Tech Guides - Data

Amazon and Microsoft, the pioneer tech giants have collaborated their efforts to bring in a compelling, easy, and powerful deep learning interface known as Gluon. If you are into physics, you must be aware of the term Gluon. Gluon is a hypothetical particle believed to be exchanged between quarks in order to bind them. If we go by the literal meaning, Gluon, similar to a glue, works as a binding agent. Having gained inspiration from this, Amazon and Microsoft have glued in their efforts to bring deep learning to a wider developer audience with the launch of Gluon. It is a simple, efficient, and compact API for deep learning. Why is Gluon essential? Any Neural network, has three important phases: First, the manual coding, where the developer explains the specific behaviour of the network. Then is the training phase where the error of the output is calculated and subsequently, the weights are adjusted. This activity requires memory and is computationally exhaustive. After the training phase, the network is used to make predictions. The process of building up a neural network is labor-intensive as well as time consuming. These networks have to be trained to parse large and complex data sets and therefore they are  usually  constructed manually. Thus, making them difficult to debug and reuse. Also, manual construction requires expertise and advanced skill-sets which are possessed by experienced data scientists. However, the reach of machine learning technique is at every doorstep now. A large number of developers are looking for solutions that can help them build deep learning models with ease and practicality without compromising on the power. Gluon is a flexible and approachable way to train and construct neural networks. It comes across as a more concise and easy-to-use programming interface providing developers the ability to quickly prototype, build, and train deep learning models without sacrificing performance. The API plays around with MXNet to reduce the complexity of deep learning making it reachable to a large number of developers. How is it different? A few compelling advantages that makes Gluon stand out: An Easy to Use API A strong differentiating feature of Gluon is  that it provides interface, in the form of an API. Making it easier for the developers to grasp and develop DL models with the help of modular components. This functionality is simpler  to comprehend than the formal neural net definition methods. Data Structure Approach Deep learning models in Gluon can be defined, flexed, and modified in a way similar to a data structure. This ability makes it a  familiar interface especially for  developers who have just recently  stepped into the machine learning world. Dynamic networks can be easily managed with Gluon as it mixes the programming models from TensorFlow (symbolic representations) and PyTorch (imperative definitions of networks). Network Defining Ability Gluon provides the ability to define the network. Thus, the dynamic adjustment of the network is possible during the definition and the training process. This essentially means that the training algorithm and the neural model can inform one another. Due to this, developers can make use of standard programming structures to build, and can also use sophisticated algorithms and models to advance neural nets. High Speed Training Friendly APIs and flexible approaches are all great, but they shouldn't be incurred at the cost of training speed. Gluon is better than the manual approach as it can perform  all of the tasks without compromising on performance while providing abstractions without losing out on training speed. This is because, Gluon blends the formal definitions and specific details of the network under the hood of a concise API, allowing users to implement models, rather than doing tasks like compiler optimizations manually. Easy algorithmic implementations using Gluon Gluon supports a wide range of prebuilt and optimized components for building neural networks. Developers can build deep learning models using the MXNet framework in the Gluon interface. Gluon allows building neural nets from predefined layers. It  can also keep a note of when to record or not to record the computation graph. It can invoke highly optimized layers written in C++. Training of parallel data can also be accomplished easily. As compared to other interfaces, Gluon can run code, faster on both CPUs and GPUs.  Also, movement from one to multiple devices and initializing network parameters over them is pretty easy. Even for a simple problem like a linear regression model, Gluon can help in writing quick, and clean code. For linear regression, it eliminates the need of allocating parameters individually, implementing a stochastic gradient descent, or defining a loss function. Subsequently, it also reduces the workload required for multiclass logistic regression. On similar terms, Gluon can be used to transform the logic of a logistic regression model to a multilayer perceptron with a few additional lines of code. A convolutional neural network can also be designed easily and concisely. Limitations: On the flip side In spite of Gluon being easy, compact and efficient, it has certain limitations. Currently it  is available on Apache MXNet, and is awaiting a release in the Microsoft Cognitive Toolkit. However, not much has been known about other frameworks. For instance, it currently lacks support for the two most widely used deep learning frameworks, Caffe2 and TensorFlow. This could pose an issue for Gluon because most interfaces released, provide integration with multiple frameworks. Ultimately, it boils down to the project requirements including the model requirements and the difficulty associated with building networks from a particular tool. So, for a computer vision project people would prefer using Caffe. While TensorFlow is popular  among the developers because of the existing community, the complex nature of the platform, makes a digestible deep learning interface like Gluon highly appreciated.  Hence, each framework performs on its own tradeoffs. Conclusion Gluon comes as a boon, for both experienced data scientists and nascent developers alike. For developers, this interface, models like a data structure, providing more familiarity.  On the other side, for researchers and data scientists, it provides an interface to build prototypes quickly and easily for complex neural networks, without sacrificing training speeds. Overall, Gluon will accelerate the development of advanced neural networks and models, resulting in robust artificial intelligence based applications.

Today, machine learning in Python has become almost synonymous with scikit-learn. The "Big Bang" moment for scikit-learn was in 2007 when a gentleman named David Cournapeau decided to write this project as part of Google Summer of Code 2007. Let's take a moment to thank him. Matthieu Brucher later came on board and developed it further as part of his thesis. From that point on, sklearn never looked back. In 2010, the prestigious French research organization INRIA took ownership of the project with great developers like Gael Varoquaux, Alexandre Gramfort et al. starting work on it. Here's the oldest pull request I could find in sklearn’s repository. The title says "we're getting there"! Starting from there to today where sklearn receives funding and support from Google, Telecom ParisTech and Columbia University among others, it surely must’ve been quite a journey. Sklearn is an open source library which uses the BSD license. It is widely used in industry as well as in academia. It is built on Numpy, Scipy and Matplotlib while also having wrappers around various popular libraries such LIBSVM. Sklearn can be used “out of the box” after installation. Can I trust scikit-learn? Scikit-learn, or sklearn, is a very active open source project having brilliant maintainers. It is used worldwide by top companies such as Spotify, booking.com and the like. That it is open source where anyone can contribute might make you question the integrity of the code, but from the little experience I have contributing to sklearn, let me tell you only very high-quality code gets merged. All pull requests have to be affirmed by at least two core maintainers of the project. Every code goes through multiple iterations. While this can be time-consuming for all the parties involved, such regulations ensure sklearn’s compliance with the industry standard at all times. You don’t just build a library that’s been awarded the “best open source library” overnight! How can I use scikit-learn? Sklearn can be used for a wide variety of use-cases ranging from image classification to music recommendation to classical data modeling. Scikit-learn in various industries: In the Image classification domain, Sklearn’s implementation of K-Means along with PCA has been used for handwritten digit classification very successfully in the past. Sklearn has also been used for facial/ faces recognition using SVM with PCA. Image segmentation tasks such as detecting Red Blood Corpuscles or segmenting the popular Lena image into sections can be done using sklearn. A lot of us here use Spotify or Netflix and are awestruck by their recommendations. Recommendation engines started off with the collaborative filtering algorithm. It basically says “if people like me like something, I’ll also most probably like that.” To find out users with similar tastes, a KNN algorithm can be used which is available in sklearn. You can find a good demonstration of how it is used for music recommendation here. Classical data modeling can be bolstered using sklearn. Most people generally start their kaggle competitive journeys with the titanic challenge. One of the better tutorials out there on starting out is by dataquest and generally acts as a good introduction on how to use pandas and sklearn (a lethal combination!) for data science. It uses the robust Logistic Regression, Random Forest and the Ensembling modules to guide the user. You will be able to experience the user-friendliness of sklearn first hand while completing this tutorial. Sklearn has made machine learning literally a matter of importing a package. Sklearn also helps in Anomaly detection for highly imbalanced datasets (99.9% to 0.1% in credit card fraud detection) through a host of tools like EllipticEnvelope and OneClassSVM. In this regard, the recently merged IsolationForest algorithm especially works well in higher dimensional sets and has very high performance. Other than that, sklearn has implementations of some widely used algorithms such as linear regression, decision trees, SVM and Multi Layer Perceptrons (Neural Networks) to name a few. It has around 39 models in the “linear models” module itself! Happy scrolling here! Most of these algorithms can run very fast compared to raw python code since they are implemented in Cython and use Numpy and Scipy (which in-turn use C) for low-level computations. How is sklearn different from TensorFlow/MLllib? TensorFlow is a popular library to implement deep learning algorithms (since it can utilize GPUs). But while it can also be used to implement machine learning algorithms, the process can be arduous. For implementing logistic regression in TensorFlow, you will first have to “build” the logistic regression algorithm using a computational graph approach. Scikit-learn, on the other hand, provides the same algorithm out of the box however with the limitation that it has to be done in memory. Here's a good example of how LogisticRegression is done in Tensorflow. Apache Spark’s MLlib, on the other hand, consists of algorithms which can be used out of the box just like in Sklearn, however, it is generally used when the ML task is to be performed in a distributed setting. If your dataset fits into RAM, Sklearn would be a better choice for the task. If the dataset is massive, most people generally prototype on a small subset of the dataset locally using Sklearn. Once prototyping and experimentation are done, they deploy in the cluster using MLlib. Some sklearn must-knows Scikit-learn can be used for three different kinds of problems in machine learning namely supervised learning, unsupervised learning and reinforcement learning (ahem AlphaGo). Unsupervised learning happens when one doesn’t have ‘y’ labels in their dataset. Dimensionality reduction and clustering are typical examples. Scikit-learn has implementations of variations of the Principal Component Analysis such as SparsePCA, KernelPCA, and IncrementalPCA among others. Supervised learning covers problems such as spam detection, rent prediction etc. In these problems, the ‘y’ tag for the dataset is present. Models such as Linear regression, random forest, adaboost etc. are implemented in sklearn. From sklearn.linear_models import LogisticRegression Clf = LogisticRegression().fit(train_X, train_y) Preds = Clf.predict(test_X) Model evaluation and analysis Cross-validation, grid search for parameter selection and prediction evaluation can be done using the Model Selection and Metrics module which implements functions such as cross_val_score and f1_score respectively among others. They can be used as such: Import numpy as np From model_selection import cross_val_score From sklearn.metrics import f1_score Cross_val_avg = np.mean(cross_val_score(clf, train_X, train_y, scoring=’f1’)) # tune your parameters for better cross_val_score # for model results on a certain classification problem F_measure = f1_score(test_y, preds) Model Saving Simply pickle your model using pickle.save and it is ready to be distributed and deployed! Hence a whole machine learning pipeline can be built easily using sklearn. Finishing Remarks There are many good books out there talking about machine learning, but in context to Python,  Sebastian Raschka`s  (one of the core developers on sklearn) recently released his book titled “ Python Machine Learning” and it’s in great demand. Another great blog you could follow is Erik Bernhardsson’s blog. Along with writing about machine learning, he also discusses software development and other interesting ideas. Do subscribe to the scikit-learn mailing list as well. There are some very interesting questions posted there and a lot of learnings to take home. The machine learning subreddit also collates information from a lot of different sources and is thus a good place to find useful information. Scikit-learn has revolutionized the machine learning world by making it accessible to everyone. Machine learning is not like black magic anymore. If you use scikit-learn and like it, do consider contributing to sklearn. There is a huge clutter of open issues and PRs on the sklearn GitHub page. Scikit-learn needs contributors! Have a look at this page to start contributing. Contributing to a library is easily the best way to learn it! [author title="About the Author"]Devashish Deshpande started his foray into data science and machine learning in 2015 with an online course when the question of how machines can learn started intriguing him. He pursued more online courses as well as courses in data science during his undergrad. In order to gain practical knowledge he started contributing to open source projects beginning with a small pull request in Scikit-Learn. He then did a summer project with Gensim and delivered workshops and talks at PyCon France and India in 2016. Currently, Devashish works in the data science team at belong.co, India. Here's the link to his GitHub profile.[/author]

Tommy’s a brilliant young man, who loves programming. He’s so occupied with computers that he hardly has any time for friends. Tommy programs a very intelligent robot called Polly, using Artificial Intelligence, so that he has someone to talk to. One day, Tommy gets hurt real bad about something and needs someone to talk to. He rushes home to talk to Polly and pours out his emotions to her. To his disappointment, Polly starts giving him advice like she does for any other thing. She doesn’t understand that he needs someone to “feel” what he’s feeling rather than rant away on what he should or shouldn’t be doing. He naturally feels disconnected from Polly. My Friend doesn’t get me Have you ever wondered what it would be like to have a robot as a friend? I’m thinking something along the lines of Siri. Siri’s pretty good at holding conversations and is quick witted too. But Siri can’t understand your feelings or emotions, neither can “she” feel anything herself. Are we missing that “personality” from the artificial beings that we’re creating? Even if you talk about chatbots, although we gain through convenience, we lose the emotional aspect, especially at a time when expressive communication is the most important. Do we really need it? I remember watching the Terminator, where Arnie asks John, “Why do you cry?” John finds it difficult to explain to him, why humans cry. The fact is though, that the machine actually understood there was something wrong with the human, thanks to the visual effects associated with crying. We’ve also seen some instances of robots or AI analysing sentiment through text processing as well. But how accurate is this? How would a machine know when a human is actually using sarcasm? What if John was faking it and could cry at the drop of a hat or he just happened to be chopping onions? That’s food for thought now. On the contrary, you might wonder though, do we really want our machines to start analysing our emotions? What if they take advantage of our emotional state? Well, that’s a bit of a far fetched thought and what we need to understand is that it’s necessary for robots to gauge a bit of our emotions to enhance the experience of interacting with them. There are several wonderful applications for such a technology. For instance, Marketing organisations could use applications that detect users facial expressions when they look at a new commercial to gauge their “interest”. It could also be used by law enforcement as a replacement to the polygraph. Another interesting use case would be to help autism affected individuals understand the emotions of others better. The combination of AI and EI could find a tonne of applications right from cars that can sense if the driver is tired or sleepy and prevent an accident by pulling over, to a fridge that can detect if you’re stressed and lock itself, to prevent you from binge eating! Recent Developments in Emotional Intelligence There are several developments happening from the past few years, in terms of building systems that understand emotions. Pepper, a Japanese robot, for instance, can tell feelings such as joy, sadness and anger, and respond by playing you a song. A couple of years ago, Microsoft released a tool, the Emotion API, that could breakdown a person’s emotions based only on their picture. Physiologists, Neurologists and Psychologists, have collaborated with engineers to find measurable indicators of human emotion that can be taught to computers to look out for. There are projects that have attempted to decode facial expressions, the pitch of our voices, biometric data such as heart rate and even our body language and muscle movements. Bronwyn van der Merwe, General Manager of Fjord in the Asia Pacific region revealed that big companies like Amazon, Google and Microsoft are hiring comedians and script writers in order to harness the human-like aspect of AI by inducing personality into their technologies. Jerry, Ellen, Chris, Russell...are you all listening? How it works Almost 40% of our emotions are conveyed through tone of voice and the rest is read through facial expressions and gestures we make. An enormous amount of data is collected from media content and other sources and is used as training data for algorithms to learn human facial expressions and speech. One type of learning used is Active Learning or human-assisted machine learning. This is a kind of supervised learning, where the learning algorithm is able to interactively query the user to obtain new data points or an output. Situations might exist where unlabeled data is plentiful but manually labeling the data is expensive. In such a scenario, learning algorithms can query the user for labels. Since the algorithm chooses the examples, the number of examples to learn a concept turns out to be lower than what is required for usual supervised learning. Another approach is to use Transfer Learning, a method that focuses on storing the knowledge that’s gained while solving one problem and then applying it to a different but related problem. For example, knowledge gained while learning to recognize fruits could apply when trying to recognize vegetables. This works by analysing a video for facial expressions and then transfering that learning to label speech modality. What’s under the hood of these machines? Powerful robots that are capable of understanding emotions would most certainly be running Neural Nets under the hood. Complementing the power of these Neural Nets are beefy CPUs and GPUs on the likes of the Nvidia Titan X GPU and Intel Nervana CPU chip. Last year at NIPS, amongst controversial body shots and loads of humour filled interactions, Kory Mathewson and Piotr Mirowski entertained audiences with A.L.Ex and Pyggy, two AI robots that have played alongside humans in over 30 shows. These robots introduce audiences to the “comedy of speech recognition errors” by blabbering away to each other as well as to humans. Built around a Recurrent Neural Network that’s trained on dialogue from thousands of films, A.L.Ex. communicates with human performers, audience participants, and spectators through speech recognition, voice synthesis, and video projection. A.L.E.x is written in Torch and Lua code and has a word vocabulary of 50,000 words that have been extracted from 102,916 movies and it is built on an RNN with Long-Short Term Memory architecture and 512 dimensional layers. The unconquered challenges today The way I see it, there are broadly 3 challenge areas that AI powered robots face in this day: Rationality and emotions: AI robots need to be fed with initial logic by humans, failing which, they cannot learn on their own. They may never have the level of rationality or the breadth of emotions to take decisions the way humans do. Intuition, strategic thinking and emotions: Machines are incapable of thinking into the future and taking decisions the way humans can. For example, not very far into the future, we might have an AI powered dating application that measures a subscriber’s interest level while chatting with someone. It might just rate the interest level lower, if the person is in a bad mood due to some other reason. It wouldn’t consider the reason behind the emotion and whether it was actually linked to the ongoing conversation. Spontaneity, empathy and emotions: It may be years before robots are capable of coming up with a plan B, the way humans do. Having a contingency plan and implementing it in an emotional crisis is something that AI fails at accomplishing. For example, if you’re angry at something and just want to be left alone, your companion robot might just follow what you say without understanding your underlying emotion, while an actual human would instantly empathise with your situation and rather try to be there for you. Bronwyn van der Merwe said, "As human beings, we have contextual understanding and we have empathy, and right now there isn't a lot of that built into AI. We do believe that in the future, the companies that are going to succeed will be those that can build into their technology that kind of an understanding". What’s in store for the future If you ask me, right now we’re on the highway to something really great. Yes, there are several aspects that are unclear about AI and robots making our lives easier vs disrupting them, but as time passes, science is fitting the pieces of the puzzle together to bring about positive changes in our lives. AI is improving on the emotional front as I write, although there are clearly miles to go. Companies like Affectiva are pioneering emotion recognition technology and are working hard to improve the way AI understands human emotions. Biggies like Microsoft had been working on bringing in emotional intelligence into their AI since before 2015 and have come a long way since then. Perhaps, in the next Terminator movie, Arnie might just comfort a weeping Sarah Connor, saying, “Don’t cry, Sarah dear, he’s not worth it”, or something of the sort. As a parting note and just for funsies, here’s a final question for you, “Can you imagine a point in the future when robots have such high levels of EQ, that some of us might consider choosing them as a partner over humans?”

[box type="note" align="" class="" width=""]This article is an excerpt taken from a book Big Data Analytics with Java written by Rajat Mehta. This book will help you learn to perform big data analytics tasks using machine learning concepts such as clustering, recommending products, data segmentation and more. [/box] With this post, you will learn what is sentiment analysis and how it is used to analyze emotions associated within the text. You will also learn key NLP concepts such as Tokenization, stemming among others and how they are used for sentiment analysis. What is sentiment analysis? One of the forms of text analysis is sentimental analysis. As the name suggests this technique is used to figure out the sentiment or emotion associated with the underlying text. So if you have a piece of text and you want to understand what kind of emotion it conveys, for example, anger, love, hate, positive, negative, and so on you can use the technique sentimental analysis. Sentimental analysis is used in various places, for example: To analyze the reviews of a product whether they are positive or negative This can be especially useful to predict how successful your new product is by analyzing user feedback To analyze the reviews of a movie to check if it's a hit or a flop Detecting the use of bad language (such as heated language, negative remarks, and so on) in forums, emails, and social media To analyze the content of tweets or information on other social media to check if a political party campaign was successful or not  Thus, sentimental analysis is a useful technique, but before we see the code for our sample sentimental analysis example, let's understand some of the concepts needed to solve this problem. [box type="shadow" align="" class="" width=""]For working on a sentimental analysis problem we will be using some techniques from natural language processing and we will be explaining some of those concepts.[/box] Concepts for sentimental analysis Before we dive into the fully-fledged problem of analyzing the sentiment behind text, we must understand some concepts from the NLP (Natural Language Processing) perspective. We will explain these concepts now. Tokenization From the perspective of machine learning one of the most important tasks is feature extraction and feature selection. When the data is plain text then we need some way to extract the information out of it. We use a technique called tokenization where the text content is pulled and tokens or words are extracted from it. The token can be a single word or a group of words too. There are various ways to extract the tokens, as follows: By using regular expressions: Regular expressions can be applied to textual content to extract words or tokens from it. By using a pre-trained model: Apache Spark ships with a pre-trained model (machine learning model) that is trained to pull tokens from a text. You can apply this model to a piece of text and it will return the predicted results as a set of tokens. To understand a tokenizer using an example, let's see a simple sentence as follows: Sentence: "The movie was awesome with nice songs" Once you extract tokens from it you will get an array of strings as follows: Tokens: ['The', 'movie', 'was', 'awesome', 'with', 'nice', 'songs'] [box type="shadow" align="" class="" width=""]The type of tokens you extract depends on the type of tokens you are interested in. Here we extracted single tokens, but tokens can also be a group of words, for example, 'very nice', 'not good', 'too bad', and so on.[/box] Stop words removal Not all the words present in the text are important. Some words are common words used in the English language that are important for the purpose of maintaining the grammar correctly, but from conveying the information perspective or emotion perspective they might not be important at all, for example, common words such as is, was, were, the, and so. To remove these words there are again some common techniques that you can use from natural language processing, such as: Store stop words in a file or dictionary and compare your extracted tokens with the words in this dictionary or file. If they match simply ignore them. Use a pre-trained machine learning model that has been taught to remove stop words. Apache Spark ships with one such model in the Spark feature package. Let's try to understand stop words removal using an example: Sentence: "The movie was awesome" From the sentence we can see that common words with no special meaning to convey are the and was. So after applying the stop words removal program to this data you will get: After stop words removal: [ 'movie', 'awesome', 'nice', 'songs'] [box type="shadow" align="" class="" width=""]In the preceding sentence, the stop words the, was, and with are removed.[/box] Stemming Stemming is the process of reducing a word to its base or root form. For example, look at the set of words shown here: car, cars, car's, cars' From our perspective of sentimental analysis, we are only interested in the main words or the main word that it refers to. The reason for this is that the underlying meaning of the word in any case is the same. So whether we pick car's or cars we are referring to a car only. Hence the stem or root word for the previous set of words will be: car, cars, car's, cars' => car (stem or root word) For English words again you can again use a pre-trained model and apply it to a set of data for figuring out the stem word. Of course there are more complex and better ways (for example, you can retrain the model with more data), or you have to totally use a different model or technique if you are dealing with languages other than English. Diving into stemming in detail is beyond the scope of this book and we would encourage readers to check out some documentation on natural language processing from Wikipedia and the Stanford nlp website. [box type="shadow" align="" class="" width=""]To keep the sentimental analysis example in this book simple we will not be doing stemming of our tokens, but we will urge the readers to try the same to get better predictive results.[/box] N-grams Sometimes a single word conveys the meaning of context, other times a group of words can convey a better meaning. For example, 'happy' is a word in itself that conveys happiness, but 'not happy' changes the picture completely and 'not happy' is the exact opposite of 'happy'. If we are extracting only single words then in the example shown before, that is 'not happy', then 'not' and 'happy' would be two separate words and the entire sentence might be selected as positive by the classifier However, if the classifier picks the bi-grams (that is, two words in one token) in this case then it would be trained with 'not happy' and it would classify similar sentences with 'not happy' in it as 'negative'. Therefore, for training our models we can either use a uni-gram or a bi-gram where we have two words per token or as the name suggest an n-gram where we have 'n' words per token, it all depends upon which token set trains our model well and it improves its predictive results accuracy. To see examples of n-grams refer to the following table:   Sentence The movie was awesome with nice songs Uni-gram ['The', 'movie', 'was', 'awesome', 'with', 'nice', 'songs'] Bi-grams ['The movie', 'was awesome', 'with nice', 'songs'] Tri-grams ['The movie was', 'awesome with nice', 'songs']   For the purpose of this case study we will be only looking at unigrams to keep our example simple. By now we know how to extract words from text and remove the unwanted words, but how do we measure the importance of words or the sentiment that originates from them? For this there are a few popular approaches and we will now discuss two such approaches. Term presence and term frequency Term presence just means that if the term is present we mark the value as 1 or else 0. Later we build a matrix out of it where the rows represent the words and columns represent each sentence. This matrix is later used to do text analysis by feeding its content to a classifier. Term Frequency, as the name suggests, just depicts the count or occurrences of the word or tokens within the document. Let's refer to the example in the following table where we find term frequency:   Sentence The movie was awesome with nice songs and nice dialogues. Tokens (Unigrams only for now) ['The', 'movie', 'was', 'awesome', 'with', 'nice', 'songs', 'and', 'dialogues'] Term Frequency ['The = 1', 'movie = 1', 'was = 1', 'awesome = 1', 'with = 1', 'nice = 2', 'songs = 1', 'dialogues = 1']   As seen in the preceding table, the word 'nice' is repeated twice in the preceding sentence and hence it will get more weight in determining the opinion shown by the sentence. Bland term frequency is not a precise approach for the following reasons: There could be some redundant irrelevant words, for example, the, it, and they that might have a big frequency or count and they might impact the training of the model There could be some important rare words that could convey the sentiment regarding the document yet their frequency might be low and hence they might not be inclusive for greater impact on the training of the model Due to this reason, a better approach of TF-IDF is chosen as shown in the next sections. TF-IDF TF-IDF stands for Term Frequency and Inverse Document Frequency and in simple terms it means the importance of a term to a document. It works using two simple steps as follows: It counts the number of terms in the document, so the higher the number of terms the greater the importance of this term to the document. Counting just the frequency of words in a document is not a very precise way to find the importance of the words. The simple reason for this is there could be too many stop words and their count is high so their importance might get elevated above the importance of real good words. To fix this, TF-IDF checks for the availability of these stop words in other documents as well. If the words appear in other documents as well in large numbers that means these words could be grammatical words such as they, for, is, and so on, and TF-IDF decreases the importance or weight of such stop words. Let's try to understand TF-IDF using the following figure: As seen in the preceding figure, doc-1, doc-2, and so on are the documents from which we extract the tokens or words and then from those words we calculate the TF-IDFs. Words that are stop words or regular words such as for , is, and so on, have low TF-IDFs, while words that are rare such as 'awesome movie' have higher TF-IDFs. TF-IDF is the product of Term Frequency and Inverse document frequency. Both of them are explained here: Term Frequency: This is nothing but the count of the occurrences of the words in the document. There are other ways of measuring this, but the simplistic approach is to just count the occurrences of the tokens. The simple formula for its calculation is:      Term Frequency = Frequency count of the tokens Inverse Document Frequency: This is the measure of how much information the word provides. It scales up the weight of the words that are rare and scales down the weight of highly occurring words. The formula for inverse document frequency is: TF-IDF: TF-IDF is a simple multiplication of the Term Frequency and the Inverse Document Frequency. Hence: This simple technique is very popular and it is used in a lot of places for text analysis. Next let's look into another simple approach called bag of words that is used in text analytics too. Bag of words As the name suggests, bag of words uses a simple approach whereby we first extract the words or tokens from the text and then push them in a bag (imaginary set) and the main point about this is that the words are stored in the bag without any particular order. Thus the mere presence of a word in the bag is of main importance and the order of the occurrence of the word in the sentence as well as its grammatical context carries no value. Since the bag of words gives no importance to the order of words you can use the TF-IDFs of all the words in the bag and put them in a vector and later train a classifier (naïve bayes or any other model) with it. Once trained, the model can now be fed with vectors of new data to predict on its sentiment. Summing it up, we have got you well versed with sentiment analysis techniques and NLP concepts in order to apply sentimental analysis. If you want to implement machine learning algorithms to carry out predictive analytics and real-time streaming analytics you can refer to the book Big Data Analytics with Java.    

Machine learning is a useful and effective tool to have when it comes to building prediction models or to build a useful data structure from an avalanche of data. Many ML algorithms are in use today for a variety of real-world use cases. Given a sample dataset, a machine learning model can give predictions with only certain accuracy, which largely depends on the quality of the training data fed to it. Is there a way to increase the prediction accuracy by somehow involving humans in the process? The answer is yes, and the solution is called as ‘Interactive Machine Learning’. Why we need interactive machine learning As we already discussed above, a model can give predictions only as good as the quality of the training data fed to it. If the quality of the training data is not good enough, the model might: Take more time to learn and then give accurate predictions Quality of predictions will be very poor This challenge can be overcome by involving humans in the machine learning process. By incorporating human feedback in the model training process, it can be trained faster and more efficiently to give more accurate predictions. In the widely adopted machine learning approaches, including supervised and unsupervised learning or even active learning for that matter, there is no way to include human feedback in the training process to improve the accuracy of predictions. In case of supervised learning, for example, the data is already pre-labelled and is used without any actual inputs from the human during the training process. For this reason alone, the concept of interactive machine learning is seen by many machine learning and AI experts as a breakthrough. How interactive machine learning works Machine Learning Researchers Teng Lee, James Johnson and Steve Cheng have suggested a novel way to include human inputs to improve the performance and predictions of the machine learning model. It has been called as the ‘Transparent Boosting Tree’ algorithm, which is a very interesting approach to combine the advantages of machine learning and human inputs in the final decision making process. The Transparent Boosting Tree, or TBT in short, is an algorithm that would visualize the model and the prediction details of each step in the machine learning process to the user, take his/her feedback, and incorporate it into the learning process. The ML model is in charge of updating the assigned weights to the inputs, and filtering the information shown to the user for his/her feedback. Once the feedback is received, it can be incorporated by the ML model as a part of the learning process, thus improving it. A basic flowchart of the interactive machine learning process is as shown: Interactive Machine Learning More in-depth information on how interactive machine learning works can be found in their paper. What can Interactive machine learning do for businesses With the rising popularity and applications of AI across all industry verticals, humans may have a key role to play in the learning process of an algorithm, apart from just coding it. While observing the algorithm’s own outputs or evaluations in the form of visualizations or plain predictions, humans can suggest way to to improve that prediction by giving feedback in the form of inputs such as labels, corrections or rankings. This helps the models in two ways: Increases the prediction accuracy Time taken for the algorithm to learn is shortened considerably Both the advantages can be invaluable to businesses, as they look to incorporate AI and machine learning in their processes, and look for faster and more accurate predictions. Interactive Machine Learning is still in its nascent stage and we can expect more developments in the domain to surface in the coming days. Once production-ready, it will undoubtedly be a game-changer. Read more Active Learning: An approach to training machine learning models efficiently Anatomy of an automated machine learning algorithm (AutoML) How machine learning as a service is transforming cloud

Tools to perform Deep Learning tasks are in abundance. You have programming languages that are adapted for the job or those specifically created to get the job done. Then, you have several frameworks and libraries which allow data scientists to design systems that sift through tonnes of data and learn from it. But a major challenge for all tools lies in tackling two primary issues: The size of the data The speed of computation Now, with petabytes and exabytes of data, it’s become way more taxing for researchers to handle. Take image processing for example. ImageNet itself is such a massive dataset consisting of trillions of images from several distinct classes that tackling this scale is a serious lip-biting affair. The speed at which researchers are able to get actionable insights from the data is also an important factor. Powerful hardware like multi-core GPUs, rumbling with raw power and begging to be tamed, have waltzed into the mosh pit of big data. You may try to humble these mean machines with old school machine learning stacks like R, SciPy or NumPy, but in vain. So, the deep learning community developed several powerful libraries to solve this problem, and they succeeded to an extent. But two major problems still existed - the frameworks failed to solve the problems of efficiency and flexibility together. This is where a one-of-a-kind, powerful, and flexible library like MXNet rises up to the challenge and makes developers’ lives a lot easier. What is MXNet? MXNet sits happy at over 10k stars on Github and has recently been inducted into the Apache Software Foundation. It focuses on accelerating the development and deployment of Deep Neural Networks at scale. This means exploiting the full potential of multi-core GPUs to process tonnes of data at blazing fast speeds. We’ll take a look at some of MXNet’s most interesting features over the next few minutes. Why is MXNET so good? Efficient MXNet is backed by a C++ backend which allows it to be extremely fast on even a single machine. It allows for automatically parallelizing computation across devices as well as synchronising the computation when multithreading is introduced. Moreover, the support for linear scaling means that not only the number of GPUs can be scaled, but MXNet also supports heavily distributed computing by scaling the number of machines as well. Moreover, MXNet has a graph optimisation layer that sits on top of a dynamic dependency scheduler, which enhances memory efficiency and speed. Extremely portable MXNet is extremely portable, especially as it can be programmed in umpteen languages such as C++, R, Python, Julia, JavaScript, Go, and more. It’s widely supported across most operating systems like Linux, Windows, iOS as well as Android making it multi-platform, including low level platforms. Moreover, it works well in cloud environments, like AWS - one of the reasons AWS has officially adopted MXNet as its deep learning framework of choice. You can now run MXNet from the Deep Learning AMI. Great for data visualization MXNet uses not only the mx.viz.plot_network method for visualising neural networks but it also has in-built support for Graphviz, to visualise neural nets as a computation graph. Check Joseph Paul Cohen’s blog for a great side-by-side visualisation of CNN architectures in MXNet. Alternatively, you could strip the TensorBoard off TensorFlow and use it with MXNet. Jump here for more info on that. Flexible MXNet supports both imperative and declarative/symbolic styles of programming, allowing you to blend both styles for increased efficiency. Libraries like Numpy and Torch support plain imperative programming, while TensorFlow, Theano, and Caffe support plain declarative programming. You can get a closer look at what these styles actually are here. MXNet is the only framework so far that mixes both styles to maximise efficiency and productivity. In-built profiler MXNet comes packaged with an in-built profiler that lets you profile execution times, layer-by-layer, in the network. Now, while you’ll be interested in using your own general profiling tools like gprof and nvprof to profile at kernel, function, or instruction level, the in-built profiler is specifically tuned to provide detailed information at a symbol or operator level. Limitations While MXNet has a host of attractive features that explain why it has earned public admiration, it has its share of limitations just like any other popular tool. One of the biggest issues encountered with MXNet is that it tends to give varied results when compile settings are modified. For example, a model would work well with cuDNN3 but wouldn’t with cuDNN4. To overcome issues like this, you might have to spend some time on forums. Moreover, it is a daunting task to write your own operators or layers in C++, to achieve efficiency. Although, with v0.9, the official documentation mentions that it has become easier. Finally, the documentation is introductory and is not organised well enough to create custom operators or to perform other advanced tasks. So, should I use MXNet? MXNet is the new kid on the block that supports modern deep learning models like CNNs and LSTMs. It boasts of immense speed, scalability, and flexibility to solve your deep learning problems and consumes as little as 4 Gigs of memory when running deep networks with almost a thousand layers. The core library including its dependencies can mash into a single C++ source file, which can be compiled on both Android and iOS, as well as a browser with the JavaScript extensions. But like all other libraries, it has it’s own hurdles, which aren’t that life threatening enough, to prevent one from getting the job done, and a good one at that! Is that enough to get you excited and start using MXNet? Go get working then! And don’t forget to tell us your experiences of working with MXNet.

[box type="shadow" align="" class="" width=""]Matt Kowalski (in Gravity): Houston, in the blind.[/box] Being an oncologist is a difficult job. Every year, 50,000 research papers are published on just Oncology. If an Oncologist were to read every one of them, it will take nearly 29 hours of reading every workday to stay updated on this plethora of information. Added to this is the challenge of dealing with nearly 1000 patients every year. Needless to say, a modern-day physician is bombarded with information that doubles every three years. This wide gap between the availability of information and the ability to access it in a manner that’s practically useful is simply getting wider. No wonder doctors and other medical practitioners can feel overwhelmed and lost in space, sometimes! [box type="shadow" align="" class="" width=""]Mission Control: Shariff, what's your status? Shariff: Nearly there.[/box] Advances in the field of Big Data and cognitive computing are helping make strides in solving this kind of pressing problems facing the healthcare industry. IBM Watson is at the forefront of solving such scenarios and as time goes by the system will only become more robust. From a strict technological standpoint, the new applications of Watson are impressive and groundbreaking: The system is capable of combing through 600,000 pieces of medical evidence, 2 million pages of text from 42 medical journals and clinical trials in the area of oncology research, and 1.5 million patient records to provide on-the-spot treatment recommendations to health care providers. According to IBM, more than 90 percent of the nurses who have worked with Watson follow the guidance the system gives to them. - Infoworld Watson, who? IBM Watson is an interactive expert system that uses cognitive computing, natural language processing, and evidence-based learning to arrive at answers to questions posed to it by its users in plain English. Watson doesn’t just stop with hypotheses generation but goes ahead and proposes a list of recommendations to the user. Let’s pause and try to grasp what this means for a healthcare professional. Imagine a doctor typing in his/her iPad “A cyst found in the under-arm of the patient and biopsy suggesting non-Hodgkin's Lymphoma”. With so many cancers and alternative treatments available to treat them, to zero down on the right cure at the right time is a tough job for an oncologist. IBM Watson taps into the collective wisdom of Oncology experts - practitioners, researchers, and academicians across the globe to understand the latest advances happening inside the rapidly evolving field of Oncology. It then culls out information most relevant to the patient’s particular situation after considering their medical history. Within minutes, Watson then comes up with various tailored approaches that the doctor can adopt to treat his/her patient. Watson can help healthcare professionals narrow down on the right diagnosis, take informed and timely decisions and put in place treatment plans for their patients. All the doctor has to do is ask a question while mentioning the symptoms a patient is experiencing. This question-answer format is pretty revolutionary in that it can completely reshape how healthcare exists. How is IBM Watson redefining Healthcare? As more and more information is fed into IBM Watson, doctors will get highly customised recommendations to treat their patients. The impact on patient care and hospital cost can be tremendous. For Healthcare professionals, Watson can Reduce/Eliminate time spent by healthcare professionals on insight mining from an ever-growing body of research Provide a list of recommended options for treatment with a score of confidence attached Design treatment plans based on option chosen In short, it can act as a highly effective personal assistant to this group. This means these professionals are more competent, more successful and have the time and energy to make deep personal connections with their patients thereby elevating patient care to a whole different level. For patients, Watson can Act an interactive interface answering their queries and connecting them with their healthcare professionals Provide at home diagnostics and healthcare advice Keep their patient records updated and synced up with their hospitals Thus, Watson can help patients make informed medical choices, take better care of themselves and alleviate the stress and anxiety induced by a chaotic and opaque hospital environment. For the healthcare industry, it means a reduction in overall cost to hospitals, reduced investment in post-treatment patient care, higher rates of success, reduction in errors due to oversight, misdiagnosis, and other human errors. This can indirectly improve key administrative metrics, lower employee burnout/churn rate, improve morale and result in other intangible benefits more.   The implications of such a transformation are not limited to health care alone. What about Insurance, Watson? IBM Watson can have a substantial impact on insurance companies too. Insurance, a fiercely debated topic, is a major cost for healthcare. Increasing revenue potential, better customer relationship and reducing cost are some areas where Watson will start disrupting medical insurance. But that’s just the beginning. Tighter integration with hospitals, more data on patient care, and more information on newer remedies will provide ground-breaking insights to insurance companies. These insights will help them figure out the right premiums and the underwriting frameworks. Moreover, the above is not a scene set in some distant future. In Japan, insurance company Fukoku Mutual Life Insurance replaced 34 employees and deployed IBM Watson. Customers of Fukoku can now directly discuss with an AI robot instead of a human being to settle payments. Fukoku made a one-time fee of $1,70,000 along with yearly maintenance of $1,28,000 to IBM for its Watson’s services. They plan to recover this cost by replacing their team of sales personnel, insurance agents, and customer care personnel - potentially saving nearly a million dollars in annual savings. These are interesting times and some may even call it anxiety-inducing. [box type="shadow" align="" class="" width=""]Shariff: No, no, no, Houston, don't be anxious. Anxiety is bad for the heart.[/box]

With the advent of automation, artificial intelligence(AI), and machine learning, we hear about their applications regularly in news across industries. This has been especially true for healthcare where various hospitals, health insurance companies, healthcare units, etc. have been impacted in more substantial and concrete ways by AI when compared to other industries. In the recent years, healthcare startups and life science organizations have ventured into Artificial Intelligence technology and are one of the most heavily invested areas by VCs. Various organizations with ties to healthcare are leveraging the advances in artificial intelligence algorithms for remote patient monitoring, medical imaging and diagnostics, and implementing newly developed sophisticated methods, and applications into the system. Let’s explore some of the most popular AI applications which have revamped the healthcare industry. Proper maintenance and management of medical records Assembling, analyzing, and maintaining medical information and records is one of the most commonly used applications of AI. With the coming of digital automation, robots are being used for collecting and tracing data for proper data management and analysis. This has brought down manual labor to a considerable extent. Computerized medical consultation and treatment path The existence of medical consultation apps like DocsApp allows a user to talk to experienced and specialist doctors on chat or call directly from their phone in a private and secure manner. Users can report their symptoms into the app and this ensures the users are connected to the right specialist physicians as per the user’s medical history. This has been made possible due to the existence of AI systems. AI also aids in treatment design like analyzing data, making notes and reports from a patient’s file, thereby helping in choosing the right customized treatment as per the patient’s medical history. Eliminates monotonous manual labor Various medical tasks like analyzing X-Ray reports, test reports, CT scans and other common tasks can be executed by robots and other mechanical devices more accurately. Radiology is one such discipline wherein human supervision and control have dropped to a substantial level due to the extensive use of AI. Aids in drug manufacture and creation Generally, billions of dollars are spent on developing pharmaceuticals through clinical trials and they take almost a decade or two to manufacture a life-saving drug. But now, with the arrival of AI, the entire drug creation procedure has been simplified and has become pretty reasonable as well. Even in the recent outbreak of the Ebola virus, AI was used for drug discovery, to redesign solutions and to scan the current existing medicines to eradicate the plague. Regular health monitoring In the current era of digitization, there are certain wearable health trackers – like Garmin, Fitbit, etc. which can monitor your heart rate and activity levels. These devices help the user to keep a close check on their health by setting up their exercise plan, or reminding them to stay hydrated. All this information can also be shared with your physician to track your current health status through AI systems. Helps in the early and accurate detection of medical disorders AI helps in spotting carcinogenic and cardiovascular disorders at an early stage and also aids in predicting health issues that people are likely to contract due to hereditary or genetic reasons. Enhances medical diagnosis and medication management Medical diagnosis and medication management are the ultimate data-based problems in the healthcare industry. IBM’s Watson, a deep learning system has simplified medical investigation and is being applied to oncology, specifically for cancer diagnosis. Previously, human doctors used to collect patient data, research on it and conduct clinical trials. But with AI, the manual efforts have reduced considerably. For medication management, certain apps have been developed to monitor the medicines taken by a patient. The cellphone camera in conjunction with AI technology to check whether the patients are taking the medication as prescribed. Further, this also helps in detecting serious medical problems and tracking patients medicine adaptability and participants behavior in certain scientific trials. To conclude, we can connote that we are gradually embarking on the new era of cognitive technology with the power of AI-based systems. In the coming years, we can expect AI to transform every area of the healthcare industry that it brushes up with. Experts are constantly looking for ways and means to organize the existing structure and power up healthcare on the basis of new AI technology. The ultimate goals being to improve patient experience, build a better public health management and reduce costs by automating manual labor. Author Bio Maria Thomas is the Content Marketing Manager and Product Specialist at GreyCampus with eight years rich experience on professional certification courses like PMI- Project Management Professional, PMI-ACP, Prince2, ITIL (Information Technology Infrastructure Library), Big Data, Cloud, Digital Marketing and Six Sigma. Healthcare Analytics: Logistic Regression to Reduce Patient Readmissions How IBM Watson is paving the road for Healthcare 3.0

Social media companies are facing a lot of heat presently because of their privacy issues. One of them is Facebook. The Cambridge analytica scandal had even prompted a senate hearing for Mark Zuckerberg. On the other end of this spectrum, there is another messaging app known as Telegram, registered in London, United Kingdom, founded by the Russian entrepreneur Pavel Durov. Telegram has been in the news for an absolutely opposite situation. It’s often touted as one of the most secure and secretive messaging apps. The end to end encryption ensures that security agencies across the world have a tough time getting access to any suspicious piece of information. For this reason Russia has banned the use of Telegram app on April 2018. Telegram updated their privacy policies on . These updates have further ensured that Telegram will retain the title of the most secure messaging application in the planet. It’s imperative for any messaging app to get access to our data. But how they choose to use it makes you either vulnerable or secure. Telegram in their latest update have stated that they process personal data on the grounds that such processing caters to the following two goals: Providing effective and innovative Services to our users To detect, prevent or otherwise address fraud or security issues in respect of their provision of Services. The caveat for the second point being the security interests shall not override the space of fundamental rights and freedoms that require protection of personal data. This clause is an excellent example on how applications can prove to be a torchbearer for human rights and basic human privacy amidst glaring loopholes. Telegram have listed the the kind of user data accessed by the app. They are as follows: Basic Account Data Telegram stores basic account user data that includes mobile number, profile name, profile picture and about information, which are needed  to create a Telegram account. The most interesting part of this is Telegram allows you to only keep your username (if you choose to) public. The people who have you in their contact list will see you as you want them to - for example you might be a John Doe in public, but your mom will still see you as ‘Dear Son’ in their contacts. Telegram doesn’t require your real name, gender, age or even your screen name to be your real name. E-mail Address When you enable 2-step-verification for your account or store documents using the Telegram Passport feature, you can opt to set up a password recovery email. This address will only be used to send you a password recovery code if you forget it. They are particular about not sending any unsolicited marketing emails to you. Personal Messages Cloud Chats Telegram stores messages, photos, videos and documents from your cloud chats on their servers so that you can access your data from any of your devices anytime without having to rely on third-party backups. All data is stored heavily encrypted and the encryption keys in each case are stored in several other data centers in different jurisdictions. This way local engineers or physical intruders cannot get access to user data. Secret Chats Telegram has a feature called Secret chats that uses end-to-end encryption. This means that all data is encrypted with a key that only the sender and the recipients know. There is no way for us or anybody else without direct access to your device to learn what content is being sent in those messages. Telegram does not store ‘secret chats’ on their servers. They also do not keep any logs for messages in secret chats, so after a short period of time there is no way of determining who or when you messaged via secret chats. Secret chats are not available in the cloud — you can only access those messages from the device they were sent to or from. Media in Secret Chats When you send photos, videos or files via secret chats, before being uploaded, each item is encrypted with a separate key, not known to the server. This key and the file’s location are then encrypted again, this time with the secret chat’s key — and sent to your recipient. They can then download and decipher the file. This means that the file is technically on one of Telegram’s servers, but it looks like a piece of random indecipherable garbage to everyone except for you and the recipient. This complete process is random and there random data packets are periodically purged from the storage disks too. Public Chats In addition to private messages, Telegram also supports public channels and public groups. All public chats are cloud chats. Like everything else on Telegram, the data you post in public communities is encrypted, both in storage and in transit — but everything you post in public will be accessible to everyone. Phone Number and Contacts Telegram uses phone numbers as unique identifiers so that it is easy for you to switch from SMS and other messaging apps and retain your social graph. But the most important thing is that permissions from the users are a must before the cookies are allowed into your browser. Cookies Telegram promises that the only cookies they use are those to operate and provide their Services on the web. They clearly state that they don’t use cookies for profiling or advertising. Their cookies are small text files that allow them to provide and customize their Services, and provide an enhanced user experience. Also, whether or not to use these cookies is a choice made by the users. So, how does Telegram remain in business? The Telegram business model doesn’t match that of a revenue generating service. The founder Pavel Durov is also the founder of the popular Russian social networking site VK. Telegram doesn’t charge for any messaging services, it doesn’t show ads yet. Some new in app purchase features might be included in the new version. As of now, the main source of revenue for Telegram are donations and mainly the earnings of Pavel Durov himself (from the social networking site VK). What can social networks learn from Telegram? Telegram’s policies elevate privacy standards that many are asking from other social messaging apps. The clamour for stopping the exploitation of user data, using their location details for targeted marketing and advertising campaigns is increasing now. Telegram shows that privacy can be achieved, if intended, in today’s overexposed social media world. But there is are also costs to this level of user privacy and secrecy, that are sometimes not discussed enough. The ISIS members behind the 2015 Paris attacks used Telegram to spread propaganda. ISIS also used the app to recruit the perpetrators of the Christmas market attack in Berlin last year and claimed credit for the massacre. More recently, a Turkish prosecutor found that the shooter behind the New Year’s Eve attack at the Reina nightclub in Istanbul used Telegram to receive directions for it from an ISIS leader in Raqqa. While these incidents can never negate the need for a secure and less intrusive social media platform like Telegram, there should be workarounds and escape routes designed for stopping extremists and terrorist activities. Telegram have assured that all ISIS messaging channels are deleted from their network which is a great way to start. Content moderation, proactive sentiment and pattern recognition and content/account isolation are the next challenges for Telegram. One thing is for sure, Telegram’s continual pursuance of user secrey and user data privacy is throwing an open challenge to others to follow suite. Whether others will oblige or not, only time will tell. To read about Telegram’s updated privacy policies in detail, you can check out the official Telegram Privacy Settings. How to stay safe while using Social Media Time for Facebook, Twitter and other social media to take responsibility or face regulation What RESTful APIs can do for Cloud, IoT, social media and other emerging technologies

Unless you have been living under a rock, you have most likely heard about Bitcoin, the world's most popular cryptocurrency that is growing by leaps and bounds. In fact, recently, Bitcoin broke the threshold of $6000 and is now priced at an all-time high. Bitcoin is not alone in this race as another cryptocurrency named Ethereum is hot on its heels. Despite being only three years old, Ethereum is quickly emerging as a popular choice especially among enterprise users. Ethereum’s YTD price growth has been more than a whopping 3000%. In terms of market cap as well Ethereum has shown a significant increase. Its overall share of the 'total cryptocurrency market' rose from 5% at the beginning of the year to 30% YTD.  In absolute terms, today it stands at around $28 Billion.  On the other hand, Bitcoin’s market cap as a percentage of the market has shrunk from 85% at the start of the year to 55%  and is valued at around $90 Billion. Bitcoin played a huge role in bringing Ethereum into existence. The co-creator and inventor of Ethereum, Vitalik Buterin, was only 19 when his father introduced him to bitcoin and by extension, to the fascinating world of cryptocurrency. In a span of 3 years, Vitalik had written several blogs on the topic and also co-founded the Bitcoin Magazine in 2011. Though Bitcoin served as an excellent tool for money transaction eliminating the need for banks, fees, or third party, its scripting language had limitations. This led to Vitalik, along with other developers, to found Ethereum - A platform that aimed to extend beyond Bitcoin’s scope and make internet decentralized. How Ethereum differs from the reigning cryptocurrency - Bitcoin Both Bitcoin and Ethereum are built on top of Blockchain technology allowing them to build a decentralized public network. However, Ethereum’s capability extends beyond being a cryptocurrency and differs from Bitcoin substantially in terms of scope and potential. Exploiting the full spectrum blockchain platform Bitcoin leverages Blockchain's distributed ledger technology to perform secured peer-to-peer cash transactions. It thus disrupted traditional financial transaction instruments such as PayPal. Meanwhile, Ethereum aims to offer much more than digital currency by helping developers build and deploy any kind of decentralized applications on top of Blockchain. The following are some Ethereum based features and applications that make it superior to bitcoin. DApps A decentralized app or DApp refers to a program running on the internet through a network but is not under the control of any single entity. A white paper on DApp highlights the four conditions that need to be satisfied to call an application a DApp: It must be completely open-source Data and records of operation must be cryptographically stored It should utilize a cryptographic token It must generate tokens The whitepaper also goes on to suggest that DApps are the future: “decentralized applications will someday surpass the world’s largest software corporations in utility, user-base, and network valuation due to their superior incentivization structure, flexibility, transparency, resiliency, and distributed nature.” Smart Contracts and EVM Another feature that Ethereum boasts over Bitcoin is a smart contract. A smart contract works like a traditional contract. You can use it to perform a task or transfer money in return for any asset or task in an efficient manner without needing interference from a middleman. Though Bitcoin is fast, secure, and saves cost it has limitations in terms of the ability to run operations. Ethereum solves this problem by allowing operations to work as a contract by converting them to pieces of code and have them supervised by a network of computers. A tool that helps Ethereum developers build and experiment with different contracts is Ethereum Virtual Machine. It acts as a testing environment to build blockchain operations and is isolated from the main network. Thus, it gives developers a perfect platform to build and test smart as well as robust contracts across different industries. DAOs One can also create Decentralized Autonomous Organizations (DAO) using Ethereum. DAO eliminates the need for human managerial involvement. The organization runs through smart contracts that convert rules, core tasks and structure of the organization to codes monitored by a fault-tolerant network. An example of DAO is Slock.it, a DAO version of Airbnb. Performance An important factor for cryptocurrency transaction is the amount of time it takes to finalize the transaction. This is called as Block Time. In terms of performance, the Bitcoin network takes 10 minutes to make a transaction whereas Ethereum is much more efficient and boasts a block time of just 14-15 seconds. Development Ethereum’s programming language Solidity is based on JavaScript. This is great for web developers who want to use their knowledge of JavaScript to build cool DApps and extend the Ethereum platform. Moreover, Ethereum is Turing complete, meaning it can compute anything that is computable provided enough resources are available. Bitcoin, on the other hand, is based on C++ which comparatively is not a popular choice among the new generation of app developers. Community and Vision One can say Bitcoin works as a DAO with no involvement of individuals in managing the cryptocurrency and is completely decentralized and owned by the community. Satoshi Nakamoto, who prefers to stay behind the curtains, is the only name that one comes across when it comes to relating an individual with Bitcoin. The community, therefore, lacks a figurehead when it comes to seeking future directions. Meanwhile, Vitalik Buterin is hugely popular amongst Ethereum enthusiasts and is very much involved in designing the future roadmap with other co-founders. Cryptocurrency Supply Similar to Bitcoin, Ethereum has Ether which works as a digital asset that fuels the network and transactions performed on the platform. Bitcoin has a fixed supply cap of around 21 million coins. It’s going to take more than 100 years to mine the last Bitcoin after which Bitcoin would behave as a deflationary cryptocurrency. Ethereum, on the other hand, has no fixed supply cap but has restricted its annual supply to 18 million Ethers. With no upper cap on the number of Ether that can be mined, Ethereum behaves as an inflationary currency and may lose value with time. However, the Ethereum community is now planning to move from proof-of-work to proof-of-stake model which should limit the number of ethers being mined and also offer benefits such as energy efficiency and security. Some real-world applications using Ethereum The Decentralized applications’ growth has been on the rise with people starting to recognize the value offered by Blockchain and decentralization such as security, immutability, tamper-proofing, and much more. While Bitcoin uses blockchain purely as a list of transactions, Ethereum manages to transfer value and information through its platform. Thus, it allows for immense possibilities when it comes to building different DApps across a wide range of industries. The financial domain is obviously where Ethereum is finding a lot of traction. Projects such as Branche - a Decentralized Consumer Micro­credit and Financial Services and Augur, a decentralized prediction market that has raised more than $ 5 million are some prominent examples. But financial applications are only the tip of the iceberg when it comes to possibilities that Ethereum offers and potential it holds when it comes disrupting industries across various sectors. Some other sectors where Ethereum is making its presence felt are: Firstblood is a decentralized eSports platform which has raised more than $5.5 million. It allows players to test their skills and bet using Ethereum while the tournaments are tracked on smart contracts and blockchain. Alice.Si a charitable trust that lets donors invest in noble causes knowing the fact that they only pay for causes where the charity makes an impact. Chainy is an Ethereum-based authentication and verification system that permanently stores records on blockchain using timestamping. Flippening is happening! If you haven’t heard of Flippening, it’s a term coined by cryptocurrency enthusiasts on Ethereum chances of beating Bitcoin to claim the number one spot to become the largest capitalized blockchain. Comparing Ethereum to Bitcoin may not be right as both serve different purposes. Bitcoin will continue to dominate cryptocurrency but as more industries adopt Ethereum to build Smart Contracts, DApps, or DAOs of their choice, its popularity is only going to grow, subsequently making Ether more valuable. Thus, the possibility of Ether displacing Bitcoin is strong. With the pace at which Ethereum is growing and the potential it holds in terms of unleashing Blockchain’s power to transform industries, it is definitely a question of when rather than if Flippening would happen!

To stay competitive in today’s economic environment, organizations can no longer be reliant on just their IT team for all their data consumption needs. At the same time, the need to get quick insights to make smarter and more accurate business decisions is now stronger than ever. As a result, there has been a sharp rise in a new kind of analytics where the information seekers can themselves create and access a specific set of reports and dashboards - without IT intervention. This is popularly termed as Self-service Analytics. Per Gartner, Self-service analytics is defined as: “A  form of business intelligence (BI) in which line-of-business professionals are enabled and encouraged to perform queries and generate reports on their own, with nominal IT support.” Expected to become a $10 billion market by 2022, self-service analytics is characterized by simple, intuitive and interactive BI tools that have basic analytic and reporting capabilities with a focus on easy data access. It empowers business users to access relevant data and extract insights from it without needing to be an expert in statistical analysis or data mining. Today, many tools and platforms for self-service analytics are already on the market - Tableau, Microsoft Power BI, IBM Watson, Qlikview and Qlik Sense being some of the major ones. Not only have these empowered users to perform all kinds of analytics with accuracy, but their reasonable pricing, in-tool guidance and the sheer ease of use have also made them very popular among business users. Rise of the Citizen Data Scientist The rise in popularity of self-service analytics has led to the coining of a media-favored term - ‘Citizen Data Scientist’. But what does the term mean? Citizen data scientists are business users and other professionals who can perform less intensive data-related tasks such as data exploration, visualization and reporting on their own using just the self-service BI tools. If Gartner’s predictions are to be believed, there will be more citizen data scientists in 2019 than the traditional data scientists who will be performing a variety of analytics-related tasks. How Self-service Analytics benefits businesses Allowing the end-users within a business to perform their own analysis has some important advantages as compared to using the traditional BI platforms: The time taken to arrive at crucial business insights is drastically reduced. This is because teams don’t have to rely on the IT team to deliver specific reports and dashboards based on the organizational data. Quicker insights from self-service BI tools mean businesses can take decisions faster with higher confidence and deploy appropriate strategies to maximize business goals. Because of the relative ease of use, business users can get up to speed with the self-service BI tools/platform in no time and with very little training as compared to being trained on complex BI solutions. This means relatively lower training costs and democratization of BI analytics which in turn reduces the workload on the IT team and allows them to focus on their own core tasks. Self-service analytics helps the users to manage the data from disparate sources more efficiently, thus allowing organizations to be agiler in terms of handling new business requirements. Challenges in Self-service analytics While the self-service analytics platforms offer many benefits, they come with their own set of challenges too.  Let’s see some of them: Defining a clear role for the IT team within the business by addressing concerns such as: Identifying the right BI tool for the business - Among the many tools out there, identifying which tool is the best fit is very important. Identifying which processes and business groups can make the best use of self-service BI and who may require assistance from IT Setting up the right infrastructure and support system for data analysis and reporting Answering questions like - who will design complex models and perform high-level data analysis Thus, rather than becoming secondary to the business, the role of the IT team becomes even more important when adopting a self-service business intelligence solution. Defining a strict data governance policy - This is a critical task as unauthorized access to organizational data can be detrimental to the business. Identifying the right ‘power users’, i.e., the users who need access to the data and the tools, the level of access that needs to be given to them, and ensuring the integrity and security of the data are some of the key factors that need to be kept in mind. The IT team plays a major role in establishing strict data governance policies and ensuring the data is safe, secure and shared across the right users for self-service analytics. Asking the right kind of questions on the data - When users who aren’t analysts get access to data and the self-service tools, asking the right questions of the data in order to get useful, actionable insights from it becomes highly important. Failure to perform correct analysis can result in incorrect or insufficient findings, which might lead to wrong decision-making. Regular training sessions and support systems in place can help a business overcome this challenge. To read more about the limitations of self-service BI, check out this interesting article. In Conclusion IDC has predicted that spending on self-service BI tools will grow 2.5 times than spending on traditional IT-controlled BI tools by 2020. This is an indicator that many organizations worldwide and of all sizes will increasingly believe that self-service analytics is a feasible and profitable way to go forward. Today mainstream adoption of self-service analytics still appears to be in the early stages due to a general lack of awareness among businesses. Many organizations still depend on the IT team or an internal analytics team for all their data-driven decision-making tasks. As we have already seen, this comes with a lot of limitations - limitations that can easily be overcome by the adoption of a self-service culture in analytics, and thus boost the speed, ease of use and quality of the analytics. By shifting most of the reporting work to the power users,  and by establishing the right data governance policies, businesses with a self-service BI strategy can grow a culture that fuels agile thinking, innovation and thus is ready for success in the marketplace. If you’re interested in learning more about popular self-service BI tools, these are some of our premium products to help you get started:   Learning Tableau 10 Tableau 10 Business Intelligence Cookbook Learning IBM Watson Analytics QlikView 11 for Developers Microsoft Power BI Cookbook    

With the rise in popularity of deep learning as a concept and a paradigm, neural networks are captivating the interest of machine learning enthusiasts and developers alike, by being able to replicate the human brain for efficient predictions, image recognition, text recognition, and much more. However, can these neural networks do something more, or are they just limited to predictions? Can they self-generate new data by learning from a training dataset? Generative Adversarial networks (GANs) are here, to answer all these questions. So, what are GANs all about? Generative Adversarial Networks follow unsupervised machine learning, unlike traditional neural networks. When a neural network is taught to identify a bird, it is fed with a huge number of images including birds, as training data. Each picture is labeled before it is put to use in training the models. This labeling of data is both costly and time-consuming. So, how can you train your neural networks by giving it less data to train on? GANs are of a great help here. They cast out an easy way to train the DL algorithms by slashing out the amount of data required to train the neural network models, that too, with no labeling of data required. The architecture of a GAN includes a generative network model(G), which produces fake images or texts, and an adversarial network model--also known as the discriminator model (D)--that distinguishes between the real and the fake productions by comparing the content sent by the generator with the training data it has. Both of these are trained separately by feeding each of them with training data and a competitive goal. Source: Learning Generative Adversarial Networks GANs in action GANs were introduced by Ian Goodfellow, an AI researcher at Google Brain. He compares the generator and the discriminator models with a counterfeiter and a police officer. “You can think of this being like a competition between counterfeiters and the police,” Goodfellow said. “Counterfeiters want to make fake money and have it look real, and the police want to look at any particular bill and determine if it’s fake.” Both the discriminator and the generator are trained simultaneously to create a powerful GAN architecture. Let’s peek into how a GAN model is trained- Specify the problem statement and state the type of manipulation that the GAN model is expected to carry out. Collect data based on the problem statement. For instance, for image manipulation, a lot of images are required to be collected to feed in. The discriminator is fed with an image; one from the training set and one produced by the generator The discriminator can be termed as ‘successfully trained’ if it returns 1 for the real image and 0 for the fake image. The goal of the generator is to successfully fool the discriminator and getting the output as 1 for each of its generated image. In the beginning of the training, the discriminator loss--the ability to differentiate real and fake image or data--is minimal. As the training advances, the generator loss decreases and the discriminator loss increases, This means, the generator is now able to generate real images. Real world applications of GANs The basic application of GANs can be seen in generating photo-realistic images. But there is more to what GANs can do. Some of the instances where GANs are majorly put to use include: Image Synthesis Image Synthesis is one of the primary use cases of GANs. Here, multilayer perceptron models are used in both the generator and the discriminator to generate photo-realistic images based on the training dataset of the images. Text-to-image synthesis 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. Both Generator and Discriminator are then trained based on this encoded data. Image Inpainting Images that have missing parts or have too much of noise are given as an input to the generator which produces a near to real image. For instance, using TensorFlow framework, DCGANs (Deep Convolutional GANs), can generate a complete image from a broken image. DCGANs are a class of CNNs that stabilizes GANs for efficient usage. Video generation Static images can be transformed into short scenes with plausible motions using GANs. These GANs use scene dynamics in order to add motion to static images. The videos generated by these models are not real but illusions. Drug discovery Unlike text and image manipulation, Insilico medicine uses GANs to generate an artificially intelligent drug discovery mechanism. To do this, the generator is trained to predict a drug for a disease which was previously incurable.The task of the discriminator is to determine whether the drug actually cures the disease. Challenges in training a GAN Whenever a competition is laid out, there has to be a distinct winner. In GANs, there are two models competing against each other. Hence, there can be difficulties in training them. Here are some challenges faced while training GANs: Fair training: While training both the models, precaution has to be taken that the discriminator does not overpower the generator. If it does, the generator would fail to train effectively. On the other hand, if the discriminator is lenient, it would allow any illegitimate content to be generated. Failure to understand the number of objects and the dimensions of objects, present in a particular image. This usually occurs during the initial learning phase. For instance, GANs, at times output an image which ends up having more than two eyes, which is not normal in the real world. Sometimes, it may present a 3D image like a 2D one. This is because they cannot differentiate between the two. Failure to understand the holistic structure: GANs lack in identifying universally correct images. It may generate an image which can be totally opposed to how they look in real. For instance, a cat having an elongated body shape, or a cow standing on its hind legs, etc. Mode collapse is another challenge, which occurs when a low variation dataset is processed by a GANs. Real world includes complex and multimodal distributions, where data may have different concentrated sub-groups. The problem here is, the generator would be able to yield images based on anyone sub-group resulting in an inaccurate output. Thus, causing a mode collapse. To tackle these and other challenges that arise while training GANs, researchers have come up with DCGANs (Deep Convolutional GANs), WassersteinGANs, CycleGANs to ensure fair training, enhance accuracy, and reduce the training time. AdaGANs are implemented to eliminate mode collapse problem. Conclusion Although the adoption of GANs is not as widespread as one might imagine, there’s no doubt that they could change the way unsupervised machine learning is used today. It is not too far-fetched to think that their implementation in the future could find practical applications in not just image or text processing, but also in domains such as cryptography and cybersecurity. Innovations in developing newer GAN models with improved accuracy and lesser training time is the key here - but it is something surely worth keeping an eye on.

When people talk about tech giants such as Amazon, Facebook, and Google, they usually talk about the great and powerful innovations that they’ve brought to the table, that have perpetually transformed the contemporary world. Of late, criticism of these same tech titans holding back the power of innovation from other smaller companies as they have become, what you may call, a tech monopoly has been gain traction. In a podcast episode of Innovation For All, Sheana Ahlqvist talked to Sally Hubbard, an antitrust expert, and investigative journalist at The Capitol Forum, regarding tech giants building monopolies. Here are some key highlights from the podcast.   Let’s recall the definition of monopoly. “A market structure characterized by a single seller, selling a unique product in the market. In a monopoly market, the seller faces no competition, as he is the sole seller of goods with no close substitute. Monopoly market makes the single seller the market controller as well as the price maker. He enjoys the power of setting the price for his goods”. In a nutshell, decrease the prices of your service and drive everyone else out of the business. A popular example is John D Rockefeller, Standard Oil’s chief executive, who ruined other competitors by cutting the prices of the oil until they went bankrupt, immediately after which the higher prices returned. Now although there is no price-fixing in the case of Google or Facebook since they offer completely free services, they’re still a monopoly. Let’s have a look. How are Amazon, Google, and Facebook tech monopolies? If you look at each one of these organizations - Amazon, Facebook, and Google have carved out their own markets, with gargantuan and durable market power vested in the hands of each one of them. According to the US Department of Justice, a market share of greater than 50% has been necessary for courts to find the existence of monopolistic power. A dominant market share is a useful starting point in determining monopoly power. Going by this rule, Google has dominated the search engine market, maintaining an 86.02 % market share as of July 2018, as per Statista. This is way over 50%, making Google a monopoly. The majority of Google revenues are generated through advertising. Similarly, Facebook dominates the social media market, with its worldwide market share of 66.67%, making it a monopoly too. Amazon, on the other hand, has 41% market share in the e-commerce retail market which is expected to increase significantly to 50% of the entire e-commerce retail market’s GMV, by 2021. This brings it pretty close to being a monopoly soon in the e-commerce market soon. Another factor that is considered under the Sherman Act, a part of the antitrust law, when identifying a firm that possesses monopoly power, is the existence of anti-competitive effect i.e. companies trying to maintain or acquire a dominant position by excluding competitors or preventing new entry. One recent example that comes to mind is when Google was fined with $5 billion in July this year for breaching EU’s antitrust laws. Google was fined for 3 types of illegal restrictions on the use of Android, cementing the dominance of its search engine. As per EU, Google denied its rivals a chance to innovate and compete on merits, which is illegal under EU’s antitrust laws. Also Read: A quick look at E.U.’s antitrust case against Google’s Android Monopolies and Social Injustice Hubbard points out how these tech titans don’t have any major competitors or substitutes, and even if you don’t pay most of these tech giants with money, you pay them with your data. This is more than enough for world domination, which is always an underlying aspiration for tech companies as they strive to be “the one” in the eyes of their customers, by carefully leveraging their data. This data also put these companies at an unfair advantage over other smaller and newer businesses. As Clive Humby, a British mathematician rightly said, “data is the new oil” in the digital economy. Hubbard explains how the downsides of this monopoly might not be visible to the consumer but affects entrepreneurs and small businesses who are greatly harmed by the practices of these companies. Taking Amazon, for instance, no one wishes to be dependent on their competitor, however, since Amazon has integrated the service of selling products on its platform, not only is everyone competing against Amazon but are also dependent on Amazon, as it is Amazon who decides the rules for the sellers. Add to this the fact that Amazon comprises a ginormous amount of consumer data in hand, putting it in an unfair advantage over others as it can dominate its products over others. There is currently an ongoing EU investigation into Amazon’s use of consumer and seller data collected on its platform to better its own products sold on its platform. Similarly, Google’s monopoly is evident in the fact that it gets to decide the losers and winners of the internet on its Google search, prioritizing its products over the others. An example of this is Google getting fined with 2.7 billion dollars by EU, last year after it ruled the company had abused its power by promoting its own shopping comparison service at the top of search results. Facebook, on the other hand, doesn’t have a direct competition, leaving users with less choice in terms of Social network sites, making it a monopoly. Add to that the fact that other major social media platforms such as Instagram and Whatsapp are also owned by Facebook. Hubbard explains how Facebook doesn't have competition, so it can prioritize its profits over the other factors such as user data as it's really not concerned about user loss. This is evident in the number of scandals that Facebook has gotten itself into regarding user data.  Facebook is facing a whole lot of data and privacy-related controversies, Cambridge Analytica scandal being the most popular one. Facebook suffered the largest security breach in its history that left 50M user accounts compromised, last month. Department of Housing and Urban Development UD) filed a complaint against Facebook in August, alleging the platform is selling ads that discriminate against users based on race, religion, and sexuality. ACLU also sued Facebook in September for enabling sex and age discrimination through targeted ads. Last week, the New York Times published a bombshell report on how Facebook has been following the strategy of ‘delaying, denying and deflecting’ the blame under the leadership of Sheryl Sandberg for all the controversies surrounding it. Scandals aside, even if a user finds the content hosted by Facebook displeasing, they don’t really have a choice to “stop using Facebook” as their friends and family continue to use the platform to stay in touch. Also, Facebook charges advertisers depending on how many people see a message instead of being based on ad clicks. This is why Facebook’s algorithm is programmed in a way that it prioritizes more engaging branded content and ads over the others. Monopoly and Gender Inequality As the market power of these tech giants increases, so does their wealth. Hubbard points out that the wealth from the many among the working and middle classes get transferred to the few belonging to the 1% and 0.1% at the top of the income and wealth distribution. The concentration of market power hurts workers and results in depresses wages, affecting women and other minority workers the most. “When general wages go down or stagnate, female workers are even worse off. Women make 78 cents to a man’s dollar, with black women making 64 cents and Latina women making 54 cents for every dollar a white man makes. As wages by the bottom 99% of earners continue to shrink, women get paid a mere percentage of fewer dollars. And the top 1% of earners are predominantly men”, mentions Sally Hubbard. There have also been declines in employee mobility as there are lesser firms competing due to giant firms acquiring smaller firms. This leads to reduced bargaining power in the hands of an employee. Moreover, these firms also t impose non-compete clauses and no-poach agreements putting a damper on workers’ ability to switch jobs. As eloquently put by Hubbard, “these tech platforms are the ones controlling the rules of the arena in which the game is played and are also the ones playing the game”. Taking into consideration this analogy, it’s anyone’s guess who’ll win the game. OK Google, why are you ok with mut(at)ing your ethos for Project DragonFly? How far will Facebook go to fix what it broke: Democracy, Trust, Reality Amazon splits HQ2 between New York and Washington, D.C. after a making 200+ states compete over a year; public sentiments largely negative

Learning something by rote i.e., repeating it many times, perfecting a skill by practising it over and over again or building something by making minor adjustments progressively to a prototype are things that comes to us naturally as human beings. Machines can also learn this way and this is called ‘Iterative machine learning’. In most cases, iteration is an efficient learning approach that helps reach the desired end results faster and accurately without becoming a resource crunch nightmare. Now, you might wonder, isn’t iteration inherently part of any kind of machine learning? In other words, modern day machine learning techniques across the spectrum from basic regression analysis, decision trees, Bayesian networks, to advanced neural nets and deep learning algorithms have some inherent iterative component built into them. What is the need, then, for discussing iterative learning as a standalone topic? This is simply because introducing iteration externally to an algorithm can minimize the error margin and therefore help in accurate modelling.  How Iterative Learning works Let’s understand how iteration works by looking closely at what happens during a single iteration flow within a machine learning algorithm. A pre-processed training dataset is first introduced into the model. After processing and model building with the given data, the model is tested, and then the results are matched with the desired result/expected output. The feedback is then returned back to the system for the algorithm to further learn and fine tune its results. This clearly shows that two iteration processes take place here: Data Iteration - Inherent to the algorithm Model Training Iteration - Introduced externally Now, what if we did not feedback the results into the system i.e. did not allow the algorithm to learn iteratively but instead adopted a sequential approach? Would the algorithm work and would it provide the right results? Yes, the algorithm would definitely work. However, the quality of the results it produces is going to vary vastly based on a number of factors. The quality and quantity of the training dataset, the feature definition and extraction techniques employed, the robustness of the algorithm itself are among many other factors. Even if all of the above were done perfectly, there is still no guarantee that the results produced by a sequential approach will be highly accurate. In short, the results will neither be accurate nor reproducible. Iterative learning thus allows algorithms to improve model accuracy. Certain algorithms have iteration central to their design and can be scaled as per the data size. These algorithms are at the forefront of machine learning implementations because of their ability to perform faster and better. In the following sections we will discuss iteration in different sets of algorithms each from the three main machine learning approaches - supervised ML, unsupervised ML and reinforcement learning. The Boosting algorithms: Iteration in supervised ML The boosting algorithms, inherently iterative in nature, are a brilliant way to improve results by minimizing errors. They are primarily designed to reduce bias in results and transform a particular set of weak learning classifier algorithms to strong learners and to enable them to reduce errors. Some examples are: AdaBoost (Adaptive Boosting) Gradient Tree Boosting XGBoost How they work All boosting algorithms have a common classifiers which are iteratively modified to reach the desired result. Let’s take the example of finding cases of plagiarism in a certain article. The first classifier here would be to find a group of words that appear somewhere else or in another article which would result in a red flag. If we create 10 separate group of words and term them as classifiers 1 to 10, then our article will be checked on the basis of this classifier and any possible matches will be red flagged. But no red flags with these 10 classifiers would not mean a definite 100% original article. Thus, we would need to update the classifiers, create shorter groups perhaps based on the first pass and improve the accuracy with which the classifiers can find similarity with other articles. This iteration process in Boosting algorithms eventually leads us to a fairly high rate of accuracy. The reason being after each iteration, the classifiers are updated based on their performance. The ones which have close similarity with other content are updated and tweaked so that we can get a better match. This process of improving the algorithm inherently, is termed as boosting and is currently one of the most popular methods in Supervised Machine Learning. Strengths & weaknesses The obvious advantage of this approach is that it allows minimal errors in the final model as the iteration enables the model to correct itself every time there is an error. The downside is the higher processing time and the overall memory requirement for a large number of iterations. Another important aspect is that the error fed back to train the model is done externally, which means the supervisor has control over the model and how it modifies. This in turn has a downside that the model doesn’t learn to eliminate error on its own. Hence, the model is not reusable with another set of data. In other words, the model does not learn how to become error-free by itself and hence cannot be ported to another dataset as it would need to start the learning process from scratch. Artificial Neural Networks: Iteration in unsupervised ML Neural Networks have become the poster child for unsupervised machine learning because of their accuracy in predicting data models. Some well known neural networks are: Convolutional Neural Networks   Boltzmann Machines Recurrent Neural Networks Deep Neural Networks Memory Networks How they work Artificial neural networks are highly accurate in simulating data models mainly because of their iterative process of learning. But this process is different from the one we explored earlier for Boosting algorithms. Here the process is seamless and natural and in a way it paves the way for reinforcement learning in AI systems. Neural Networks consist of electronic networks simulating the way the human brain is works. Every network has an input and output node and in-between hidden layers that consist of algorithms. The input node is given the initial data set to perform a set of actions and each iteration creates a result that is output as a string of data. This output is then matched with the actual result dataset and the error is then fed back to the input node. This error then enables the algorithms to correct themselves and reach closer and closer to the actual dataset. This process is called training the Neural Networks and each iteration improve the accuracy. The key difference between the iteration performed here as compared to how it is performed by Boosting algorithms is that here we don’t have to update the classifiers manually, the algorithms change themselves based on the error feedback. Strengths & weaknesses The main advantage of this process is obviously the level of accuracy that it can achieve on its own. The model is also reusable because it learns the means to achieve accuracy and not just gives you a direct result. The flip side of this approach is that the models can go wrong heavily and deviate completely in a different direction. This is because the induced iteration takes its own course and doesn’t need human supervision. The facebook chat-bots deviating from their original goal and communicating within themselves in a language of their own is a case in point. But as is the saying, smart things come with their own baggage. It’s a risk we would have to be ready to tackle if we want to create more accurate models and smarter systems.    Reinforcement Learning Reinforcement learning is a interesting case of machine learning where the simple neural networks are connected and together they interact with the environment to learn from their mistakes and rewards. The iteration introduced here happens in a complex way. The iteration happens in the form of reward or punishment for arriving at the correct or wrong results respectively. After each interaction of this kind, the multilayered neural networks incorporate the feedback, and then recreate the models for better accuracy. The typical type of reward and punishment method somewhat puts it in a space where it is neither supervised nor unsupervised, but exhibits traits of both and also has the added advantage of producing more accurate results. The con here is that the models are complex by design. Multilayered neural networks are difficult to handle in case of multiple iterations because each layer might respond differently to a certain reward or punishment. As such it may create inner conflict that might lead to a stalled system - one that can’t decide which direction to move next. Some Practical Implementations of Iteration Many modern day machine learning platforms and frameworks have implemented the iteration process on their own to create better data models, Apache Spark and MapR are two such examples. The way the two implement iteration is technically different and they have their merits and limitations. Let’s look at MapReduce. It reads and writes data directly onto HDFS filesystem present on the disk. Note that for every iteration to be read and written from the disk needs significant time. This in a way creates a more robust and fault tolerant system but compromises on the speed. On the other hand, Apache Spark stores the data in memory (Resilient Distributed DataSet) i.e. in the RAM. As a result, each iteration takes much less time which enables Spark to perform lightning fast data processing. But the primary problem with the Spark way of doing iteration is that dynamic memory or RAM is much less reliable than disk memory to store iteration data and perform complex operations. Hence it’s much less fault tolerant that MapR.   Bringing it together To sum up the discussion, we can look at the process of iteration and its stages in implementing machine learning models roughly as follows: Parameter Iteration: This is the first and inherent stage of iteration for any algorithm. The parameters involved in a certain algorithm are run multiple times and the best fitting parameters for the model are finalized in this process. Data Iteration: Once the model parameters are finalized, the data is put into the system and the model is simulated. Multiple sets of data are put into the system to check the parameters’ effectiveness in bringing out the desired result. Hence, if data iteration stage suggests that some of the parameters are not well suited for the model, then they are taken back to the parameter iteration stage and parameters are added or modified. Model Iteration: After the initial parameters and data sets are finalized, the model testing/ training happens. The iteration in model testing phase is all about running the same model simulation multiple times with the same parameters and data set, and then checking the amount of error, if the error varies significantly in every iteration, then there is something wrong with either the data or the parameter or both. Iterations are done to data and parameters until the model achieves accuracy.   Human Iteration: This step involves the human induced iteration where different models are put together to create a fully functional smart system. Here, multiple levels of fitting and refitting happens to achieve a coherent overall goal such as creating a driverless car system or a fully functional AI. Iteration is pivotal to creating smarter AI systems in the near future. The enormous memory requirements for performing multiple iterations on complex data sets continue to pose major challenges. But with increasingly better AI chips, storage options and data transfer techniques, these challenges are getting easier to handle. We believe iterative machine learning techniques will continue to lead the transformation of the AI landscape in the near future.  

Deep Reinforcement Learning (Deep RL) is the new buzzword in the machine learning world. Deep RL is an approach which combines reinforcement learning and deep learning in order to achieve human-level performance. It brings together the self-learning approach to learn successful strategies that lead to the greatest long-term rewards and allows the agents to construct and learn their own knowledge directly from raw inputs. With the fusion of these two approaches, we saw the introduction of many algorithms, starting with DeepMind’s Deep Q Network (DQN). It is a deep variant of the Q-learning algorithm. This algorithm reached human-level performance in playing Atari games. Combining Q-learning with reasonably sized neural networks and some optimization tricks, you can achieve human or superhuman performance in several Atari games. Deep RL resulted in one of the notable advancements in the game of AlphaGo.The AI agent by DeepMind was able to beat the human world champions Lee Sedol (4-1) and Fan Hui (5-0). DeepMind then further released advanced versions of their Agent called AlphaGO Zero and AlphaZero. Many recent works from the researchers at UC Berkeley have shown how both reinforcement learning and deep reinforcement learning have enabled the control of complex robots, both for locomotion and navigation. Despite these successes, it is quite difficult to find cases where deep RL has added any practical real-world value. The current status is that it is still a research topic. One of its limitations is that it assumes the existence of a reward function, which is either given or is hand-tuned offline. To get the desired results, your reward function must capture exactly what you want. RL has an annoying tendency to overfit to your reward, resulting in things you haven’t expected. This is the reason why Atari is a benchmark, as it is not only easy to get a lot of samples, but the goal is fairly straightforward i.e to maximize score. With so many researchers working towards introducing improved Deep RL algorithms, it surely is a treat. AlphaZero: The genesis of machine intuition DeepMind open sources TRFL, a new library of reinforcement learning building blocks Understanding Deep Reinforcement Learning by understanding the Markov Decision Process [Tutorial]