Tech News - Data

Microsoft has announced multiple updates to its Microsoft Azure Cloud Platform today. These updates are meant to help companies migrate database workloads to its data centers and making it easier to run them in Azure. SQL Server customers can now try the preview for SQL Database Managed Instance, Azure Hybrid Benefit for SQL Server, and Azure Database Migration Service preview for Managed Instance. Additionally, Microsoft has also announced the preview for Apache Tomcat® support in Azure App Service and the general availability of Azure Database for MySQL and PostgreSQL in the coming weeks, making it easier to bring open source powered applications to Azure. Microsoft SQL Database Managed Instance Azure SQL Database Managed Instance allows seamless movement of any SQL Server application to Azure without application changes. Managed Instance offers full engine compatibility with existing SQL Server deployments including capabilities like SQLAgent, DBMail, and Change Data Capture, to name a few. Microsoft Azure Database Migration Service The Azure Database Migration Service is designed as an end-to-end solution to help customers moving databases from on-premises SQL Server instances to SQL Database Managed Instances. Microsoft Azure Hybrid Benefit program With the Azure Hybrid Benefit program customers can now move their on-premises SQL Server licenses with active Software Assurance to Managed Instance and soon the SQL Server Integration Services licenses to Azure Data Factory with upto 30% discounted pricing. Apache Tomcat® support in Microsoft Azure App Service Microsoft also announced a preview of built-in support for Apache Tomcat and OpenJDK 8 from Azure App Service. This will help Java developers easily deploy web applications and APIs to Azure’s market leading PaaS. Once deployed, customers can then extend it with the Azure SDK for Java to work with various Azure services such as Storage, Azure Database for MySQL, and Azure Database for PostgreSQL.  General availability of Azure database services for MySQL and PostgreSQL Azure Database Services for MySQL and PostgreSQL provide customers with fully managed homes for their open source databases in Microsoft’s cloud. These reduce a company's time spent in managing things like database scaling and patching. SQL Information Protection Preview SQL Information Protection lets organizations discover, classify, label and protect potentially sensitive data that's stored in a database management system, either in Microsoft's cloud or in an organization's datacenters. This service can be used with the Azure SQL Database service or with SQL Server on premises. More information about these updates is available on the Microsoft Azure blog.

Snips, open source Snips NLU, a Natural Language Understanding python library that allows parsing sentences written in natural language, and extract structured information. Snips NLU is a Python library that can be used to easily train models to make predictions on new queries.  Snips have also open sourced Snips NLU-rs, a Rust implementation focused on the prediction aspect. Snip NLU-rs consists of a traditional flat model called a linear chain Conditional Random Field (CRF), instead of CNNs or bi-LSTMs. The Snips team has replaced heavy word embeddings with a carefully crafted set of features that capture semantic and structural signals from the sentence. The Snips NLU-rs inference engine can run literally anywhere, from a 5$ Raspberry Pi Zero to an AWS EC2 free-tier instance. This library can be used on most modern architectures: on small connected devices, on mobile, on a desktop, or on a server. It can currently handle 5 languages (English, French, German, Spanish and Korean), with more to be added regularly. Unlike other chatbots and voice assistants that mostly rely on cloud services for their NLU, Snips NLU can run on the Edge or on a server.  Moreover, the platform is the first ‘Private by Design’ alternative to traditional voice assistants. This means user data is not touched, processed or collected, unlike most voice assistants. Researchers at Snip compared their NLU engine with other leading voice assistants/chatbots including API.ai (now DialogFlow, Google), Wit.ai (Facebook), Luis.ai (Microsoft), and Amazon Alexa by training them all using the same dataset, and testing them on the same out-sample test set. Experimental results showed that Snips NLU is as accurate or better than other cloud solutions at slot extraction tasks, regardless of how much training data was used. If you want to know more, check out the Github repository. To start building your own Snip NLU assistant go on to the Snips console.

Microsoft unveils plans to introduce more artificial intelligence and machine learning capabilities inside Windows 10. The next major Windows 10 update will now include a new AI platform, Windows ML. The new platform will enable developers to build machine-learning models, trained in Azure, right into their apps using Visual Studio and run them on their PCs. Here are a few noteworthy features: Windows ML has an abstraction layer at its core that can automatically optimize an application’s ML model for the underlying hardware. It adapts itself to every machine. So for example, if your computer includes a graphics card that supports Microsoft’s DirectX framework, Windows ML can use the software’s performance boosting features to enhance response times. On a less sophisticated machine, it might simply run AI models on the CPU. Developers can also import existing learning models from different AI platforms and run them locally on PCs and devices running on Windows 10. Microsoft researchers point out 3 benefits of using the Windows ML AI platform: Low latency, real-time results. Windows can perform AI evaluation tasks using the local processing capabilities of the PC, enabling real-time analysis of large local data such as images and video. Reduced operational costs. Developers can build affordable, end-to-end AI solutions that combine training models in Azure with deployment to Windows devices for evaluation. Flexibility. Developers can choose to perform AI tasks on the device or in the cloud based on what their customers and scenarios need. Microsoft also plans to provide support for specialized chips to power AI software. As part of the effort, the company is collaborating with Intel Corp. to make Windows ML compatible with its Movidius vision processing units. Developers can get an early look at the AI platform on Windows with Visual Studio Preview 15.7.  For all others, the Windows ML API in standard desktops apps and Universal Windows Apps will be available across all editions of Windows 10 this year. To read about all release features, have a look at the official Windows blog.

Reptile, developed by Open AI, is a simple meta-learning algorithm. Meta-learning is the process of learning how to learn. A meta-learning algorithm takes in a distribution of tasks, where each task is a learning problem, and it produces a quick learner. This means a learner must be able to generalize from a small number of examples. An example of a meta-learning problem is few-shot classification. Here, each task is a classification problem within which the learner after seeing only 1 - 5 input-output examples from each class must classify new inputs. What Reptile does It samples a task repeatedly, performs stochastic gradient descent on it, and finally updates the initial parameters towards the final parameters learned on the task. Any Comparisons? Reptile performs as well as MAML, which is also a broadly applicable meta-learning algorithm. Unlike MAML, Reptile is simple to implement and more computationally efficient. Some features of Reptile : Reptile seeks an initialization for the parameters of a neural network, such that the network can be fine-tuned using a small amount of data from a new task. Unlike MAML, Reptile simply performs stochastic gradient descent (SGD) on each task in a standard way. This means it does not unroll a computation graph or calculate any second derivatives. Hence, Reptile takes less computation and memory than MAML. The current Reptile implementation uses TensorFlow for the computations involved, and includes code for replicating the experiments on Omniglot and Mini-ImageNet. To Read more on how Reptile works, visit the OpenAI blog. To view Reptile implementations, visit its GitHub Repository.  

The MongoDB Go driver team has announced the second alpha release of the official Go driver. The official MongoDB Driver API for Go has certain changes as a part of the second alpha release. This release mainly contains improvements to the user experience and bug fixes. A point to be noted here is that this is an alpha software, so it is not recommended for production use. The MongoDB Go Driver team said, “following semantic versioning, the v0 version of the public API should not be considered stable and could change.” Changes since the prior release include: New Features: Examples for sample shell commands created Marshal, Unmarshal, and UnmarshalDocument functions added to BSON library Stringer for objectid.ObjectID implemented Improvements: New BSON library is tested against BSON corpus ReplaceOptions replaced from UpdateOptions CRUD tests resynced to update insertMany test format to a map Namespace type added for options in mongo package. DecodeBytes method added to the Cursor A method added to bson.Value to get the offset into the underlying []byte mongo.Cursor is made its own interface Document.ElementAt usability improved bson.ArrayConstructor renamed to bson.ValueConstructor FromHex function added to the objectid package Bugs: Lookup should properly traverse Arrays Documentation for the bson package wrt the builder.Builder type needs to be clarified Ensure methods of *Document handle the case where *Document is nil Update bson.ErrTooSmall bson.Reader.Lookup should return ErrElementNotFound if no element is found The official documentation is available on GoDoc. Questions and inquiries can be asked on the mongo-go-driver Google Group.

Open AI’s Reptile meta-learning algorithm, MongoDB Go driver Alpha 2 release, Microsoft’s built-in AI for Windows 10, and more in today’s top stories around machine learning, deep learning, and data science news. Top Data science News Stories of the Day Reptile: Open AI's scalable meta-learning Algorithm MongoDB Go Driver Alpha 2 released! Windows ML: Microsoft’s planned built-in AI platform for developers in Windows 10 Other Data Science News at a Glance IBM Research has launched PAIRS (Physical Analytics Integrated Repository and Services) Geoscope, a cloud analytics service to connect apps with a range of big geospatial datasets, covering maps, satellite, weather, and population changes. This service is available for developers to use IBM's REST API to add geospatial and time-based data to their own apps.      Read more on ZDNet. 2. Microsoft and Esri offer the GeoAI Data Science Virtual Machine (DSVM) as part of their Data Science          Virtual Machine/Deep Learning Virtual Machine family of products on Azure.     Read more on Microsoft Azure Blog. 3. Hitachi Vantara announces additional capabilities for machine learning orchestration to help data                scientists monitor, test, retrain and redeploy supervised models in production.     Read more on Dataquest. 4. AtScale Inc. today updated its business intelligence abstraction platform with an added support for          data lakes of any size and simpler migration of analytics workloads across business intelligence tools.     Read more at SiliconAngle. 5. Power BI Desktop March Feature Summary is here. Features include, making SAP HANA and several            popular connectors generally available. Bookmarking is also now generally available to create bookmarks            from scratch in the Power BI Service.     Read more on Microsoft Blog. 6. Instagram engineering team announces that it is open sourcing Rocksandra, an Apache Cassandra            storage engine built on RocksDB, a persistent key-value store for fast storage.     Read more on Medium.

Alteryx, one of the leading providers of self-service analytics have released Alteryx Analytics 2018.1.  This collaborative platform helps data scientists and business analysts to discover, prepare, and blend from more data sources, and easily operationalize models. Let’s have a quick sneak-peek at the features of the new 2018.1 platform: Collaborative Insights: These insights will help in gaining quick access to the right data at the right time in a governed manner. Using the Alteryx Connect Loaders one can directly access metadata stored in DB2, HDFS, and SAP Hana. One can also evaluate and display analytic assets. The option to find, view and launch assets stored in Alteryx Connect directly from Alteryx Designer using the global search is possible using these insights. One can also establish Alteryx Connect lineage from Designer workflows that use In-Database processes/tools. Analytic Flexibility: The 2018.1 platform can be utilized for harnessing the full value of one’s existing architecture and emerging data assets. The user can experience expanded data connections i.e., new connectors for AWS Athena, Redshift spectrum, and an enhanced integration with Excel. This new version has a new Tableau support, which outputs directly from an Alteryx workflow into Tableau Hyper. One can execute code from R, Python or Scala directly against the Spark cluster with the new code tool for Apache Spark. Generating fast and accurate suggestions, error notifications, and auto-completion of expressions are all now possible. Operationalize Models : One can easily deploy predictive models (built in Alteryx, R, or Python) into production with Alteryx Promote.  Also, this new version allows managing models from development to production to ensure they deliver the best impact on the business. It also aids in monitoring model performance and health in order to understand if the model need to be retrained or discarded. Read more about this topic in detail on the Alteryx Community Blog.  

Two of the leading open-source projects, Apache Spark and Kubernetes now collaborate: Apache Spark 2.3 has native Kubernetes support. Kubernetes : A natural fit for Apache Spark Apache Spark is a framework for large-scale data processing and an important tool for data scientists. It offers a robust platform to carry out major tasks; be it data transformation, analytics, or machine learning. Recently, data scientists have been embracing the concept of working with containers in order to improve their workflows. Benefits such as packaging of dependencies and creating reproducible artifacts can be leveraged by the container adoption. This is where Kubernetes, an open-source system for automating deployment, to scale and manage containerized environments, comes to the rescue. It enables one to run containerized applications within Spark. This combination of Apache Spark and Kubernetes has dual benefits. Firstly, data scientists get to use their principal tool i.e., Apache Spark’s ability to manage distributed data processing tasks and secondly, they can work with containers using Kubernetes API. With Apache Spark 2.3, users can run Spark workloads in an existing Kubernetes 1.7+ cluster. This means Apache Spark workloads can make direct use of Kubernetes clusters for multi-tenancy and sharing through Namespaces and Quotas. It can also make use of administrative features such as Pluggable Authorization and Logging. Also, Spark workloads require no changes or new installations on the Kubernetes cluster. One simply has to create a container image and set up the right RBAC roles for the Spark Application and it is ready. The native Kubernetes support offers a fine-grained management of Spark Applications along with improved elasticity, and seamless integration with logging and monitoring solutions. The community is also exploring advanced use cases such as managing streaming workloads and leveraging service meshes like Istio. Visit Databricks blog to read more on this topic.

In an attempt to deepen neural network interpretability, Google has released Google Lucid, a neural network visualization library along with publishing an article “The Building Blocks of Interpretability”, which answers one of the most popular questions in Deep Learning: how do neural networks make decisions? Google Lucid is a neural network visualization library building off Google’s work on DeepDream. You may remember DeepDream as Google’s earlier attempt to visualize how neural networks understand images, which led to the creation of psychedelic images. Google Lucid adds feature visualizations to create more artistic DeepDream images. It is basically a collection of infrastructure and tools for research in neural network interpretability. In particular, it provides state of the art implementations of feature visualization techniques, and flexible abstractions that make it very easy to explore new research directions. To add more flexibility and ease of work, Google is also releasing colab notebooks. These notebooks make it extremely easy to use Lucid to reproduce visualizations. Just open the notebook and click a button to run code without worrying about setup requirements. To further make things exciting, Google’s new Distill article, titled, “The Building Blocks of Interpretability,” shows how feature visualization in combination with other interpretability techniques allows a clear cut view of the neural network. This is helpful to see how a neural network makes some decisions at a point, and how they influence the final output. For example, Google says, “we can see things like how a network detects a floppy ear, and then that increases the probability it gives to the image being a “Labrador retriever” or “beagle”. The article explores techniques for understanding which neurons fire in the network by attaching visualizations to each neuron, almost a kind of MRI for neural networks. It can also zoom out and show how the entire image was perceived at different layers. Thus detecting very simple combinations of edges, to rich textures and 3d structure, to high-level structures. The purpose of this research, Google says is to “address one of the most exciting questions in Deep Learning: how do neural networks do what they do?” However, it adds, “This work only scratches the surface of the kind of interfaces that we think it’s possible to build for understanding neural networks. We’re excited to see what the community will do.” You can read the entire article on Distill.

Google’s Lucid for ML interpretability, Native Kubernetes support in Apache Spark 2.3, Cloudera Altus with SDX, and more in today’s top stories around machine learning, deep learning, and data science news. Top Data science News Stories of the Day Google strides forward in deep learning: open sources Lucid to answer how neural networks make decisions. Apache Spark 2.3 now has native Kubernetes support. Alteryx Analytics 2018.1 is here: The analytics platform for enterprises. Other Data Science News at a Glance 1. Cloudera announces Cloudera Altus with SDX, the first machine learning and analytics Platform-as-a-Service (PaaS), for simplifying Multi-function Big Data Analytics. Read more on PR Newswire. 2. Prisma Cloud , a new GraphQL Database Platform launched. It offers powerful data workflows like exploring and editing data in an intuitive data browser as well as automatic rollbacks. Further features include team collaboration, performance metrics, health checks and more. Read more on the Graphcool blog. 3. Pentagon uses AI to analyze drone footage. Google is reportedly working on a pilot project with the US Defense Department to use TensorFlow APIs to assist in object recognition on unclassified data. Read more on ZDNet. 4. CRAN gets a new addition of textfeatures, a simple package for extracting useful features from character objects. Read more on CRAN-project. 5. Allscripts Healthcare Solutions unveiled Avenel, its new electronic health record, which uses machine learning to reduce time for clinical documentation and is designed to work like an app. Read more on Nasdaq.

Microsoft Cloud AI Research Challenge, Neblio - a next-generation blockchain network, Salesforce Einstein Analytics’ ‘Conversational Queries’, and more in today’s top stories around machine learning, deep learning, and data science news. Top Data science News Stories of the Day Google Bristlecone: A New Quantum processor by Google’s Quantum AI lab Google-Landmarks, a novel dataset for instance-level image recognition Pandas on Ray: Make Pandas faster by replacing one line of your code Other Data Science News at a Glance 1. The Microsoft Cloud AI Research Challenge invites any researcher—from students to academics to employees of public and private organizations—to build AI applications on Microsoft AI services, and the two best will be awarded USD25,000. Read more at Microsoft Research. 2. Neblio, a next-generation blockchain network that aims to make enterprise integration seamless, simple, and cost-effective. It offers a suite of solutions that are intended to streamline the process of integrating blockchain technology in a simple and efficient manner. Read more at CryptoSlate. 3. Big-data company HVR Software B.V. today launched its real-time data integration platform. The new architecture provides a more efficient method moves data continuously using a log-based change data capture method, which is a low-impact way of moving data from a variety of sources into target systems. Read more at SiliconANGLE. 4. Google open sources a protocol buffer implementation of the Fast Healthcare Interoperability Resources (FHIR) standard to make healthcare data work better with machine learning. To enable large-scale machine learning, the protocol buffer have a few additions such as, implementations in various programming languages, an efficient way to serialize large amounts of data to disk, and a representation that allows analyses of large datasets. Read more at Google Research. 5. MXNet is now faster and more scalable with the 1.1.0 release. With this release, MXNet makes it easier for developers to build vocabulary and load pre-trained word embeddings by adding experimental API.It also includes improved batching for GEMM/TRSM operators with large matrices on GPU makes it faster for you to train models. Read more at  The Apache Blog. 6. Paxata announced the general availability of Spring ’18, the next major release of the company’s award-winning Adaptive Information Platform. The latest offering significantly accelerates how business consumers prepare enterprise data volumes at speed and creates high-quality information for analysis and collaboration across global organizations with new enhancements that include one-click profiling, rapid data onboarding, and multi-tenancy capabilities. Read more at Paxata Press releases. 7. Axoni announces AxLang, a new programming language based on Scala that supports functional programming and enables formal verification of smart contracts for Ethereum-compatible networks. Read more on Medium. 8. Salesforce introduced a new feature to Einstein Analytics today called ‘Conversational Queries’. With Conversational Queries, users can type phrases related to their data — such as “show me top accounts by annual revenue” or “rank accounts decreasing by annual revenue and billing country” — and instantly view answers in automatically configured dynamic charts. Read more on  Techcrunch.

Pandas on Ray is the latest development in the Ray framework. It is a DataFrame library that wraps Pandas and provides a transparent distribution of data and computation. 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. It accelerates Pandas queries by 4 times on an 8-core machine. This requires users to change just a single line of code in their notebooks. Ray: A machine learning substitute for Apache Spark Developed by two Ph.D.students, Philipp Moritz and Robert Nishihara, at the RISELab, Ray is a a distributed execution framework for AI applications and also a potential project to replace Apache Spark. RISELab is the successor to the U.C.Berkeley group, which created Apache Spark. Apache Spark was designed to be faster than its forerunner, MapReduce, but still faced issues with design decisions which made it difficult to write applications that included Complex task dependencies. This was mainly because of Spark’s internal synchronization mechanisms. Ray was designed to provide better speeds than Apache Spark. Ray, is designed to provide better speeds than even Apache Spark. It is written in C++ and aims at accelerating the execution of machine learning algorithms developed in Python. It makes use of an immutable object model--any objects that can be made immutable don’t need to be synchronized across the cluster--which save a lot of time. Also, Ray maintains a state of computation among various other nodes in the cluster, which in turn maximizes robustness. Additional features include: Ray can handle heterogeneous hardware (where some application workload is being executed on CPUs and others on GPUs) as it has a number of schedulers that can bring both CPUs and GPUs together. It can also borrow task-dependency attributes from MPI, the low-level distributed programming environment. Ray is also useful for building an array of applications that require fast decision-making on real-world data such as what’s required for autonomous driving and so on. Pandas on Ray On comparing Pandas with Pandas on Ray, following results were obtained: Pandas on Ray: 100 loops, best of 3: 4.14 ms per loop Pandas: The slowest run took 32.21 times longer than the fastest. This could mean that an intermediate result is being cached. 1 loop, best of 3: 17.3 ms per loop This concluded, Pandas on Ray is about 4 times faster than Pandas. This was run on a machine with eight cores, so the speedup isn't perfect because of the overheads. Here, no special optimizations were done for Pandas on Ray; only the default settings were used in this experimentation. Also, Ray uses Eager execution and thus one cannot have query planning or have advanced knowledge of the best way to compute a given workflow. To know more about Ray in detail, visit its GitHub repository. Also, to more about Pandas on Ray at the RISELab blog.

The quest to conquer the Quantum World is rapidly advancing! Another contender in this conquest is Google, who has launched the preview of Bristlecone, a new Quantum Processor. Google’s Bristlecone was unveiled at the annual American Physical Society meeting in Los Angeles on March 5, 2018. According to Google, “Bristlecone would be a compelling proof-of-principle for building larger scale quantum computers.” The purpose of this quantum processor is to provide a testbed for research into system error rates and scalability of Google’s qubit technology along with applications in quantum simulation, optimization, and machine learning. A Preview of Bristlecone, Google’s New Quantum Processor. On the right, is a cartoon of the device: each “X” represents a qubit, with nearest neighbor connectivity. Google Bristlecone uses a new architecture that allows 72 quantum bits on a single array with an overlapping design that puts two different grids together. Google has optimized Bristlecone for the lowest possible error rate using a specialized process called Quantum Error Correction. The previous 9-qubit linear quantum computers by Google demonstrated error rates of 1% readout, 0.1% single-qubit gates and 0.6% two-qubit gates. Google Bristlecone uses the same scheme for coupling, control, and readout, but is scaled to a square array of 72 qubits. Google researchers chose a device of this size to demonstrate quantum supremacy in the future, to investigate first and second order error-correction using the surface code, and to facilitate quantum algorithm development on actual hardware. The intended research direction of the Quantum AI Lab is to access near-term applications on the road to building an error corrected quantum computer. For this, Google says, “would require harmony between a full stack of technology ranging from software and control electronics to the processor itself. Getting this right requires careful systems engineering over several iterations.” More information about Google Bristlecone is available in the Google research blog.

Image retrieval and image recognition are fundamental problems in the machine learning and computer vision world. Image classification technology has shown remarkable progress over the past few years. An obstacle in this research, however, is the unavailability of large annotated datasets. Google has made an attempt to solve this challenge by introducing Google-Landmarks, a worldwide dataset for recognition of human-made and natural landmarks. This dataset was made with the intention of solving fine-grained and instance-level recognition problems. Examples of this include identifying important landmarks in images (Eiffel Tower, Mount Fuji, Taj Mahal, etc), which accounts for a large portion of what people like to photograph. Landmark recognition can help predict landmark labels directly from image pixels to help people better understand and organize their photo collections. The Google-Landmarks dataset contains more than 2 million images depicting 30 thousand unique landmarks from across the world, a number of classes that is almost 30x larger than what is available in commonly used datasets. Geographic distribution of landmarks in the Landmark dataset Google has also open-sourced Deep Local Features DELF, an attentive local feature descriptor, which is useful for large-scale instance-level image recognition, in order to advance research in this area. DELF detects and describes semantic local features which can be geometrically verified between images showing the same object instance. It is also optimized for landmark recognition. Google-Landmarks is being released as part of the Landmark Recognition and Landmark Retrieval Kaggle challenges. The Landmark recognition challenge calls for developers to build models that recognize the correct landmark (if any) in a dataset of challenging test images. In the retrieval challenge, developers are given query images and for each query, they are expected to retrieve all database images containing the same landmarks (if any). Participants are encouraged to compete in both these challenges as the test set for both the problems is same. Participants may also use the training data from the recognition challenge to train models which could be useful for the retrieval challenge. However, there are no landmarks in common between the training/index sets of the two challenges. This challenge is the focal point of the CVPR’18 Landmarks workshop. More details of the challenge and the dataset can be found in the Google research blog.  

After a sneak-peek into TensorFlow’s release candidates 1.6.0-rc0 and 1.6.0-rc1, its major release 1.6.0 is finally here! Tensorflow 1.6.0 includes two new breaking changes, feature improvements and bug fixes in its list. The previous version, TensorFlow 1.5, introduced us to jaw dropping inclusions such as TensorFlow Lite developer preview and TensorFlow Eager Execution.   Let’s have a look at what’s in store with the newly released TensorFlow version 1.6.0. The two most important changes include: The prebuilt binaries are now built against CUDA 9.0 and cuDNN 7 These prebuilt binaries would now use AVX instructions, which may break TensorFlow on older CPUs. List of major feature improvements: A new optimizer internal API for non-slot variables. tf.estimator.{FinalExporter,LatestExporter} can now export stripped SavedModels, which improves forward compatibility of the SavedModel. FFT support has been added to XLA CPU/GPU. Also, Android TF can now be built with CUDA acceleration on compatible Tegra devices. Few API changes in 1.6.0: Introducing prepare_variance boolean with default setting to False for backward compatibility. Move layers_dense_variational_impl.py to layers_dense_variational.py. Minor bug fixes include: Addition of a  client-side throttle in the Google Cloud Storage (GCS). Addition of a FlushCaches() method to the FileSystem interface, with an implementation for GcsFileSystem. In addition to these, TensorFlow 1.6.0 includes a second version of the Getting started guide exclusively for newcomers in Machine learning. Not only this, documentation for TPUs is a must-watch. It also includes certain other changes which you will be able to read at the GitHub version release page.