must-do for training the models? A well-known technique used in deep learning that adapts an existing trained model for a similar task to the task at hand is known as Transfer Learning. This book will showcase Transfer learning using some hands-on examples. Further, a combination of both Transfer learning and GANs, to generate high-resolution realistic images with facial datasets is explained. Thus, you will also understand how to create creating artistic hallucination on images beyond GAN. 5. GANs help you take Machine Learning models to Production Most Machine learning tutorials, video courses, and books, explain the training and the evaluation of the models. But how do we take this trained model to production and put it to use and make it available to customers? In this book, the author has taken an example, i.e. developing a facial correction system using an LFW dataset, to automatically correct corrupted images using your trained GAN model. This book also contains several techniques of deploying machine learning or deep learning models in production both on data centers and clouds with micro-service based containerized environments.You will also learn the way of running deep models in a serverless environment and with managed cloud services. This article just scratches the surface of what is possible with GANs and why learning it would change your thinking about deep neural networks. To know more grab your copy of Kuntal Ganguly’s book on GANs: Learning Generative Adversarial Networks.     .

The investment industry has evolved dramatically over the last several decades and continues to do so amid increased competition, technological advances, and a challenging economic environment. In this article, we will review several key trends that have shaped the investment environment in general, and the context for algorithmic trading more specifically. This article is an excerpt taken from the book 'Hands on Machine Learning for algorithmic trading' written by Stefan Jansen. The book explores the strategic perspective, conceptual understanding, and practical tools to add value from applying ML to the trading and investment process. The trends that have propelled algorithmic trading and ML to current prominence include: Changes in the market microstructure, such as the spread of electronic trading and the integration of markets across asset classes and geographies The development of investment strategies framed in terms of risk-factor exposure, as opposed to asset classes The revolutions in computing power, data-generation and management, and analytic methods The outperformance of the pioneers in algorithmic traders relative to human, discretionary investors In addition, the financial crises of 2001 and 2008 have affected how investors approach diversification and risk management and have given rise to low-cost passive investment vehicles in the form of exchange-traded funds (ETFs). Amid low yield and low volatility after the 2008 crisis, cost-conscious investors shifted $2 trillion from actively-managed mutual funds into passively managed ETFs. Competitive pressure is also reflected in lower hedge fund fees that dropped from the traditional 2% annual management fee and 20% take of profits to an average of 1.48% and 17.4%, respectively, in 2017. Let's have a look at how ML has come to play a strategic role in algorithmic trading. Factor investing and smart beta funds The return provided by an asset is a function of the uncertainty or risk associated with financial investment. An equity investment implies, for example, assuming a company's business risk, and a bond investment implies assuming default risk. To the extent that specific risk characteristics predict returns, identifying and forecasting the behavior of these risk factors becomes a primary focus when designing an investment strategy. It yields valuable trading signals and is the key to superior active-management results. The industry's understanding of risk factors has evolved very substantially over time and has impacted how ML is used for algorithmic trading. Modern Portfolio Theory (MPT) introduced the distinction between idiosyncratic and systematic sources of risk for a given asset. Idiosyncratic risk can be eliminated through diversification, but systematic risk cannot. In the early 1960s, the Capital Asset Pricing Model (CAPM) identified a single factor driving all asset returns: the return on the market portfolio in excess of T-bills. The market portfolio consisted of all tradable securities, weighted by their market value. The systematic exposure of an asset to the market is measured by beta, which is the correlation between the returns of the asset and the market portfolio. The recognition that the risk of an asset does not depend on the asset in isolation, but rather how it moves relative to other assets, and the market as a whole, was a major conceptual breakthrough. In other words, assets do not earn a risk premium because of their specific, idiosyncratic characteristics, but because of their exposure to underlying factor risks. However, a large body of academic literature and long investing experience have disproved the CAPM prediction that asset risk premiums depend only on their exposure to a single factor measured by the asset's beta. Instead, numerous additional risk factors have since been discovered. A factor is a quantifiable signal, attribute, or any variable that has historically correlated with future stock returns and is expected to remain correlated in future. These risk factors were labeled anomalies since they contradicted the Efficient Market Hypothesis (EMH), which sustained that market equilibrium would always price securities according to the CAPM so that no other factors should have predictive power. The economic theory behind factors can be either rational, where factor risk premiums compensate for low returns during bad times, or behavioral, where agents fail to arbitrage away excess returns. Well-known anomalies include the value, size, and momentum effects that help predict returns while controlling for the CAPM market factor. The size effect rests on small firms systematically outperforming large firms, discovered by Banz (1981) and Reinganum (1981). The value effect (Basu 1982) states that firms with low valuation metrics outperform. It suggests that firms with low price multiples, such as the price-to-earnings or the price-to-book ratios, perform better than their more expensive peers (as suggested by the inventors of value investing, Benjamin Graham and David Dodd, and popularized by Warren Buffet). The momentum effect, discovered in the late 1980s by, among others, Clifford Asness, the founding partner of AQR, states that stocks with good momentum, in terms of recent 6-12 month returns, have higher returns going forward than poor momentum stocks with similar market risk. Researchers also found that value and momentum factors explain returns for stocks outside the US, as well as for other asset classes, such as bonds, currencies, and commodities, and additional risk factors. In fixed income, the value strategy is called riding the yield curve and is a form of the duration premium. In commodities, it is called the roll return, with a positive return for an upward-sloping futures curve and a negative return otherwise. In foreign exchange, the value strategy is called carry. There is also an illiquidity premium. Securities that are more illiquid trade at low prices and have high average excess returns, relative to their more liquid counterparts. Bonds with higher default risk tend to have higher returns on average, reflecting a credit risk premium. Since investors are willing to pay for insurance against high volatility when returns tend to crash, sellers of volatility protection in options markets tend to earn high returns. Multifactor models define risks in broader and more diverse terms than just the market portfolio. In 1976, Stephen Ross proposed arbitrage pricing theory, which asserted that investors are compensated for multiple systematic sources of risk that cannot be diversified away. The three most important macro factors are growth, inflation, and volatility, in addition to productivity, demographic, and political risk. In 1992, Eugene Fama and Kenneth French combined the equity risk factors' size and value with a market factor into a single model that better explained cross-sectional stock returns. They later added a model that also included bond risk factors to simultaneously explain returns for both asset classes. A particularly attractive aspect of risk factors is their low or negative correlation. Value and momentum risk factors, for instance, are negatively correlated, reducing the risk and increasing risk-adjusted returns above and beyond the benefit implied by the risk factors. Furthermore, using leverage and long-short strategies, factor strategies can be combined into market-neutral approaches. The combination of long positions in securities exposed to positive risks with underweight or short positions in the securities exposed to negative risks allows for the collection of dynamic risk premiums. As a result, the factors that explained returns above and beyond the CAPM were incorporated into investment styles that tilt portfolios in favor of one or more factors, and assets began to migrate into factor-based portfolios. The 2008 financial crisis underlined how asset-class labels could be highly misleading and create a false sense of diversification when investors do not look at the underlying factor risks, as asset classes came crashing down together. Over the past several decades, quantitative factor investing has evolved from a simple approach based on two or three styles to multifactor smart or exotic beta products. Smart beta funds have crossed $1 trillion AUM in 2017, testifying to the popularity of the hybrid investment strategy that combines active and passive management. Smart beta funds take a passive strategy but modify it according to one or more factors, such as cheaper stocks or screening them according to dividend payouts, to generate better returns. This growth has coincided with increasing criticism of the high fees charged by traditional active managers as well as heightened scrutiny of their performance. The ongoing discovery and successful forecasting of risk factors that, either individually or in combination with other risk factors, significantly impact future asset returns across asset classes is a key driver of the surge in ML in the investment industry. Algorithmic pioneers outperform humans at scale The track record and growth of Assets Under Management (AUM) of firms that spearheaded algorithmic trading has played a key role in generating investor interest and subsequent industry efforts to replicate their success. Systematic funds differ from HFT in that trades may be held significantly longer while seeking to exploit arbitrage opportunities as opposed to advantages from sheer speed. Systematic strategies that mostly or exclusively rely on algorithmic decision-making were most famously introduced by mathematician James Simons who founded Renaissance Technologies in 1982 and built it into the premier quant firm. Its secretive Medallion Fund, which is closed to outsiders, has earned an estimated annualized return of 35% since 1982. DE Shaw, Citadel, and Two Sigma, three of the most prominent quantitative hedge funds that use systematic strategies based on algorithms, rose to the all-time top-20 performers for the first time in 2017 in terms of total dollars earned for investors, after fees, and since inception. DE Shaw, founded in 1988 with $47 billion AUM in 2018 joined the list at number 3. Citadel started in 1990 by Kenneth Griffin, manages $29 billion and ranks 5, and Two Sigma started only in 2001 by DE Shaw alumni John Overdeck and David Siegel, has grown from $8 billion AUM in 2011 to $52 billion in 2018. Bridgewater started in 1975 with over $150 billion AUM, continues to lead due to its Pure Alpha Fund that also incorporates systematic strategies. Similarly, on the Institutional Investors 2017 Hedge Fund 100 list, five of the top six firms rely largely or completely on computers and trading algorithms to make investment decisions—and all of them have been growing their assets in an otherwise challenging environment. Several quantitatively-focused firms climbed several ranks and in some cases grew their assets by double-digit percentages. Number 2-ranked Applied Quantitative Research (AQR) grew its hedge fund assets 48% in 2017 to $69.7 billion and managed $187.6  billion firm-wide. Among all hedge funds, ranked by compounded performance over the last three years, the quant-based funds run by Renaissance Technologies achieved ranks 6 and 24, Two Sigma rank 11, D.E. Shaw no 18 and 32, and Citadel ranks 30 and 37. Beyond the top performers, algorithmic strategies have worked well in the last several years. In the past five years, quant-focused hedge funds gained about 5.1% per year while the average hedge fund rose 4.3% per year in the same period. ML driven funds attract $1 trillion AUM The familiar three revolutions in computing power, data, and ML methods have made the adoption of systematic, data-driven strategies not only more compelling and cost-effective but a key source of competitive advantage. As a result, algorithmic approaches are not only finding wider application in the hedge-fund industry that pioneered these strategies but across a broader range of asset managers and even passively-managed vehicles such as ETFs. In particular, predictive analytics using machine learning and algorithmic automation play an increasingly prominent role in all steps of the investment process across asset classes, from idea-generation and research to strategy formulation and portfolio construction, trade execution, and risk management. Estimates of industry size vary because there is no objective definition of a quantitative or algorithmic fund, and many traditional hedge funds or even mutual funds and ETFs are introducing computer-driven strategies or integrating them into a discretionary environment in a human-plus-machine approach. Morgan Stanley estimated in 2017 that algorithmic strategies have grown at 15% per year over the past six years and control about $1.5 trillion between hedge funds, mutual funds, and smart beta ETFs. Other reports suggest the quantitative hedge fund industry was about to exceed $1 trillion AUM, nearly doubling its size since 2010 amid outflows from traditional hedge funds. In contrast, total hedge fund industry capital hit $3.21 trillion according to the latest global Hedge Fund Research report. The market research firm Preqin estimates that almost 1,500 hedge funds make a majority of their trades with help from computer models. Quantitative hedge funds are now responsible for 27% of all US stock trades by investors, up from 14% in 2013. But many use data scientists—or quants—which, in turn, use machines to build large statistical models (WSJ). In recent years, however, funds have moved toward true ML, where artificially-intelligent systems can analyze large amounts of data at speed and improve themselves through such analyses. Recent examples include Rebellion Research, Sentient, and Aidyia, which rely on evolutionary algorithms and deep learning to devise fully-automatic Artificial Intelligence (AI)-driven investment platforms. From the core hedge fund industry, the adoption of algorithmic strategies has spread to mutual funds and even passively-managed exchange-traded funds in the form of smart beta funds, and to discretionary funds in the form of quantamental approaches. The emergence of quantamental funds Two distinct approaches have evolved in active investment management: systematic (or quant) and discretionary investing. Systematic approaches rely on algorithms for a repeatable and data-driven approach to identify investment opportunities across many securities; in contrast, a discretionary approach involves an in-depth analysis of a smaller number of securities. These two approaches are becoming more similar to fundamental managers take more data-science-driven approaches. Even fundamental traders now arm themselves with quantitative techniques, accounting for $55 billion of systematic assets, according to Barclays. Agnostic to specific companies, quantitative funds trade patterns and dynamics across a wide swath of securities. Quants now account for about 17% of total hedge fund assets, data compiled by Barclays shows. Point72 Asset Management, with $12 billion in assets, has been shifting about half of its portfolio managers to a man-plus-machine approach. Point72 is also investing tens of millions of dollars into a group that analyzes large amounts of alternative data and passes the results on to traders. Investments in strategic capabilities Rising investments in related capabilities—technology, data and, most importantly, skilled humans—highlight how significant algorithmic trading using ML has become for competitive advantage, especially in light of the rising popularity of passive, indexed investment vehicles, such as ETFs, since the 2008 financial crisis. Morgan Stanley noted that only 23% of its quant clients say they are not considering using or not already using ML, down from 44% in 2016. Guggenheim Partners LLC built what it calls a supercomputing cluster for $1 million at the Lawrence Berkeley National Laboratory in California to help crunch numbers for Guggenheim's quant investment funds. Electricity for the computers costs another $1 million a year. AQR is a quantitative investment group that relies on academic research to identify and systematically trade factors that have, over time, proven to beat the broader market. The firm used to eschew the purely computer-powered strategies of quant peers such as Renaissance Technologies or DE Shaw. More recently, however, AQR has begun to seek profitable patterns in markets using ML to parse through novel datasets, such as satellite pictures of shadows cast by oil wells and tankers. The leading firm BlackRock, with over $5 trillion AUM, also bets on algorithms to beat discretionary fund managers by heavily investing in SAE, a systematic trading firm it acquired during the financial crisis. Franklin Templeton bought Random Forest Capital, a debt-focused, data-led investment company for an undisclosed amount, hoping that its technology can support the wider asset manager. We looked at how ML plays a role in different industry trends around algorithmic trading. If you want to learn more about design and execution of algorithmic trading strategies, and use cases of ML in algorithmic trading, be sure to check out the book 'Hands on Machine Learning for algorithmic trading'. Using machine learning for phishing domain detection [Tutorial] Anatomy of an automated machine learning algorithm (AutoML) 10 machine learning algorithms every engineer needs to know

In 2012, a year which feels a lot like the very early years of the era of data, Wired published this article on Narrative Science, an organization based in Chicago that uses Machine Learning algorithms to write news articles. Its founder and CEO, Kris Hammond, is a man whose enthusiasm for algorithmic possibilities is unparalleled. When asked whether an algorithm would win a Pulitzer in the next 20 years he goes further, claiming that it could happen in the next 5 years. Hammond’s excitement at what his organization is doing is not unwarranted. But his optimism certainly is. Unless 2017 is a particularly poor year for journalism and literary nonfiction, a Pulitzer for one of Narrative Science’s algorithms looks unlikely to say the least. But there are a couple of problems with Hammond’s enthusiasm. He fails to recognise the limitations of algorithms, the fact that the job of even the most intricate and complex Deep Learning algorithm is very specific is quite literally determined by the people who create it. “We are humanising the machine” he’s quoted as saying in a Guardian interview from June 2015. “Based on general ideas of what is important and a close understanding of who the audience is, we are giving it the tools to know how to tell us stories”. It’s important to notice here how he talks - it’s all about what ‘we’re’ doing. The algorithms that are central to Narrative Science’s mission are things that are created by people, by data scientists. It’s easy to read what’s going on as a simple case of the machines taking over. True, perhaps there is cause for concern among writers when he suggests that in 25 years 90% of news stories will be created by algorithms, but in actual fact there’s just a simple shift in where labour is focused. It's time to rethink algorithms We need to rethink how we view and talk about data science, Machine Learning and algorithms. We see, for example, algorithms as impersonal, blandly futuristic things. Although they might be crucial to our personalized online experiences, they are regarded as the hypermodern equivalent of the inauthentic handshake of a door to door salesman. Similarly, at the other end, the process of creating them are viewed as a feat of engineering, maths and statistics nerds tackling the complex interplay of statistics and machinery. Instead, we should think of algorithms as something creative, things that organize and present the world in a specific way, like a well-designed building. If an algorithm did indeed win a Pulitzer, wouldn’t it really be the team behind it that deserves it? When Hammond talks, for example, about “general ideas of what is important and a close understanding who the audience is”, he is referring very much to a creative process. Sure, it’s the algorithm that learns this, but it nevertheless requires the insight of a scientist, an analyst to consider these factors, and to consider how their algorithm will interact with the irritating complexity and unpredictability of reality. Machine Learning projects, then, are as much about designing algorithms as they are programming them. There’s a certain architecture, a politics that informs them. It’s all about prioritization and organization, and those two things aren’t just obvious; they’re certainly not things which can be identified and quantified. They are instead things that inform the way we quantify, the way we label. The very real fingerprints of human imagination, and indeed fallibility are in algorithms we experience every single day. Algorithms are made by people Perhaps we’ve all fallen for Hammond’s enthusiasm. It’s easy to see the algorithms as the key to the future, and forget that really they’re just things that are made by people. Indeed, it might well be that they’re so successful that we forget they’ve been made by anyone - it’s usually only when algorithms don’t work that the human aspect emerges. The data-team have done their job when no one realises they are there. An obvious example: You can see it when Spotify recommends some bizarre songs that you would never even consider listening to. The problem here isn’t simply a technical one, it’s about how different tracks or artists are tagged and grouped, how they are made to fit within a particular dataset that is the problem. It’s an issue of context - to build a great Machine Learning system you need to be alive to the stories and ideas that permeate within the world in which your algorithm operates - if you, as the data scientist lack this awareness, so will your Machine Learning project. But there have been more problematic and disturbing incidents such as when Flickr auto tags people of color in pictures as apes, due to the way a visual recognition algorithm has been trained. In this case, the issue is with a lack of sensitivity about the way in which an algorithm may work - the things it might run up against when it’s faced with the messiness of the real-world, with its conflicts, its identities, ideas and stories. The story of Solid Gold Bomb too, is a reminder of the unintended consequences of algorithms. It’s a reminder of the fact that we can be lazy with algorithms; instead of being designed with thought and care they become a surrogate for it - what’s more is that they always give us a get out clause; we can blame the machine if something goes wrong. If this all sounds like I’m simply down on algorithms, that I’m a technological pessimist, you’re wrong. What I’m trying to say is that it’s humans that are really in control. If an algorithm won a Pulitzer, what would that imply – it would mean the machines have won. It would mean we’re no longer the ones doing the thinking, solving problems, finding new ones. Data scientists are designers As the economy becomes reliant on technological innovation, it’s easy to remove ourselves, to underplay the creative thinking that drives what we do. That’s what Hammond’s doing, in his frenzied excitement about his company - he’s forgetting that it’s him and his team that are finding their way through today’s stories. It might be easier to see creativity at work when we cast our eyes towards game development and web design, but data scientists are designers and creators too. We’re often so keen to stress the technical aspects of these sort of roles that we forget this important aspect of the data scientist skillset.