Tech News - Data

Blizzard, an American video game development company is all set to demonstrate the progress made by Google’s DeepMind AI at StarCraft II, a real-time strategy video game, tomorrow. “The StarCraft games have emerged as a "grand challenge" for the AI community as they're the perfect environment for benchmarking progress against problems such as planning, dealing with uncertainty and spatial reasoning”, says the Blizzard team. Blizzard had partnered up with DeepMind during the 2016 BlizzCon, where they announced that they’re opening up the research platform for StarCraft II so that everyone in the StarCraft II community can contribute towards advancement in the AI research. Ever since then, much progress has made on the AI research front when it comes to StarCraft II. It was only two months back when, Oriol Vinyals, Research Scientist, Google DeepMind, shared the details of the progress that the AI had made in StarCraft II, states the Blizzard team. Vinyals stated how the AI, or agent, had learned to perform basic macro focused strategies along with defence moves against cheesy and aggressive tactics such as “cannon rushes”. Blizzard also posted an update during BlizzCon 2018, stating that DeepMind had been working really hard at training their AI (or agent) to better understand and learn StarCraft II. “Once it started to grasp the basic rules of the game, it started exhibiting amusing behaviour such as immediately worker rushing its opponent, which actually had a success rate of 50% against the 'Insane' difficulty standard StarCraft II AI”, mentioned the Blizzard team. It has almost become a trend for DeepMind to measure the capabilities of its advanced AI against human opponents in video games. For instance, it made headlines in 2016 when its AlphaGo AI program, managed to successfully defeat Lee Sedol, world champion, in a five-game match. AlphaGo had also previously defeated the professional Go player, Fan Hui in 2015 who was a three-time European champion of the game at the time. Also, recently in December 2018, DeepMind researchers published a full evaluation of its AlphaZero in the journal Science, confirming that it is capable of mastering Chess, Shogi, and Go from scratch. Other examples of AI making its way into advanced game learning includes OpenAI Five, a team of AI algorithms that beat a team of amateur human video game players in Dota 2 – the popular battle arena game, back in June 2018. Later in August, it managed to beat semi-professional players at the Dota 2 game. The demonstration for DeepMind AI in StarCraft II is all set for tomorrow at 10 AM Pacific Time. Check out StarCraft’s Twitch channel or DeepMind’s YouTube channel to learn about other recent developments that have been made. Deepmind’s AlphaFold is successful in predicting the 3D structure of a protein making major inroads of AI use in healthcare Graph Nets – DeepMind’s library for graph networks in Tensorflow and Sonnet DeepMind open sources TRFL, a new library of reinforcement learning building blocks

Yesterday at the Microsoft Ignite conference, Microsoft, Adobe, and SAP came together to announce the Open Data Initiative. This initiative aims to help companies to better govern their data and support privacy and security initiatives. What is the Open Data Initiative? Open Data Initiative is an open alliance that aims to eliminate silos and enables a seamless flow of customer data. For this initiative, the trios (Microsoft, Adobe, and SAP) are enhancing interoperability and data exchange between their applications and platforms through a single data model. These applications and platforms include Adobe Experience Cloud and Adobe Experience Platform, Microsoft Dynamics 365, SAP C/4HANA and S/4HANA. How this initiative will help companies? This initiative will help other companies in the following ways: Companies will be able to build and adopt intelligent applications that will understand data, relationships, and metadata spanning multiple services from Adobe, SAP, Microsoft and their partners It will help companies to use the information trapped in internal and external silos to extract more value from its own data in real time to better serve customers Based on their preference or needs, companies will be able to move transactional, operational, customer or IoT data to and from the common data lake Enable companies to create data-powered digital feedback loops for greater business impact Top retail companies are showing support and excitement for the Open Data Initiative. Barry Simpson, chief information officer at the Coca-Cola Company, said: “This initiative from Adobe, Microsoft and SAP is an important and strategic development for the Coca-Cola System. Our digital growth plans centered around our customers are fueled by these platforms and open standards. A more unified approach to the management and control of our data strengthens our ability to support our growth agenda and our ability to satisfy security, privacy and GDPR compliance requirements. The industry needs to follow these leaders.” To know more about the Open Data Initiative, check out the press release on Microsoft’s official website. Microsoft announces the first public preview of SQL Server 2019 at Ignite 2018 SAP creates AI ethics guidelines and forms an advisory panel Adobe set to acquire Marketo putting Adobe Experience Cloud at the heart of all marketing

After almost a year of Cloud ML Engine release, Google has finally announced the use of Cloud TPU for faster training and running of machine learning models on Cloud ML Engine. This beta release allow customers of Cloud ML and Google Cloud Platform to use the revolutionary TPUs and accelerate the TensorFlow based machine learning models. Key features of Cloud TPU: High-level Performance - Each Cloud TPU offers a potential of up to 180 teraflops of computing performance and 64 gigabytes of ultra-high bandwidth memory. Availability of Reference Models - Solve challenges faced in image classification and object detection applications on Cloud TPUs with access to models like RetinaNet and ResNet 50. Access to Custom Machine Types - Get an an advantage of balancing processor speeds, memory, storage resources by connecting to Cloud TPU from various custom Virtual Machine types. Key Benefits: Speed Up Machine Learning Workloads - The newly innovated Cloud TPUs are designed to help in accelerating machine learning workloads with TensorFlow. Each of the Cloud TPU are buckled up with 180 teraflops of computational power for the cutting-edge machine learning models. Such large amounts of processing speed can help you create the next research breakthrough across Machine Learning and AI technology. On-Demand Machine Learning Supercomputing - You can access to powerful and high-performance machine learning accelerators on demand with absolute zero capital investment. Easy Ramping on Google Cloud - Knowing that TensorFlow is open-source, you can simply push your machine learning workloads of TensorFlow on Cloud TPUs.You can use TensorFlow high-level APIs and move your machine learning models to CPUs, GPUs, and TPUs with few line of codes. The Cloud TPU also offers models and training environment which can easily suffice your image classification and machine translation needs. Read more about Cloud TPU features at the official  CLOUD TPU page. Tensor Processing Unit (TPU) 3.0: Google’s answer to cloud-ready Artificial Intelligence Nvidia Tesla V100 GPUs publicly available in beta on Google Compute Engine and Kubernetes Engine How to Build TensorFlow Models for Mobile and Embedded devices

Honda has come up with a program called Curious Minded Machine (CMM) to expand cognitive robotics research. This is a program to create artificial intelligence that enables ‘learning’ with a human-like sense of curiosity. What is the Curious Minded Machine program? The idea is to build a model based on how children ‘learn to learn’. By observing human interactions and how they perform tasks, CMM can learn better ways to achieve goals. This initiative explores Cooperative Intelligence (CI), AI embedded in a social context enabling people to confidence and trust with AI systems. Soshi Iba, a principal scientist at Honda Research Institute USA, Inc says: “Our ultimate goal is to create new types of machines that can acquire an interest in learning and knowledge, and the ability to interact with the world and others. We want to develop Curious Minded Machines that use curiosity to serve the common good by understanding people's needs, empowering human capability, and ultimately addressing complex societal issues.” Who is in the program by Honda? This three-year program will include efforts from the Computer Science & Artificial Intelligence Laboratory (CSAIL) at Massachusetts Institute of Technology (MIT), the School of Engineering and Applied Science at the University of Pennsylvania (Penn), and the Paul G. Allen School of Computer Science & Engineering at the University of Washington. The areas tackled by the research teams participating with Honda are: MIT CSAIL: They are addressing a key limitation in robotic action planning. The focus is on establishing a causal theory of sensor percepts, which will help in predicting future percepts and the effect of future actions. Penn Engineering: This team from Pennsylvania is focusing on challenges in machine perception by learning from biological systems. Then applying an embodied, active and efficient approach towards acquiring representations of the surrounding world and actions. University of Washington: They are addressing the challenges to enable robots working effectively in human environments. Similar to a human child learning through exploration and curiosity, they aim to build a mathematical model of curiosity. After three years, the participating universities will have to show demonstrations of working systems that will be the foundation of CMM. To know more about the initiative, visit the Curious mind machine website. SingularityNET and Mindfire unite talents to explore artificial intelligence MIT plans to invest $1 billion in a new College of computing that will serve as an interdisciplinary hub for computer science, AI, data science “Deep meta reinforcement learning will be the future of AI where we will be so close to achieving artificial general intelligence (AGI)”, Sudharsan Ravichandiran

Last week, Google announced the formation of Advanced Technology External Advisory Council, to help Google with the major issues in AI such as facial recognition and machine learning fairness. The group was announced by Kent Walker, Google's senior vice president of global affairs, and according to Kent, the council will provide diverse perspectives to Google. Google appointed eight members for the council coming from diverse fields including, behavioural economy, privacy, applied mathematics, machine learning, industrial engineering, AI ethics, digital ethics, foreign policy, and public policy. Now a group of Google employees on the selection of the council is insisting the company on the removal of, Kay Coles James, the Heritage Foundation President who promotes anti-trans and anti-immigrant thoughts. Her tweets are proof of her thoughts against the idea of LGBTQ. Heritage has even hosted a panel of anti-transgender activists, and the panel lobbied against LGBTQ discrimination protections that were proposed by congressional Democrats. https://twitter.com/KayColesJames/status/1108768455141007360 https://twitter.com/KayColesJames/status/1108365238779498497 Yesterday, a group of employees which was known as ‘Googlers Against Transphobia and Hate’ filed a petition. The petition reads, "In selecting James, Google is making clear that its version of 'ethics' values proximity to power over the wellbeing of trans people, other LGBTQ people, and immigrants. Such a position directly contravenes Google’s stated values." The petition is already been signed by more than 1k Google employees. The employees voiced their opinion in the petition, “By appointing James to the ATEAC, Google elevates and endorses her views, implying that hers is a valid perspective worthy of inclusion in its decision making, this is unacceptable.” Few researchers and civil society activists have joined in the force against the idea of anti-trans and anti-LGBTQ.  Alessandro Acquisti, a behavioral economist and privacy researcher, has declined an invitation to join the council. https://twitter.com/ssnstudy/status/1112099054551515138 Google employees and researchers wrote that appointing James to the council "significantly undermines Google’s position on AI ethics and fairness” pointing out that there have been consistent civil rights concerns around some AI technology. The petition further reads, "Not only are James’ views counter to Google’s stated values, but they are directly counter to the project of ensuring that the development and application of AI prioritizes justice over profit.” According to a few people, James’ views are uncommon and they are taking a stand for her. Cal Smith, on Medium wrote, “Her views are not uncommon, and in fact are shared by a good percentage of Americans. If you are to have a truly representative AI that prioritizes non-discrimination then you must have a wide range of views included, including those you disagree with.” It seems the petition by Google employees will definitely put some pressure over the company, considering that the intention is more about strengthening the Human Rights than anything else. But it is yet to be known what Google finally decides! Check out the letter by the Google employees here. Is Google trying to ethics-wash its decisions with its new Advanced Tech External Advisory Council? European Union fined Google 1.49 billion euros for antitrust violations in online advertising Google Podcasts is transcribing full podcast episodes for improving search results  

A shout out to all Pythonistas! Microsoft has something in store for you if you enjoy scripting in Excel. Python to be among the official Excel scripting languages According to a topic on Excel’s feedback hub opened last month, Microsoft considers adding Python for Excel. This topic has turned out to be the highly voted feature request, ever since it was put up on the hub. Microsoft recently rolled out a survey in order to gather a detailed understanding on how users would like to make use of Python within Excel. Python turns complexity to simplicity Python is one of the most popular programming languages among developers, due to its simplicity in coding and its versatility. Talking about its ranking among other programming languages, Python ranks second on the PYPL programming languages ranking. It ranks third in the RedMonk Programming Language Rankings, and fourth in the TIOBE index. If Python for Excel is sanctioned by Microsoft, one can easily work with Excel documents, Excel data, its core functions, using Python scripts replacing the current VBA scripts. This Python scripting, would not only turn out to be a substitute for VBA, but also could be a substitute to field functions (=SUM(A1:A2)). The idea of having Python as an official Excel scripting language, was highly appreciated by many users on board. Additionally, these users also chalked out that if Microsoft goes ahead in wiring Python within Excel, they also would require Python in other Microsoft office apps. As per a discussion on the Hacker news forum, a user posted that, “Much as I would love for the power of Python in Excel it is important that whatever is done is consistent across the office experience. Some of us old enough to remember the multiple versions of VB-whatever across Excel, Word, Access and that in itself was a blow to productivity” The user also added that, Microsoft should definitely choose Python and during this process it should also decide whether it would be Python with a .Net library--which has separate standard and core libraries--or IronPython. Later, this has to be done in a mechanism that provides exact same libraries and user-written code to work in the same way throughout other Microsoft Office products. Though, Microsoft would be delighted in adding such a feature for their users, still not much is known about this project. Until then we can expect great surprises from Microsoft with user side benefits.

Yesterday, Google researchers released three different research papers which describe their investigations in easy-to-adapt robotic autonomy by combining deep Reinforcement Learning with long-range planning. This research is made for people with a mobility impairment that makes them home-bound. The researchers propose to build service robots, trained using reinforcement learning to improve the independence of people with limited mobility. The researchers have trained the local planner agents in order to perform basic navigation behaviors and traverse short distances safely without collisions with moving obstacles. These local planners take noisy sensor observations, such as a 1D lidar that helps in providing distances to obstacles, and output linear and angular velocities for robot control. The researchers trained the local planner in simulation with AutoRL (AutomatedReinforcement Learning) which is a method that automates the search for RL rewards and neural network architecture. These local planners transfer to both real robots and to new, previously unseen environments. This works as building blocks for navigation in large spaces. The researchers then worked on a roadmap, a graph where nodes are locations and edges connect the nodes only if local planners can traverse between them reliably. Automating Reinforcement Learning (AutoRL) In the first paper, Learning Navigation Behaviors End-to-End with AutoRL, the researchers trained the local planners in small, static environments. It is difficult to work with standard deep RL algorithms, such as Deep Deterministic Policy Gradient (DDPG). To make it easier, the researchers automated the deep Reinforcement Learning training. AutoRL is an evolutionary automation layer around deep RL that searches for a reward and neural network architecture with the help of a large-scale hyperparameter optimization. It works in two phases, reward search, and neural network architecture search. During the reward search, AutoRL concurrently trains a population of DDPG agents, with each having a slightly different reward function. At the end of the reward search phase, the reward that leads the agents to its destination most often gets selected. In the neural network architecture search phase, the process gets repeated. The researchers use the selected reward and tune the network layers. This turns into an iterative process and which means AutoRL is not sample efficient. Training one agent takes 5 million samples while AutoRL training around 10 generations of 100 agents requires 5 billion samples which is equivalent to 32 years of training. The advantage is that after AutoRL, the manual training process gets automated, and DDPG does not experience catastrophic forgetfulness. Another advantage is that AutoRL policies are robust to the sensor, actuator and localization noise, which generalize to new environments. PRM-RL In the second paper, PRM-RL: Long-Range Robotic Navigation Tasks by Combining Reinforcement Learning and Sampling-based Planning, the researchers explain Sampling-based planners that tackle long-range navigation by approximating robot motions. In this paper, the researchers have combined PRMs with hand-tuned RL-based local planners (without AutoRL) for training robots locally and then adapting them to different environments. The researchers trained a local planner policy in a generic simulated training environment, for each robot. Then they build a PRM with respect to that policy, called a PRM-RL, over a floor plan for the deployment environment. For building a PRM-RL, the researchers connected the sampled nodes with the help of Monte Carlo simulation. The resulting roadmap can be tuned to both the abilities and geometry of the particular robot. Though the roadmaps for robots with the same geometry having different sensors and actuators will have different connectivity. At execution time, the RL agent easily navigates from roadmap waypoint to waypoint. Long-Range Indoor Navigation with PRM-RL In the third paper, the researchers have made several improvements to the original PRM-RL. They replaced the hand-tuned DDPG with AutoRL-trained local planners, which improves long-range navigation. They have also added Simultaneous Localization and Mapping (SLAM) maps, which robots use at execution time, as a source for building the roadmaps. As the SLAM maps are noisy, this change closes the “sim2real gap”, a phenomenon where simulation-trained agents significantly underperform when they are transferred to real-robots. Lastly, they have added distributed roadmap building to generate very large scale roadmaps containing up to 700,000 nodes. The team compared PRM-RL to a variety of different methods over distances of up to 100m, well beyond the local planner range. The team realized that PRM-RL had 2 to 3 times the rate of success over baseline because the nodes were connected appropriately for the robot’s capabilities. To conclude, Autonomous robot navigation can improve the independence of people with limited mobility. This is possible by automating the learning of basic, short-range navigation behaviors with AutoRL and using the learned policies with SLAM maps for building roadmaps. To know more about this news, check out the Google AI blog post. Google launches Flutter 1.2, its first feature update, at Mobile World Congress 2019 Google released a paper showing how it’s fighting disinformation on its platforms Google introduces and open-sources Lingvo, a scalable TensorFlow framework for Sequence-to-Sequence Modeling  

Yesterday, the team at GitHub along with its partners from Weights & Biases introduced the CodeSearchNet challenge evaluation environment and leaderboard. The team is also releasing a large dataset to help data scientists in building models for this task and several baseline models that highlight the current state of the art. Semantic code search involves retrieving relevant code when a natural language query is given. While dealing with other information retrieval tasks, it needs to bridge the gap between the language used in code and natural language. Also, the standard information retrieval methods don’t work effectively in the code search domain because usually there is little shared vocabulary between search terms and results. Evaluating methods for this task is very difficult, as there are no substantial datasets that were made for this task.  Considering these issues and to evaluate the progress on code search, the team is releasing CodeSearchNet Corpus and they are presenting the CodeSearchNet Challenge. The CodeSearchNet Challenge consists of 99 natural language queries and around 4k expert relevance annotations.  The CodeSearchNet Corpus  The CodeSearchNet corpus contains around 6 million functions from open-source code spanning six programming languages including Go, Java, Python, JavaScript, PHP, and Ruby. For collecting a large dataset of functions, the team used TreeSitter infrastructure, a parser generator tool and an incremental parsing library. The team is also releasing its data preprocessing pipeline for others to use as it will be a starting point in applying machine learning to code. This data is not directly related to code search but if used with related natural language description, it can help in training models.  CodeSearchNet corpus contains automatically generated query-like natural language for around 2 million functions. It also includes the metadata that indicates the original location where the data was found. CodeSearchNet Corpus collection The team collects the corpus from publicly available open-source non-fork GitHub repositories and uses libraries.io for identifying all projects which are used by at least one other project. They further sort these projects based on their ‘popularity’ by checking the number of stars and forks. The team removes the projects that do not have a license or whose license does not allow the re-distribution of parts of the project.  The team has also tokenized all the functions, including Go, JavaScript, Python, Java, PHP and Ruby with the help of TreeSitter. For generating the training data for the CodeSearchNet Challenge, the team considers those functions in the corpus hat have documentation associated with them. The CodeSearchNet Challenge The team collected an initial set of code search queries for evaluating code search models. They started by collecting the common search queries that had high click-through rates from Bing and then combined these with queries from StaQC. The team manually filtered out those queries that were clearly ‘technical keywords’ for obtaining a set of 99 natural language queries. The team then used a standard Elasticsearch installation and baseline models for obtaining 10 results per query from their CodeSearchNet Corpus. They then asked data scientists, programmers, and machine learning researchers for annotating the results for relevance to the query. For evaluating the CodeSearchNet Challenge, a method should return a set of results from CodeSearchNet Corpus for each of 99 pre-defined natural language queries.  Other interesting news in data Can a modified MIT ‘Hippocratic License’ to restrict misuse of open source software prompt a wave of ethical innovation in tech? ImageNet Roulette: New viral app trained using ImageNet exposes racial biases in artificial intelligent system GitLab 12.3 releases with web application firewall, keyboard shortcuts, productivity analytics, system hooks and more

Microsoft unveiled plans to use Windows ML and DirectX for both game development and game play at the 2018's Game Developers Conference. Announced earlier this month, Windows ML is a runtime framework for neural networks on Windows 10. The new platform enables developers to build machine-learning models, trained in Azure, right into their apps using Visual Studio and run them on their PCs. In this game-oriented reveal, Microsoft Windows ML will primarily be used to enhance the game development process. DirectX Raytracing, a new feature of the DirectX API, aims to make games look more realistic. Custom made Gameplays Microsoft will use Windows ML to help developers leverage deep neural networks (DNN) to enhance their games. They will also make use of ML to naturally adapt a game to a player's gaming style, such as a player’s in-game habits and change things on the fly. As an example, Microsoft says, "If you're someone who likes to find treasures in game but don't care to engage in combat, DNNs could prioritize and amplify those activities while reducing the amount or difficulty of battles." Better Game Development Process Apart from gameplay, Microsoft will also use Windows ML for improving the game development process. This includes using Neural networks to perform some of the more difficult parts of creating assets and graphics, leaving artists and developers free to focus on other areas. Microsoft says. "The time and money that studios could save with tools like these could get passed down to gamers in the form of earlier release dates, more realistic games, or more content to play." Visual Quality Improvements Windows ML will also be used to create and enhance Visuals. So, aliasing around objects in games can be smoothed out by tapping into machine learning models to determine the best color for each pixel. Microsoft also showcased a new part of the DirectX API. The DirectX Raytracing (DXR) will allow developers to use the DXR in DirectX 12 to bring real-time raytracing to their games. At present, it can be used to enhance certain aspects of visual quality, all the while working to be a full replacement of rasterization in the future. The full details are available on the official Microsoft Blog.

Amazon has been exploring itself in the health care sector since quite some time now. Just last year, Amazon bought the online pharmacy PillPack for $1 billion in order to sell the prescription drugs. The company introduced Amazon Comprehend Medical, a machine learning tool that allows users to extract relevant clinical information from the unstructured text in patient records. Amazon is even working with Accenture and Merck to develop a cloud-based platform for collaborators across the life sciences industry with a motive to bring innovation in the drug development research. Amazon has now taken a bigger leap by announcing its voice assistant, Alexa as HIPAA (Health Insurance Portability and Accountability Act) compliant, which means that it can work with health care and medical software developers in order to invent new programs or skills with voice and provide better experiences to their customers. With the help of Amazon Alexa, developers will design new skills to help customers manage their healthcare needs at home by simply using voice. Patients will now be able to book a medical appointment, access the hospital post-discharge instructions or check on the status of a prescription delivery, and much more just via the voice! HIPAA has been designed to protect patients in cases where their personal health information is shared with health care organizations such as hospitals. This will allow healthcare companies to build Alexa voice tools capable of securely transmitting the patient’s private information. The consumers will now be able to use new Alexa health skills for asking questions such as “Alexa, pull up my blood glucose readings” or “Alexa, find me a doctor,” and will receive a response from the voice assistant. The company further announced the launch of six voice programs including Express Scripts, My Children's Enhanced Recovery After Surgery (ERAS), Cigna Health Today, Swedish Health Connect, Atrium Health, and Livongo. These new tools allow patients to use Alexa for accessing personalized information such as prescription, progress updates after surgery, and much more. Rachel Jiang, a member of Amazon’s health and wellness team, who previously worked at Microsoft and Facebook announced that Amazon has invited six healthcare partners to use its HIPAA-compliant skills kit to build voice programs. But the company expects to get more healthcare providers on board to access its information. Jiang wrote in a post, “These new skills are designed to help customers manage a variety of healthcare needs at home simply using voice – whether it’s booking a medical appointment, accessing hospital post-discharge instructions, checking on the status of a prescription delivery, and more.” Boston Children’s Hospital now has a new HIPAA-compliant skill dubbed “ERAS” for kids that are discharged from the hospital and for their families. With the help of Alexa’s voice assistant, patients and their families or caregivers can now ask questions to the care team about their case. Even the doctors can now remotely check in on the child’s recovery process. Livongo, a digital health start-up, works with employers in order to help them in managing workers with chronic medical conditions. Livongo developed a skill for people with diabetes that uses connected glucometers that would ask about the patient’s blood sugar levels. In a statement to CNBC, Livongo’s president Jenny Schneider told that “There are lots of reasons she expects users to embrace voice technologies, versus SMS messaging or other platforms. Some of those people might have difficulty reading, or they just have busy lives and it’s just an easy option.” Express Scripts, a pharmacy benefit management organization is working towards building a way for members to check the status of their home delivery prescription via Alexa. Voice technology has been booming in the health care sector and skills like the ones mentioned above will bring health care to home and make the patients lives easy and cost-effective. John Brownstein, chief innovation officer for Boston Children’s Hospital, said, “We’re in a renaissance of voice technology and voice assistants in health care. It’s so appealing as there’s very little training, it’s low cost and convenient.” To know more about this news, check out  Amazon’s official announcement. Amazon won’t be opening its HQ2 in New York due to public protests MariaDB announces MariaDB Enterprise Server and welcomes Amazon’s Mark Porter as an advisor to the board of directors Over 30 AI experts join shareholders in calling on Amazon to stop selling Rekognition, its facial recognition tech, for government surveillance  

Data forms an integral part of every business or organization which is used to make valuable decisions based on changing circumstances. Natural Language Processing (NLP) is a widely adopted technique used by machines to understand and communicate with humans in human language. This enables human to access, analyze and extract data more intelligently from a huge amount of unstructured data. Intel AI Lab’s team of NLP researchers and developers has introduced NLP Architect, a new open-source Python library. This library can be used as a platform for future research and developing the state-of-the-art deep learning techniques for natural language processing and natural language understanding. Rapid and recent advancements in deep learning and neural network paradigms has led to the growth in NLP domain. This new library offers flexibility in implementing NLP solutions which are packed with the past and ongoing NLP research and development work of Intel AI Lab. NLP Architect overview The current version of NLP Architect offers noteworthy features which form the backbone in terms of research and practical development. All the following models are provided with required training and inference processes: It consists of NLP core models such as BIST and NP chunker that allows powerful extraction of linguistic features for NLP workflow NLU models such as intent extraction (IE), name entity recognition (NER) used for intent-based applications It consists of modules which address semantic understanding Now consists of components which hold a key for conversational AI such as chatbot applications, dialog applications and more End-to-end deep learning applications such as Q&A, reading comprehension and more Source: AI Intel Blog This library of NLP components provides the required functionality to extend NLP solutions with a range of audience. It provides excellent media for analysis and optimization of Intel software and hardware on NLP workloads. In addition to these models, new features such as data pipelines, common functional calls, and utilities related to NLP domain which are majorly used when deploying models, are added. To know more about the updates, you can refer the official Intel AI blog. How NLP Architect can be used You can train models using the provided datasets, configurations and algorithms You can train models based on your own data You can create new models or extend your existing models You can explore various common and not-so-common challenges faced in NLP domain using deep learning models You can optimize and extend the use of state-of-the-art deep learning algorithms You can integrate various modules and utilities from the library to NLP solutions Deep learning frameworks support This repository supports several open source deep learning frameworks such as: Intel Nervana Graph Intel Neon Intel-optimized TensorFlow Dynet Keras Note: We can expect the list of models to update in future. All these models will run with Python 3.5+ If you want to download the open-source Python library or want to contribute to the project by providing valuable feedback, download the code from Github. A complete documentation for all core modules with end-to-end examples can be found in their official page. Intel takes Facebook’s help on AI chip; Cisco uses AI to predict IT services; and more Introducing Intel’s OpenVINO computer vision toolkit for edge computing Facelifting NLP with Deep Learning

In a study published in Nature Machine Intelligence, researchers discovered that in some cases of machine learning it cannot be proved whether the system actually ‘learned’ something or solved the problem. They explore machine learning learnability. Axioms leading to axioms in arithmetic models We already know that machine learning systems, and AI systems in general are black boxes. You feed the system some data, you get some output or a trained system that performs some tasks but you don’t know how the system arrived at a particular solution. Now we have a published study from Ben-Davis et al that shows learnability in machine learning is undecidable. In the 1930s, Austrian logician Kurt Gödel showed that a set of axioms forming an arithmetic model lead to more axioms. In the following decades it was demonstrated that the continuum hypothesis can neither be proved nor refuted using standard mathematical axioms. The hypothesis states that no set of objects is larger in size than integers or smaller in size than real numbers. What does this have to do with machine learning? In machine learning, algorithms are designed to improve performance of certain actions with the data they are trained on. Some problems like facial recognition or recommendation engines cannot be created with regular linear programming. These are problems that can be solved today by machine learning. Machine learning learnability can be defined. A system can be considered learnable if the machine learning model can perform as the best predictor in a family of functions. This needs to be achieved under some reasonable constraints. Typically learnability in a model can be explained by analysing dimensions. But this new research shows that this is not always the case. A learning model introduced in the paper is the focus of the research: estimating the maximum (EMX) which is similar to PAC learning. The authors of the paper discover a family of functions whose learnability in EMX cannot be proved with standard mathematics. What is the EMX problem? As described in the paper, the EMX problem is: “Let X be some domain set, and let F be a family of functions from X to {0, 1} (we often think of each function f∈F as a subset of X and vice versa). Given a sample S of elements drawn i.i.d. from some unknown distribution P over X, the EMX problem is about finding a function f ∈ F that approximately maximizes the expectation EP(f) with respect to P.” In the paper, the authors present an example problem—displaying specific ads to the most frequent visitors of a website. The catch is, which visitors will visit the website is unknown. Now the EMX problem is formed as a question—what is a learner’s ability to find a function whose expected value is the largest. They show a relation between machine learning and data compression. If training data labelled by a function can be compressed, then the family from which the function originates has low complexity. Such a function is considered learnable. Monotone compression Algorithms can be used to compress data. A new one called monotone compression is introduced. They show that this compression is suitable to describe the learnability of function families in the EMX problem. A weak monotone compression is associated with the cardinality of particular infinite sets. The authors use the interval [0, 1] which contains real numbers. The results show that the finite subsets in the interval [0, 1] have monotone compression and are therefore considered learnable in EMX. But, this applied only if the continuum hypothesis is true which stands to be unprovable to date. The problem is how you define learnability In the concluding points, the paper points out an interesting perspective as to why current machine learning models get off easy without any questions about learnability. Or do they? The problem lies in how learnability is defined—as functions or as algorithms?. Current standard definitions focus on the theoretical aspect without considering computational implications. This approach in viewing learnability levies a high cost when more general types of learning is involved. You can read the research paper by Shai Ben-David and others about learnability being undecidable at the Nature journal website. Technical and hidden debts in machine learning – Google engineers’ give their perspective The US Air Force lays groundwork towards artificial general intelligence based on hierarchical model of intelligence NVIDIA demos a style-based generative adversarial network that can generate extremely realistic images; has ML community enthralled

The LF Deep Learning Foundation, a community umbrella project of The Linux Foundation, announced Horovod, started by Uber in 2017, as their new project, last year in December. Uber joined Linux Foundation in November 2018 to support LF Deep Learning Foundation open source projects. Horovod (named after a traditional Russian dance) announced at 2018 KubeCon + CloudNativeCon North America, is an open source distributed training framework for TensorFlow, Keras, MXNet, and PyTorch. It helps improve speed, as well as scales and resource allocation in machine learning training activities. The main goal of Horovod is to simplify distributed Deep Learning and make it fast. Ever since its release, Horovod has been getting leveraged across different tasks and by different companies. For instance, Uber has been using Horovod for self-driving vehicles, fraud detection, and trip forecasting. Other companies using Horovod include Alibaba, Amazon, and NVIDIA. Other contributors to the Horovod Project are Amazon, IBM, Intel, and NVIDIA. IBM uses Horovod as part of its open source deep learning solution, FfDL, and in its IBM Watson Studio. Databricks also features Horovod in their deep learning offering. Similarly, NVIDIA announced last November that it is using Uber’s Horovod to build an AI computing platform for developers of self-driving vehicles. Molly Vorwerck, Editorial Program Manager for Uber Engineering, mentioned that “Horovod was a clear choice for NVIDIA. With only a few lines of code, Horovod allowed them to scale from one to eight GPUs, optimizing model training for their self-driving sensing and perception technologies, leading to faster, safer systems”. Horovod makes it easy to take a single-GPU TensorFlow program and train it on many GPUs. Also, it is easier to achieve improved GPU resource usage figures with Horovod. It makes use of advanced algorithms and features high-performance networks that offer data scientists and other researchers the tooling to easily scale their deep learning models with high performance. Also, the open source community’s response was also very positive about Horovod. “It was very cool to see my first open source project reach so many people and be adopted so quickly..now, when I go to conferences people actually know of Horovod and they’re excited to integrate with it...all these things make me really happy”, states Alex Sergeev, Horovod Project Lead. Apart from that, Horovod also joined the existing Linux Foundation Deep Learning projects, namely, Acumos AI (an open source AI framework), Angel (a high-performance distributed machine learning platform), and EDL (Elastic Deep Learning framework). These projects have been designed to help cloud service providers build cluster cloud services using deep learning frameworks. “Uber built Horovod to make deep learning model training faster and more intuitive for AI researchers across industries. As Horovod continues to mature in its functionalities and applications, this collaboration will enable us to further scale its impact in the open source ecosystem for the advancement of AI,” said Sergeev. For more information, check out the official Horovod blog post. Uber open-sources Peloton, a unified Resource Scheduler Uber releases Ludwig, an open source AI toolkit that simplifies training deep learning models Uber releases AresDB, a new GPU-powered real-time Analytics Engine

Microsoft came out with a library, named DoWhy, earlier this week, for promoting widespread use of causal inference. Causal inference refers to the process of drawing a conclusion from a causal connection which is based on the conditions of the occurrence of an effect. Simply put, causal inference attempts to find or guess why something happened. "DoWhy" is a Python library which is aimed to spark causal thinking and analysis. It provides a unified interface for causal inference methods. There’s also automatic testing of multiple assumptions making the inference accessible to non-experts. According to Microsoft, “Our motivation for creating DoWhy comes from our experiences in causal inference studies -- ranging from estimating the impact of a recommender system to predicting likely outcomes given a life event -- we found ourselves repeating the common steps of finding the right identification strategy, devising the most suitable estimator, and conducting robustness checks, all from scratch”. DoWhy highlights the critical assumptions lying beneath causal inference analysis. It is designed using four major principles: Model a causal inference problem using assumptions. Identifying expression for the causal effect ("causal estimand"). Estimate the expression using statistical methods Verifying validity of the estimate How DoWhy works? First, DoWhy builds an underlying causal graphical model for every problem. This makes each causal assumption explicit. The graph does not have to be complete and you can provide a partial graph which represents prior knowledge about variables. The rest of the variables are automatically considered as potential confounders by DoWhy. Secondly, DoWhy distinguishes between identification and estimation. Identification of a causal effect refers to assumptions made about the data-generating process along with counterfactual expressions to specifying a target estimand. It uses the Bayesian graphical model framework to represent assumptions formally. Here the users can specify what they know and what they don’t know about the data-generation process. Thirdly, for estimation, there are methods based on the potential-outcomes framework including matching, stratification, and instrumental variables. Lastly, there are robustness tests along with sensitivity checks for testing or verifying the reliability of an obtained estimate. With this, you can test how the estimate changes with varying assumptions. The library is also capable of automatically checking the validity of obtained estimate depending on assumptions in the graphical model. DoWhy supports Python 3+ and requires packages such as numpy, scipy, scikit-learn, pandas, pygraphviz (for causal graphs plotting), networkx (for causal graphs analysis), matplotlib (for general plotting), and sympy (for symbolic expressions rendering). Microsoft plans on adding more features to the DoWhy library. This includes improved estimation support, sensitivity methods and interoperability with available estimation software. For more information, check out the official DoWhy documentation. Say hello to FASTER: a new key-value store for large state management by Microsoft NIPS 2017 Special: A deep dive into Deep Bayesian and Bayesian Deep Learning with Yee Whye Teh Microsoft launches a free version of its Teams app to take Slack head on  

“The harvesting and sharing of data by mobile phone apps is out of control” -Researchers at Oxford University A paper published on 18th October by researchers at Oxford University revealed that 90% of Google play store apps are harvesting user data and subsequently sharing it with companies like Alphabet, Twitter, Facebook, and many others. The study points out the presence of third-party trackers on nearly one million (959,000) apps from the US and UK Google Play stores. The statistics are unsettling. Around 88% of this data is handed over to ‘Alphabet’- Google’s parent company. Microsoft, Twitter, Facebook and others follow suite. Here is how they fair: The most prevalent root parent tracking companies and their subsidiaries These third-party trackers were mostly prevalent in news apps and apps aimed at children and young adults. By tracking users data- which includes information like age, location, gender, buying habits, and other miscellaneous information- companies can form a profile of users. This can then be used to send target specific ads, influence a user’s buying habits or even send political campaign messages. Considering that these trackers were hugely present in apps related to children, the paper states that allowing profiling of children without attempting to obtain parental consent, is downright unlawful. Even though there are tracker blocking software available for mobile and web, these primarily cannot control the tracking software embedded on an app’s OS. The privacy settings for an app are focussed on more specific app permissions like contact sharing, location sharing etc. In response to this research,  a Google spokesman said in a statement to Business Insider “Across Google and in Google Play, we have clear policies and guidelines for how developers and third-party apps can handle data and we require developers to be transparent and ask for user permission.” Further, they added, “If an app violates our policies, we take action.” Google also added that the researchers had “mischaracterized” some of the app’s basic functions to reach their conclusion. Head over to the research paper to obtain more information about this study. Alternatively, you can visit the dailmail.co.uk for more insights to this news. Google is missing out $50 million because of Fortnite’s decision to bypass Play Store A multimillion-dollar ad fraud scheme that secretly tracked user affected millions of Android phones. This is how Google is tackling it. All new Android apps on Google Play must target API Level 26 (Android Oreo) or higher, to publish