New code patterns from IBM, ShareInsights 2.0 a new analytics platform, NVIDIA’s CUDA Toolkit version 9.1, Google’s TFGAN, and more in today’s top stories in AI, machine learning and data science news.
Google open sources TFGAN, for easily training and evaluation of GANs
Google has open sourced TFGAN to make it easy to train and evaluate GANs. TFGAN is a lightweight library which provides an easy infrastructure to train a GAN. It also consists of well-tested loss and evaluation metrics and easy-to-use examples to highlight the flexibility of TFGAN. TFGAN provides simple function calls that cover the majority of GAN use-cases. TFGAN is lightweight, so developers can use it alongside other frameworks, or with native TensorFlow code. Developers can also select from a large number of already-implemented losses and features. The code is well-tested so numerical or statistical mistakes associated with GAN libraries are avoidable. Google has also released a tutorial that includes a high-level API to quickly get a model trained on user-provided data.
IBM releases 120 code patterns for building AI, blockchain, IoT and chatbots
IBM has released more than 120 code patterns for streamlining the development process in areas of IoT, AI, Blockchain, Analytics, and DevOps to name a few. These code patterns include packages of code, one-click GitHub repositories, documentation and other resources. With this move, IBM aims to help developers automate their tasks, giving them more time to innovate and build. In addition, IBM has also unveiled Bot Asset Exchange, an open source repository, for helping developers create easily deployable chatbots compatible with the Watson Conversation Service.
NVIDIA announces release of CUDA Toolkit version 9.1
CUDA 9 is one of the most powerful software platforms for GPU-accelerated applications. NVIDIA has added new algorithms and optimizations in its CUDA Toolkit version 9.1 for speeding up AI and HPC apps on Volta GPUs. It also includes compiler optimizations, support for new developer tool versions, and bug fixes. Using new functions in NVIDIA Performance Primitives, developers can now develop image augmentation algorithms for deep learning. They can also run batched neural machine translations and sequence modeling operations on Volta Tensor cores using new APIs in cuBLAS. The new release also has certain core optimizations to launch kernels up to 12x faster. Using new heuristics in cuFFT, developers can now solve large 2D and 3D FFT problems more efficiently on multi-GPU systems.
Accelerite launches ShareInsights 2.0, an end-to-end, self-service approach for big data analysis
ShareInsights 2.0 comes as a single, end-to-end, collaborative analytics platform for data preparation and online analytical processing and visualization from Accelerite, a cloud management vendor. The software runs natively atop a Hadoop cluster and uses existing Apache Spark instances for machine learning applications. It also boasts of more than 50 connectors and over 100 analytical widgets ranging from simple aggregation tasks to more complex machine learning. A drag-and-drop access to source data on the native cluster is available using the self-service data discovery feature. It even has its own visualization interface and also supports other data visualization tools such as Tableau.
Google reduces prices of its cloud machine learning offerings
Google backlashes at Amazon following the release of Amazon Sagemaker by slashing the prices of its own cloud machine learning service. According to a report by VentureBeat, customers using basic-tier compute for training a machine learning system will now pay 43 percent less than they did earlier. Google is also providing per-hour pricing for all of the different types of training machines available. The price reductions report was first included in a blog post that appeared briefly on Google’s website. This announcement came just a few weeks after Amazon released its Sagemaker, which is a rival to Google’s ML Engine service. In addition to the price cuts, the leaked blog also talked about Google’s plans to make its online prediction feature generally available.
Seattle plans for new machine learning apps and algorithms seeking inspiration from a Facebook Hackathon
A recent Facebook hackathon invited Seattle coders to solve the city’s civic tech problems using Machine Learning. Seeking inspiration from this hackathon, the open data leaders of the city decided to include machine learning as an integral part of the city’s open data program. The two winning projects at the hackathon were Find ‘n Park, an app that uses machine learning to help motorists find parking, and Contractor 5, a software that helps people find building contractors and estimate costs for new construction or remodeling. Following these two use cases, David Doyle, Seattle’s open data program manager, has encouraged all departments to think about leveraging machine learning opportunities in their current and future open datasets.