Tech News - Data

Tensorflow 1.6.0, Pandas on Ray, Google-Landmarks, Microsoft’s Custom Vision and  Face API and more in today’s top stories around machine learning, deep learning, and data science news. 1. Tensorflow v1.6.0 finally makes its debut Tensorflow 1.6.0 has finally released after two release candidates. The breaking changes, major features, and improvements include: Pre-built binaries are now built against CUDA 9.0 and cuDNN 7. Pre-built binaries will use AVX instructions. New Optimizer internal API for non-slot variables. tf.estimator.{FinalExporter,LatestExporter} now export stripped SavedModels. FFT support added to XLA CPU/GPU. Android TF can now be built with CUDA acceleration on compatible Tegra devices. To know about Bug Fixes and other changes, you may visit the GitHub repo. 2. Pandas on Ray, A DataFrame library for making Pandas faster The team at UC Berkeley are developing a DataFrame library that wraps Pandas and transparently distributes the data and computation. The early stage library, Pandas on Ray, can accelerate Pandas queries by 4x on an 8-core machine, only requiring users to change a single line of code in their notebooks. Pandas on Ray is targeted towards existing Pandas users who are looking to improve performance and see faster runtimes without having to switch to another API. The ultimate goal of this project is to be able to use Pandas in a cloud setting. 3. Google launches Landmarks, a new Dataset for Landmark Recognition Google has released Google-Landmarks, the largest worldwide dataset for recognition of human-made and natural landmarks. The dataset contains more than 2 million images depicting 30 thousand unique landmarks from across the world and a number of classes that is ~30x larger than what is available in commonly used datasets. Additionally, Google is also open-sourcing Deep Local Features (DELF), an attentive local feature descriptor. They have also launched two Kaggle challenges. The recognition track challenge is to build models that recognize the correct landmark in a dataset of challenging test images, while the retrieval track challenges participants to retrieve images containing the same landmark. 4. Microsoft Azure adds computer vision and image processing capabilities Microsoft has updated its Azure platform with computer vision capabilities with the launch of Custom Vision, a service that lets developers train models for processing specific kind of images. Alongside Custom Vision, the company also made its Face API service for face and emotion detection generally available. The major improvement in Face API includes a scalability boost that enables the service to recognize up to a million different individuals within images. It also launched Bing Entity Search, which allows developers to harness Microsoft’s search engine to help users find needed information within their application. 5. Intela launches Farrago, an online tool that uses machine learning to clean up dirty data Data science company Intela AI, launches Farrago, a machine learning tool to clean up dirty data. This tool can automate the manual work of identifying and removing duplicate records from databases. It can also analyze a company’s data and intelligently recommend the best way to organize, clean and transform it. According to Intela CEO Asa Cox, “Farrago could save a company, client or programme, hundreds of man-hours of time spent manually (or semi-manually) cleaning data.” The online demonstration of Farrago is readily available.  

Certifiable Distributional Robustness with Principled Adversarial Training, a paper accepted for ICLR 2018, is a collaborative effort of Aman Sinha, Hongseok Namkoong, and John Duchi. In this paper, the authors state the vulnerability of neural networks to adversarial examples and further take the perspective of a distributionally robust optimization which guarantees performance under adversarial input perturbations. Certifiable Distributional Robustness with Applying Principled Adversarial Training What problem is the paper trying to solve? Recent works have shown that neural networks are vulnerable to adversarial examples; seemingly imperceptible perturbations to data can lead to misbehavior of the model, such as misclassifications of the output. Many researchers proposed adversarial attack and defense mechanisms to counter these vulnerabilities. While these works provide an initial foundation for adversarial training, there are no guarantees on whether proposed white-box attacks can find the most adversarial perturbation and whether there is a class of attacks such defenses can successfully prevent. On the other hand, verification of deep networks using SMT (satisfiability modulo theories) solvers provides formal guarantees on robustness but is NP-hard in general. This approach requires prohibitive computational expense even on small networks. The authors take the perspective of distributionally robust optimization and provide an adversarial training procedure with provable guarantees on its computational and statistical performance. Paper summary This paper proposes a principled methodology to induce distributional robustness in trained neural nets with the purpose of mitigating the impact of adversarial examples. The idea is to train the model to perform well not only with respect to the unknown population distribution, but to perform well on the worst-case distribution in a Wasserstein ball around the population distribution. In particular, the authors adopt the Wasserstein distance to define the ambiguity sets. This allows them to use strong duality results from the literature on distributionally robust optimization and express the empirical minimax problem as a regularized ERM (empirical risk minimization) with a different cost. Key takeaways The paper provides a method for efficiently guaranteeing distributional robustness with a simple form of adversarial data perturbation. The method values strong statistical guarantees and fast optimization rates for a large class of problems. Empirical evaluations indicate that the proposed methods are in fact robust to perturbations in the data, and they outperform less-principled adversarial training techniques. The major benefit of this approach is its simplicity and wide applicability across many models and machine-learning scenarios. Reviewer comments summary Overall Score: 27/30 Average Score: 9 The reviewers have strongly accepted this paper and have stated that it is of a great quality and originality. They said that this paper is an interesting attempt, but some of the key claims seem to be inaccurate and miss comparison to proper baselines. Another reviewer said, the paper applies recently developed ideas in the literature of robust optimization, in particular distributionally robust optimization with Wasserstein metric, and showed that under this framework for smooth loss functions when not too much robustness is requested, then the resulting optimization problem is of the same difficulty level as the original one (where the adversarial attack is not concerned). The paper has also received some criticisms but at the end of all it is majorly liked by many of the reviewers.

Apache Spark 2.3 now on Databricks Runtime 4.0 Beta, Twitter donates Heron to Apache Software Foundation, new Blockchain-based platform to build AI apps, and more in today’s top stories around machine learning, deep learning, and data science news. 1. Apache Spark 2.3 Now Available on Databricks Runtime 4.0 Beta Databricks announced the availability of Apache Spark 2.3.0 on Databricks as part of its Databricks Runtime 4.0 beta. The Spark 2.3: Marks a major milestone for Structured Streaming by introducing low-latency continuous processing and stream-to-stream joins. Boosts PySpark by improving performance with pandas UDFs Runs on Kubernetes clusters by providing native support for Apache Spark applications. The release extends new functionality to SparkR, Python, MLlib, and GraphX. It also focuses on usability, stability, and refinement, resolving over 1400 tickets. For additional features and other information, read the Spark 2.3 release notes. 2. New Blockchain-Based Platform to Collectively Build AI Apps Dbrain, a new project built on the Ethereum Blockchain leverages smart contracts to develop a simple tool that allows everyone to label and validate data in exchange for cryptocurrency. Dbrain introduces a platform that targets businesses and data scientists that need the data to develop AI solutions. By building smart contracts on Ethereum, Dbrain plans to use its internal protocols to solve fundamental AI-based development, execution, and adoption challenges which include: Dataset quality Trust and security Infrastructure costs It aims to have a complete AI production line integrated with its platform. Thus it combines labeling functionality and ensures that payment validation is transparent. It also aims to provide customized AI solutions within a single product. To know more about the challenges in detail, visit Dbrain’s blogpost. 3. Now on GitHub: The Autonomous Driving Cookbook from Microsoft as Jupyter Notebooks The Autonomous Driving Cookbook from Microsoft is now available on GitHub. The cookbook is an open source collection of scenarios, tutorials, and demos to help you quickly onboard various aspects of the autonomous driving pipeline. It is an ongoing project developed and maintained by the Deep Learning and Robotics chapter of Microsoft Garage, the team that helped develop the recent expansion of AirSim to include cars for autonomous driving research. Tutorials in the cookbook are presented as Jupyter notebooks, making it very easy for you to download the instructions and get started without a lot of setup time. To help this further, wherever needed, tutorials come with their own datasets, helper scripts and binaries. Read more of this in detail at Microsoft + Open Source Blogpost here. 4. Tenet Partners Launches Data Analytics Platform Tenet Partners announced CoreBrand® Data Science, a new business unit leveraging the power of predictive analytics and data science to transform how corporations and capital markets can generate value from their brands. Tenet Partners help the C-suite to drive positive business outcomes by using a combination of research that underpins the CoreBrand® Index and advanced analytics. Read the official press release for detailed information about this launch. 5. Twitter donates Heron to Apache Software Foundation Twitter announced that it is donating Heron to the Apache Incubator where the community will continue to grow and thrive under the guidance of the Apache Software Foundation. Heron is a real-time analytics platform developed by Twitter to reliably process billions of events generated at Twitter every day. Open-sourced in 2011, it is the next generation distributed streaming engine that was built to be backwards compatible with Apache Storm. It was built to improve Twitter’s developer and operational experiences with Storm and introduced a wide array of architectural improvements and native support for Apache Aurora. Heron has become Twitter’s primary streaming system, reliabily powering all of Twitter’s real-time analytics and running hundreds of development and production topologies deployed on thousands of nodes. For more, read Twitter’s official announcement.  

Algorithmia’s AI smart contract, Microsoft’s ML server 9.3, PostgreSQL 10 supported in Amazon RDS, Bitcoin Core 0.16.0, and more in today’s top stories around machine learning, blockchain, and data science news. 1. Algorithmia has developed an AI smart contract with a neural network running on the Ethereum blockchain Algorithmia Inc, the AI and ML algorithm marketplace provider, has created the first ever AI smart contract with a neural network running on the Ethereum blockchain. The contract basically offers a bounty for developers to create an AI model that can determine voter preferences based on their latitude and longitude. The smart contract will use the blockchain to automatically validate the solution. Here’s how the model works: The buyer creates a new contract. The contract is published to the Ethereum blockchain. Machine Learning engineers download the data and train an AI/ML model. The model is submitted and run on the ethereum blockchain using the data set from the contract. If the model fulfils the criteria of the contract, the model is sent to the buyer and payment sent to the ML engineer. 2. Microsoft Machine Learning Server 9.3 releases Microsoft Machine Learning Server 9.3 has been released. Key areas of change in the 9.3 release include: Set-up and configuration of Operationalization. Platform upgrades, better-together with Azure ML. Support for local Spark. Improved revoscalepy library. Linux R-Client support for SQL Server compute context. More partnerships and solution templates.   Microsoft Machine Learning Server 9.3 can be downloaded from Visual Studio Dev Essentials, or via ML Server VMs in Azure. It comes packed with the power of the open source R and Python engines, making both R and Python ready for enterprise-class ML and advanced analytics. 3. PostgreSQL 10 is now supported in Amazon RDS Amazon RDS for PostgreSQL now supports PostgreSQL major version 10. Amazon RDS for PostgreSQL makes it easy to set up, operate, and scale PostgreSQL deployments in the cloud. To use the new versions, users can create an Amazon RDS for PostgreSQL database instance with just a few clicks in the AWS Management Console, or upgrade an existing instance using point-and-click upgrades. PostgreSQL 10 includes various new features including native table partitioning, support for improved parallelism in query execution, ICU collation support, column group statistics, enhanced postgres_fdw extension, and many more. 4. Bitcoin Core 0.16.0 is now released Bitcoin Core version 0.16.0 is now available. This is a new major version release, including new features, various bug fixes, performance improvements, as well as updated translations. Bitcoin Core 0.16.0 introduces full support for segwit in the wallet and user interfaces. Version 0.16.0 will only create hierarchical deterministic (HD) wallets. It now has more flexibility in where the wallets directory can be located. The minimum version of the GCC compiler required to compile Bitcoin Core is now 4.8. Pruned nodes can now signal BIP159's NODE_NETWORK_LIMITED using service bits, in preparation for full BIP159 support in later versions. A new RPC ‘rescanblockchain’ has been added to manually invoke a blockchain rescan. Safe mode is now disabled by default and must be manually enabled. The `validateaddress` RPC output has been extended with a few new fields, and support for segwit addresses. The detailed report is available in the change log. 5. Introducing Draw.io JupyterLab extension, a Diagram Editor for JupyterLab The Draw.io JupyterLab extension is a LaTeX editor for JupyterLab which is an easy way to live-compile text documents, diagrams, flow charts and draw figures. The Draw.io JupyterLab extension takes advantage of the JupyterLab architecture: i.e. registering a new mime type (.dio) with the file explorer to open files, and adding a launcher button and menu items. It also provides multiple synchronized views of the same diagrams, displayed at the same time. It allows a user to visualize the same content with different zoom levels, or with a bare text editor. The entire code is available on GitHub.

To keep up with Project Jupyter’s motto of developing open-source software, open-standards, and services with a goal to offer interactive computing across various programming languages they released JupyterLab beta readily available for users this month. JupyterLab is tagged as the next generation UI for Project Jupyter, and is a successor to Jupyter Notebooks, a successful and a widely adopted application launched by Project Jupyter last year. Saying hello to JupyterLab Jupyter Notebook is an open-source web application that allows users to create and share documentations that contains live code, visualizations, narrative text, and equations. Jupyter notebooks are used for tasks such as data cleaning, data transformation, numerical simulation, machine learning, and many more. It is now well established that the data science community  loves using Jupyter Notebooks for interactive computing. However, there are certain barriers they face which made their interaction with Jupyter Notebook a little less than ideal. Some of the cons include: Transition from different building blocks within a workflow is difficult Real-time collaboration of notebooks onto Dropbox or Google Drive is not possible with Jupyter Notebooks. Too many wasted spaces on the right and left of the Jupyter notebook These are some of the issues with Jupyter Notebooks, which are taken care of in the brand new JupyterLab. A swift move to JupyterLab JupyterLab has complete support for Jupyter Notebooks. So, one won’t miss working with notebooks but can do a lot more using JupyterLab. JupyterLab is an interactive environment which allows you to work with notebooks, code, and data, all under one roof. The most important feature of JupyterLab is real-time collaboration with several people on a single project. An add-on to this is its user-friendly interface, which makes it all the more easy-to-use. JupyterLab also shows a high level of integration between notebooks. This means, you can drag-and-drop notebooks cells and can also copy them between notebooks. You can also run code blocks from text files with .py, .R, .tex extensions. JupyterLab can also multi-task, i.e. you can open up notebooks, text editors, terminals, and other components, view them and edit them in different tabs simultaneously. JupyterLab offers an entire range of extensions which could be used to enhance parts of JupyterLab. One can choose from a variety of themes, editors, and renderers for rich outputs on notebooks. JupyterLab extensions are npm packages (the standard package format in Javascript development). There are also many community-developed extensions being built on GitHub. To find extensions, you can search GitHub for jupyterlab-extension. You can also check out the developer documentation guide for information on developing extensions. Some additional features of JupyterLab include: JupyterLab is more about development unlike Jupyter Notebook which focuses on presentation. Developers can perform syntax completion using the Tab key and object tool-tip inspection, using the Shift-Tab keys. Files can be opened up in variety of formats. Also, developers can run their codes interactively inside of 'consoles' and not only notebooks. This promotes an imperative programming mode for them. JupyterLab accommodates notebooks in multiple languages, provided the kernels for those languages are installed. Browsers such as Chrome, Firefox, and Safari are compatible with JupyterLab. The Jupyter community plans to unleash version 1.0 of JupyterLab some time later this year. The version 1.0 will replace the classic Jupyter Notebook. However, the notebook document format would be supported by both classic notebook as well as JupyterLab. For a further detailed information on JupyterLab beta, visit Project Jupyter’s official blogpost                                                      

4th cumulative update release for #SQLServer 2017, MongoDB 3.6.3 ready for production, Google debuts AdSense ‘auto ads’ with machine learning, and more in today’s top stories around machine learning, deep learning,and data science news. 1. SQL Server 2017 4th Cumulative Update release Microsoft released the 4th cumulative update for SQLServer 2017 RTM (Release To Market). This update includes all fixes introduced after the release of SQL Server 2017. This means one can install this to resolve issues fixed in any of the previous RTM CU. The latest 2017 update is CU4 - KB4056498. For information about the fixes and other improvements, read the release notes. 2. MongoDB 3.6.3 is out and ready for production deployment MongoDB version 3.6.3 is out. This version includes some minor fixes and is a recommended upgrade for all 3.6 users. The fixes in 3.6.3 include: 3.6 mongod crash on find with index and nested $and/$or Tailing oplog on secondary fails with CappedPositionLost specifying --bind_ip localhost results in error “address already in use” All JIRA issues closed in 3.6.3 For more information read the 3.6.3 Changelog. 3. Google releases ML-powered AdSense ‘Auto Ads’ Google announces AdSense ‘Auto Ads’. This is a brand new ad format which makes use of machine learning to read any web page. It is highly optimized, easy-to-use, and is capable of increasing revenue opportunities for any business on the web. It detects and places ads that are appropriate to be placed on that page. This also includes where to place the ads and how many ads to run. Publishers can also activate Auto ads with just a single line of code. Using machine learning not only helps to decide where the ad should be placed, but it is also used to ingest analytics for how well that ad performs. This can teach the system how to place ads better in the future. To know more about Auto Ads and how it works, visit AdSense Auto Ads’ official blog post. 4. Microsoft updates its Quantum Development Kit with support for Linux and Mac Microsoft announced a major update to its Quantum Development Kit. This update will enable more developers to experience the power of Quantum computing on more platforms. The update comprises of: Support for Mac- and Linux-based development Full open source license for our quantum development libraries and samples Interoperability with the Python programming language Faster simulator performance For a detailed know-how on these enhancements, read Microsoft’s official blog. 5. Pendo Systems Releases Version 4.0 Pendo Systems released version 4.0 of their Pendo Machine Learning Platform (PMLP). This new version includes an improved machine learning toolset, which accelerates time to implementation and is capable of tackling highly complex machine learning processing challenges. Version 4.0: Creates training data via an enhanced UI which helps streamline the complex management, classification and processing of all documents and enables users to train models against it. New connectivity options with CMIS (Content Management Interoperability Services) support and web crawling. Includes new plugins that integrate seamlessly with other systems to provide access to a range of Machine Learning algorithms. Read more about this in the official press release.

Alibaba’s 11-qubit quantum computer, InfluxDB’s support for ephemeral data, regres releases on CRAN, OpenAI’s MADDPG, and more in today’s top stories around machine learning, blockchain, and data science news. 1. Alibaba launches its 11-qubit quantum computing service Alibaba has progressed towards quantum computing with the launch of its 11-qubit quantum computing service. This is a joint venture between Alibaba’s cloud service subsidiary Aliyun and the Chinese Academy of Sciences. This service is available to the public on the Quantum Computing Cloud Platform. Alibaba Quantum Lab (AQL) has also released an ambitious 15-year roadmap. By 2025, it expects to have built quantum computers that will be the world’s fastest by today’s measure. By 2030, AQL hopes to achieve a general quantum computing prototype with 50–100 qubits. Aliyun is also offering a new 32-qubit quantum computer simulation service. By comparing simulated experiment results with real results on quantum computers, users can measure the latter’s performance, verify correctness, etc. 2. InfluxDB adds support for ephemeral data to its databases InfluxData Inc. have updated their Time-series database platforms with support for Ephemeral data. Ephemeral data refers to data that only exists for a very short period of time. It is increasingly being generated by new technology deployments such as software containers, Kubernetes and IoT sensors. The nature of this data makes it troublesome for existing database solutions to keep up with the influx of this temporary data. To counter this issue, InfluxData has built two time-series databases called InfluxDB and InfluxEnterprises, which are designed to query time-stamped metrics, events and measurements more efficiently than traditional relational databases. InfluxDB boasts significant number of users, including IBM Corp. which uses the platform to analyze operational information in real time. 3. regres releases on CRAN regres is now released in CRAN. reqres is a new (in R context) approach to working with HTTP messages, that is, the requests send to a server and the response it returns. There are two main objects in reqres, the Request class and the Response class. Both of these are built on R6 and heavily inspired by the request and response classes in Express.js (a web server framework for Node.js). With regres launched in CRAN, working directly with HTTP messages will be simplified as reqres takes care of the minimum requirements letting the developers focus on the server logic instead. 4. Open AI releases MADDPG, an algorithm for multi-agent reinforcement learning Open AI researchers have developed a new algorithm for centralized learning and decentralized execution in multiagent environments. Called the MADDPG, this algorithm allows agents to learn to collaborate and compete with each other. MADDPG extends a reinforcement learning algorithm called DDPG, taking inspiration from actor-critic reinforcement learning techniques. They treat each agent as an “actor”, and each actor gets advice from a “critic” that helps the actor decide what actions to reinforce during training. More information is available at the OpenAI blog. 5. ServiceNow launches Agent Intelligence to make machine learning more accessible to organizations ServiceNow have added machine-learning capabilities directly into the Now Platform, making it accessible to all their cloud services and other applications built on ServiceNow. Their Agent Intelligence ML solution will automate the categorization, prioritization and assignment of work to reduce resolution times, minimize human error and improve customer satisfaction. It can also quickly classify and route requests with fewer errors, increasing agent productivity. Agent Intelligence will initially be applied to improving the speed and quality of IT and customer-service processes.

Microsoft announces SQLServer update for mssql-cli, ENAS-PyTorch, NumPy v1.14.1 released, and more in today’s top stories around machine learning, deep learning,and data science news. 1. Microsoft announces SQLServer update for mssql-cli Microsoft announced a new update for its mssql-cli, which is a new and interactive command line query tool for SQL Server. Mssql-cli is an open source tool that works cross-platform and is part of the dbcli community. In this v1.0.0, the feature highlights are the special commands. Microsoft in its blog states that these special commands make various executions easier. They are shortcuts to perform common tasks and queries. All special commands start with a backslash (), and one can use the built-in IntelliSense to see a list of special commands that they can use. Read more at SQL Server Blog. 2. ENAS-PyTorch: A PyTorch implementation of "Efficient Neural Architecture Search via Parameters Sharing” Introducing ENAS-PyTorch, a PyTorch implementation of "Efficient Neural Architecture Search via Parameters Sharing. ENAS reduce the computational requirement (GPU-hours) of Neural Architecture Search (NAS) by 1000x. This is done via parameter sharing between models that are subgraphs within a large computational graph. To know more in detail, visit the GitHub repository. 3. NumPy v1.14.1 released NumPy version 1.14.1 released. This is a bugfix release for some problems reported following the 1.14.0 release. The major problems fixed include: Problems with the new array printing, particularly the printing of complex values. Problems with np.einsum due to the new optimized=True default. Some fixes for optimization have been applied and optimize=False is now the default. The sort order in np.unique when axis=<some-number> will now always be lexicographic in the subarray elements. In previous NumPy versions there was an optimization that could result in sorting the subarrays as unsigned byte strings. The change in 1.14.0 that multi-field indexing of structured arrays returns a view instead of a copy has been reverted but remains on track for NumPy 1.15. To know more, read NumPy’s release notes. 4. Feature Labs Launches Software Solutions to Automate Feature Engineering for Machine Learning and AI Applications Feature Labs, Inc., launched a set of tools to aid data scientists build machine learning algorithms more quickly. As stated by Max Kanter, CEO and founder of Feature Labs, the company plans to automate ‘feature engineering’, a time consuming and manual process for data scientists. Feature Labs uses “Deep Feature Synthesis” to automatically create features from raw relational and transactional datasets. Max Kanter also said,“Feature Labs is unique because we automate feature engineering, which is the process of using domain knowledge to extract new variables from raw data that make machine learning algorithms work.” Read more about this news in detail on Feature Labs’ official website. 5. SentryOne Releases Version 18.1 with Enhanced Support of SSAS Tabular SentryOne released Version 18.1of its SentryOne Platform. This updated version has an enhanced support of SSAS Tabular in BI Sentry. Bi Sentry is the complete performance monitoring, diagnosis, and optimization solution for SQL Server Analysis Services (SSAS). Jason Hall, SentryOne Vice President of Product, said, “This update also introduces general performance enhancements to the SentryOne client, and additional performance enhancements to our APS and Azure SQL DW products.” For a more detailed information read the official press release.  

MariaDB MaxScale v2.2, Accenture AI testing services, Qualcomm’s AI engine, AllenNLP v0.4, and more in today’s top stories around machine learning, deep learning, and data science news. 1. MariaDB MaxScale 2.2, an advanced database proxy for MariaDB, is now generally available MariaDB has announced the general availability of MariaDB MaxScale 2.2. MaxScale is an advanced database proxy for MariaDB. The version 2.2 hosts a variety of new features including Replication cluster failover management High availability of MaxScale Security features for General Data Protection Regulation (GDPR) compliance, readiness of upcoming MariaDB Server 10.3 Improved management interface Proxy Protocol, to ease out configuration and authorization of users by eliminating the need to duplicate them in both MariaDB MaxScale and MariaDB Server To know the entire changes, have a look at the release notes. 2. Accenture announces new services for testing AI systems Accenture launches new AI testing services. These services are built on a “Teach and Test” methodology designed to help companies build, monitor and measure reliable AI systems. The “Teach” phase emphasizes the choice of data, models, and algorithms that are used to train Machine Learning. In the “Test” phase, AI outputs are compared with the main performance indicators and analyzed for whether the system can explain how a decision or outcome was determined by using innovative techniques and Cloud-based tools to monitor the system. Accenture has used this methodology to train a conversational virtual agent for a financial services company’s website. The agent was trained 80 percent faster than previously possible and achieved an 85 percent accuracy rate on customer recommendations. 3. Qualcomm launches its new Artificial Intelligence Engine To help developers provide better machine learning-based enhancements, Qualcomm has launched a new AI engine. The Qualcomm Artificial Intelligence Engine consists of several hardware and software components that can be used by app developers to provide “AI-powered user experiences”, with or without a network connection. Key features include: Snapdragon Neural Processing Engine (NPE) software framework to accelerate AI user experiences on a device. The Snapdragon NPE supports Tensorflow, Caffe and Caffe2 frameworks, in addition to the Open Neural Network Exchange (ONNX) interchange format. Support for the Android Neural Networks API, giving developers access to Snapdragon platforms directly through the Android operating system. Hexagon Neural Network (NN) library allowing developers to run AI algorithms directly on the Hexagon Vector Processor. 4. Microsoft Azure Notebooks will now let users learn Data Science, free of charge Microsoft has made it easier to create and share live, working code an easier process with its Microsoft Azure Notebooks service. This notebook is now available free of charge and allows data science enthusiasts to learn programming and data science outside of traditional schooling. Microsoft Azure Notebooks lets users get started quickly on tasks such as data visualization and prototyping, all within a web browser. It's an implementation of the popular open-source Jupyter Notebooks service and is available to anyone who creates a free account. 5. AllenNLP, an open-source NLP research library built on PyTorch, releases its version 0.4 AllenNLP has released the version 0.4 of their NLP research library, which is built on PyTorch. The major changes include: Inclusion of ELMo which produces contextualized word embeddings that greatly improve model performance. Support for lazy datasets: Users can now stream data through the trainer with a lower memory footprint. First-class support for models that operate on spans instead of tokens. Support for programmatically importing additional dependencies. A simple server to create a stand-alone web demo for a model. Constrained decoding added to the ConditionalRandomField module (and to the corresponding NER tagger model) Additional features and bug fixes are available in the GitHub repo.

JupyterLab now user-ready, JetBrains announces beta version of Datalore, Baidu’s latest research paper on Speech Synthesis, and more in today’s top stories around machine learning, deep learning,and data science news. 1. JupyterLab is now ready for users Jupyter team announced the beta release of JupyterLab, the next-generation web-based interface for Project Jupyter. JupyterLab consists of an interactive development environment for working with notebooks, code and data. JupyterLab provides a high level of integration between notebooks, documents, and activities: Drag-and-drop to reorder notebook cells and copy them between notebooks. Run code blocks interactively from text files (.py, .R, .md, .tex, etc.). Link a code console to a notebook kernel to explore code interactively without cluttering up the notebook with temporary scratch work. Edit popular file formats with live preview, such as Markdown, JSON, CSV, Vega, VegaLite, and more. Read more on the Jupyter blog. 2. JetBrains announces beta version of Datalore: A web application for machine learning Jet Brains has launched a public beta of Datalore, an intelligent web application for data analysis and visualization in Python. Datalore includes features such as: Intelligent and easy-to-use code editor Incremental computations Out-of-the-box machine learning tools Real-time collaboration Different computational instances To read about the features in detail visit JetBrains’ official blog. 3. Baidu’s latest breakthrough in Speech Synthesis Baidu Research recently rolled out a new research paper on “Neural Voice Cloning with a Few Samples” This paper focuses on two fundamental approaches for solving the problems with voice cloning. Firstly, speaker adaptation and secondly, speaker encoding . Both these techniques can be adapted to a multi-speaker generative speech model with speaker embeddings, without degrading its quality. In terms of naturalness of the speech and similarity to the original speaker, both demonstrate good performance, even with very few cloning audios. Read the research paper, for a complete information on this topic. 4. IBM and Unity collaboration brings Watson into virtual reality environments IBM announced its collaboration with Unity to build a development kit for IBM platform. This platform will let companies to draw on IBM’s cloud-based Watson artificial intelligence suite into their projects. The features that this collaboration would bring in are: Ability to analyze the objects in a virtual environment using Watson Visual Recognition The development kit would allow Unity developers to configure games and projects in order to understand speech, communicate with users, and understand the intent of a user in natural language. Watson's Vision API will also allow developers to integrate real-time visual recognition into their Unity projects. Visit Unity’s official blog post for a details on the extended features of this collaboration. 5. Satoshipowered.ai wants to link VR and blockchain Satoshipowered.ai (SAI), a decentralized Autonomous Game Development and Crowd publishing Organization, stated that it wants to use blockchain’s decentralized bookkeeping to give players true ownership over digital goods, which could introduce economic scarcity to games with a focus on virtual worlds. Satoshipowered.ai (SAI) announced that it will make use of the Ethereum blockchain, a cryptocurrency that allows anyone to spin up their own customized digital coin. Developers can rework the blockchain that keeps track of Ethereum to also keep track of any other records. To know more about this in detail, read more here.

Kepler brings blockchain for AI, ARM releases two AI chip designs, Coinbase commerce plugin, and more in today’s top stories around machine learning, blockchain, and data science news. 1. Kepler Technologies plans to build a decentralized ecosystem for the development of AI and robotics. Kepler Technologies, a blockchain-based startup, plans to use blockchain-based solutions for the development of AI and robotics projects. Their decentralized platform is backed by smart contracts and powered by proprietary analytical algorithms. Every proposal presented to Kepler Technologies will be recorded to the blockchain through its’ innovative Proof-of-Creation network protocols. The platform’s innovative KEP token is the driver behind all voting and funding on the ecosystem. The KEP tokens will be used to incentivize behaviors on the platform in a decentralized way. It will also let users buy products at a discount, vote for or against projects to be developed, and funding proposals that are accepted. The company has also established a blockchain tech incubation platform for the users to exhibit their ideas and connect with global investors to gain financial support. Innovators can connect with each other from anywhere in the world to develop their project. 2. ARM releases two new AI chip designs for mobile devices ARM, has released designs for two new AI processors for delivering large amounts of computational capabilities to mobile devices. The first, ARM Machine Learning (ML) Processor, which will speed up general AI applications from machine translation to facial recognition. The second, ARM Object Detection (OD) Processor is a second-generation design optimized for processing visual data and detecting people and objects. The company said that its Arm ML processors can handle more than 4.6 trillion operations per second while drawing very little power.  Devices using the ARM ML processor will be able to perform ML independent of the cloud.  The OD processor is expected to be available to industry customers at the end of this month, while the ML processor design will be available sometime in the middle of the year. 3. Coinbase unveils a new plugin For Ethereum, Bitcoin, and other cryptocurrencies Coinbase, the popular crypto broker, has launched a new PayPal like plugin service for cryptocurrency merchants. This feature allows them to seamlessly integrate crypto payments by adding a Coinbase Commerce button. The plugin is available for Ethereum, Bitcoin, Bitcoin Cash and Litecoin. Previously, their merchants' service was directly integrated with Coinbase, requiring a Coinbase account. Now it’s just a  seamless crypto integration option, no different than paying through credit card, or Paypal. 4. IBM plans to use blockchain technology to aid the government IBM wants to use blockchain technology in US governance processes to help make services more secure. According to, IBM's vice-president of blockchain technology, Jerry Cuomo, “US government should employ the digital ledger technology for services such as paying taxes, creating secure identities, tracking food and drug shipments, among other purposes”. He preferred integrating blockchain into existing government projects and programmes rather than creating new projects based on the technology. The federal and state governments in the US are already working on several experimental projects based on  blockchain, with some states working on implementing blockchain-based drivers licenses and identification cards. IBM itself is working with the Centers for Disease Control and Prevention in implementing blockchain to increase the speed of CDC's ability to develop new drugs. 5. Hyperband, Hyperparameter Optimization for PyTorch A new PyTorch implementation of Hyperband is in development. HyperBand is a hyperparameter optimization algorithm that exploits the iterative nature of SGD and the embarrassing parallelism of random search. Unlike Bayesian optimization methods which focus on optimizing hyperparameter configuration selection, HyperBand poses the problem as a hyperparameter evaluation problem. It adaptively allocates more resources to promising configurations while quickly eliminating poor ones. This allows it to evaluate orders of magnitude more hyperparameter configurations. It is described in the paper Hyperband: A Novel Bandit-Based Approach to Hyperparameter Optimization by Lisha Li, Kevin Jamieson, Giulia DeSalvo, Afshin Rostamizadeh and Ameet Talwalkar. The implementation details are available in the GitHub repo.

This ICLR 2018 accepted paper, Deep Mean Field Games for Learning Optimal Behavior Policy of Large Populations, deals with inference in models of collective behavior, specifically at how to infer the parameters of a mean field game (MFG) representation of collective behavior. This paper is authored by Jiachen Yang, Xiaojing Ye, Rakshit Trivedi, Huan Xu, and Hongyuan Zha. The 6th annual ICLR conference is scheduled to happen between April 30 - May 03, 2018. Mean field game theory is the study of decision making in very large populations of small interacting agents. This theory understands the behavior of multiple agents each individually trying to optimize their position in space and time, but with their preferences being partly determined by the choices of all the other agents. Estimating the optimal behavior policy of large populations with Deep Mean Field Games What problem is the paper attempting to solve? The paper considers the problem of representing and learning the behavior of a large population of agents, to construct an effective predictive model of the behavior. For example, a population’s behavior directly affects the ranking of a set of trending topics on social media, represented by the global population distribution over topics. Each user’s observation of this global state influences their choice of the next topic in which to participate, thereby contributing to future population behavior. Classical predictive methods such as time series analysis are also used to build predictive models from data. However, these models do not consider the behavior as the result of optimization of a reward function and so may not provide insight into the motivations that produce a population’s behavior policy. Alternatively, methods that employ the underlying population network structure assume that nodes are only influenced by a local neighborhood and do not include a representation of a global state. Hence, they face difficulty in explaining events as the result of uncontrolled implicit optimization. MFG (mean field games) overcomes the limitations of alternative predictive methods by determining how a system naturally behaves according to its underlying optimal control policy. The paper proposes a novel approach for estimating the parameters of MFG. The main contribution of the paper is in relating the theories of MFG and Reinforcement Learning within the classic context of Markov Decision Processes (MDPs). The method suggested uses inverse RL to learn both the reward function and the forward dynamics of the MFG from data. Paper summary The paper covers the problem in three sections-- theory, algorithm, and experiment.  The theoretical contribution begins by transforming a continuous time MFG formulation to a discrete time formulation and then relates the MFG to an associated MDP problem. In the algorithm phase, an RL solution is suggested to the MFG problem. The authors relate solving an optimization problem on an MDP of a single agent with solving the inference problem of the (population-level) MFG. This leads to learning a reward function from demonstrations using a maximum likelihood approach, where the reward is represented using a deep neural network. The policy is learned through an actor-critic algorithm, based on gradient descent with respect to the policy parameters. The algorithm is then compared with previous approaches on toy problems with artificially created reward functions. The authors then demonstrate the algorithm on real-world social data with the aim of recovering the reward function and predicting the future trajectory. Key Takeaways This paper describes a data-driven method to solve a mean field game model of population evolution, by proving a connection between Mean Field Games with Markov Decision Process and building on methods in reinforcement learning. This method is scalable to arbitrarily large populations because the Mean Field Games framework represents population density rather than individual agents. With experiments on real data, Mean Field Games emerges as a powerful framework for learning a reward and policy that can predict trajectories of a real-world population more accurately than alternatives. Reviewer feedback summary Overall Score: 26/30 Average Score: 8.66 The reviewers are unanimous in finding the work in this paper highly novel and significant. According to the reviewers, there is still minimal work at the intersection of machine learning and collective behavior, and this paper could help to stimulate the growth of that intersection. On the flip side, surprisingly, the paper was criticized with the statement “scientific content of the work has critical conceptual flaws”. However, the author refutations persuaded the reviewers that the concerns were largely addressed.

Oracle Database 18c, spaCy v2.0.8 released, Apache Phoenix 5.0.0-alpha released, and more in today’s top stories around machine learning, deep learning, and data science news. 1. Oracle Database 18c: Now available on the Oracle Cloud and Oracle Engineered Systems Oracle announced that its Oracle Database 18c is now available on the Oracle Cloud and Oracle Engineered systems. The 18c is the first annual release in Oracle’s new database software release model, and is a core component of Oracle's recently announced Autonomous Database Cloud. The 18c does not contain any seismic changes in functionality but there are lots of incremental improvements, which include: High performance analytics High availability Multitenant improvements Application Development Big Data and Data Warehousing Security improvements To read more on the new features of Oracle Database 18c in detail, visit the documentation blog. 2. spaCy v2.0.8 released spaCy is an open-source software library for carrying out advanced Natural Language Processing. Its v2.0.8 has been released with some new features, performance improvements and minor bug fixes. The features and performance improvements are: NEW: Lexical attribute IS_CURRENCY via Token.is_currency for currency symbols. Add noun_chunks syntax iterator for Norwegian. Add get_beam_parse method in ArcEager. Revert changes to the Matcher in favour of the new and improved API (#1971) coming in v2.1.0. Fixes various typos and inconsistencies For bug fixes and other detailed information, visit the GitHub Repo. 3. Turi Create 4.1.1 released Turi Create is designed to simplify the development of custom machine learning models. The last major release of Turi Create was on December, 2017, which was Turi Create 4.0: the initial open source release by Apple. Version 4.1.1 of Turi Create is released, which includes two fixes: import turicreate fails on macOS 10.12.6 (#256) Miscellaneous documentation consistency fixes For source code and other information visit the GitHub repository. 4. Apache Phoenix 5.0.0-alpha released Apache Phoenix is an open source, parallel, relational database engine that supports OLTP for Hadoop using the Apache HBase as the backing store. It enables OLTP and operational analytics in Hadoop for low latency applications. The Apache Phoenix 5.0.0 is an alpha release. This release is the first version of Phoenix which is compatible with Apache Hadoop 3.0.x and Apache HBase 2.0.x. Known issues: The Apache Hive integration is known to be non-functional (PHOENIX-4423) Split/Merge logic with Phoenix local indexes are broken (PHOENIX-4440) Apache Tepha integration/transactional tables are non-functional (PHOENIX-4580) Point-in-time queries and tools that look at “old” cells are broken, e.g. IndexScrutiny (PHOENIX-4378) Developers encourage users to test this release out and report any observed issues for the official 5.0.0 release quality to be significantly improved. 5. Cloudera Enterprise 5.14 released Cloudera released its Cloudera Enterprise 5.14. This is a maintenance release and has fixed two issues, which were: Cloudera Manager upgrade workflow incorrectly requires deploying some optional management roles Logging issue slows down Hive and HDFS Replication jobs For further details, visit the Cloudera documentation page. 6. Eggplant AI 2.0:  Machine Learning brought to Software Testing Testplant released a new version 2.0 of its Eggplant AI. This version uses AI, machine learning, and analytics to intelligently navigate applications, predict quality issues, and correlate data, which can help product teams quickly identify and resolve issues. Eggplant AI 2.0 Highlights: Uses AI and neural networks to auto-generate tests and focus test execution on the user journeys most likely to find defects Enables software and app vendors to keep up with the pace of DevOps and user expectations Helps improve the user experience

Facebook introduces Tensor comprehensions, IBM adds AI to Video Cloud, Microsoft’s VS now includes Anaconda and more in today’s top stories around machine learning, deep learning, and data science news. 1. Facebook Announces Tensor Comprehensions Facebook AI Research (FAIR) announced the release of Tensor Comprehensions, a C++ library and mathematical language for running large-scale models on various hardware backends. It features a Just-In-Time compilation to produce automatic, on-demand, and high-performance codes for machine learning needs. The release includes: a mathematical notation to express a broad family of ML ideas in a simple syntax a C++ frontend for this mathematical notation based on Halide IR a polyhedral Just-in-Time (JIT) compiler based on Integer Set Library (ISL) a multi-threaded, multi-GPU autotuner based on evolutionary search For further details, you may visit the facebook research blog. 2. IBM adds AI Capabilities to IBM Video Cloud IBM Cloud Video unveiled new AI-powered Automated Watson Caption Support and Speech-to-Text capabilities to its enterprise video offering. These new AI capabilities will help in recognizing speech within videos and convert spoken words and phrases into text for video captions. Here’s how they would be helpful. Automatic transcript generation and real-time processing will slash editing workflows and costs. The advanced search and discovery features will help in optimizing employee engagement through. Increased accessibility and compliance will make content more digestible for all team members. To know more, read the official press release published online. 3. Microsoft’s Visual Studio Code is now included in the Anaconda distribution Microsoft’s Visual Studio Code will now ship as part of the popular Python data science platform Anaconda. According to Microsoft, “Visual Studio Code can easily be installed at the same time as Anaconda, providing a great editing and debugging experience for Python users, with special features tailor-made for Anaconda users.” Microsoft has previously made investments in the Python community with Python extension for VS Code and support for Python in Azure Machine Learning, SQL Server, and Azure Notebooks. According to the Anaconda team, “VS Code is a good IDE choice for its users on Windows, macOS, and Linux because of its debugging, code completion, and Git integration features.” It also offers a number of extensions that developers can tailor to their specific needs. 4. MongoDB announces support for multi-document ACID transactions in version 4.0 MongoDB has announced that it will support multi-document ACID transactions in its 4.0 release. With this release, MongoDB will now have the power of NoSQL and cross-collection ACID transaction support. This combination will make it easy for developers to write mission-critical applications leveraging the power of MongoDB.  ACID (Atomicity, Consistency, Isolation, Durability) describes the ability to guarantee that a transaction is valid, which is difficult when data is distributed across multiple documents. With these multi-document transactions, MongoDB will now provide a globally consistent view of data across replica sets and enforce all-or-nothing execution to maintain data integrity. 5. Cloudant 2.8.0 is now released Cloudant, the cloud-based service based on the Apache-backed CouchDB project, has released their version 2.8.0. The changes include: Added support for /_search_disk_size endpoint which retrieves disk size information for a specific search index. Updated default IBM Cloud Identity and Access Management token URL. Removed broken source and target parameters that constantly threw AttributeError when creating a replication document. The entire changes are available at the GitHub repo.

TensorFlow 1.6.0-rc1, node-oracledb 2.1 on npm, eXist-db v4.0.0 released, and more in today’s top stories around machine learning, deep learning,and data science news. 1. TensorFlow 1.6.0-rc1 has been released Tensorflow 1.6.0 release candidate 1 has been released with two breaking changes Prebuilt binaries are now built against CUDA 9.0 and cuDNN 7. Prebuilt binaries will use AVX instructions. This may break TensorFlow on older CPUs. There are also some major features and improvements: New Optimizer internal API for non-slot variables tf.estimator.{FinalExporter,LatestExporter} now export stripped SavedModels, which improves forward compatibility of the SavedModel. FFT support added to XLA CPU/GPU. Android TF can now be built with CUDA acceleration on compatible Tegra devices In addition to these, there are also some bug fixes and other changes such as updates in the documentation and the Google Cloud Storage(GCS) which you can read at the GitHub repo. 2. Oracle announces ‘Oracle Enterprise Data Management Cloud’ Oracle extends its Enterprise Performance Management (EPM) suite by announcing Oracle Enterprise Data Management Cloud at its Modern Finance Experience 2018, in New York. Hari Sankar, Oracle’s group vice president, EPM product management, said, “ The new offering will help customers manage important metadata structures related to their financial applications to avoid misalignment and lack of consistency.” Some benefits of the Oracle Enterprise Data Management Cloud include: Faster cloud adoption, which allows migration and mapping of enterprise data elements and on-going changes across public, private and hybrid cloud environments from Oracle or third parties. Improvised business agility allows faster business transformation through modeling M&A scenarios, reorganizations and restructuring, chart of accounts standardization and redesign. Better alignment of enterprise applications, which can manage on-going changes across front-office, back-office and performance management applications through self-service enterprise data maintenance, sharing and rationalization. System of reference for all your enterprise data provides a support enterprise data across business domains including: master data, reference data, dimensions, hierarchies, business taxonomies, associated relationships, mappings and attributes across diverse business contexts. 3. Oracle’s node-oracledb 2.1 is now available from npm Oracle announced that its Node-oracledb 2.1.0, the Node.js module for accessing Oracle Database, is now available on npm. The top features of this release include: Support for SYSDBA, SYSOPER, SYSASM, SYSBACKUP, SYSDG, SYSKM, and SYSRAC privileges in standalone connections. A new 'queryStream()' Stream 'destroy()' method Improvement in the Error object with new 'errorNum' and 'offset' properties Addition of new 'versionSuffix' and 'versionString' properties to the oracledb object to aid showing the release status and version. Node-oracledb 2.1 no longer compiles with the long-obsolete Node 0.10 or 0.12 versions. See the Change Log for complete changes in the node-oracledb 2.1 4. Oracle brings industry 4.0 capabilities to its IoT Cloud Oracle announced addition of new capabilities for its Oracle IoT Cloud applications. Oracle would be adding them to applications including Asset Monitoring, Production Monitoring, Fleet Monitoring, Connected Worker, and Service Monitoring for Connected Assets. The Industry 4.0 capabilities include: Digital Twin Augmented Reality Machine Vision Auto Data Science The advanced monitoring and analytics capabilities of these new offering allows organizations to improve efficiency, reduce costs, and identify new sources of revenue through advanced tracking of assets, workers, and vehicles, real-time issue detection, and predictive analytics. To read about these new offerings in detail, visit Oracle’s official press release. 5. eXist-db v4.0.0 released This is a major release of the eXist-db v4.0.0. The release contains API changes, several new features and bug fixes. New added features include: Addition of fn:unparsed-text, fn:unparsed-text-lines and fn:unparsed-text-available functions. Implementation of the HTML ASCII Case Insensitive Collation for XPath 3.1. Replacement of ASCIIFoldingFilter with ICUFoldingFilter in NoDiacriticsAnalyzer for better language search support. New User Manager application shipped for the Dashboard. Updated Cache Extension Module,: Implements an LRU policy with both TTL and size options. Includes new functions: cache:names(), cache:keys($name), and cache:destroy($name). Scheduled task option unschedule-on-exception is now exposed in conf.xml. Each thread that eXist creates is now explicitly named for easier identification. Bash Scripts now use /bin/env to locate bash. Updated third-party dependencies See Release notes for API changes, bug fixes, and other performance improvements.