A roundup of news about Artificial Intelligence, Machine Learning and Data Science. This is an eclectic collection of interesting blog posts, software announcements and data applications from Microsoft and elsewhere that I've noted recently.
Open Source AI, ML & Data Science News
Tensorflow 2.0.0 has been released. This major update makes many changes to improve simplicity and ease of use. Tensorflow 1.x models will need to be upgraded for Tensorflow 2.0.
Keras 2.3.0 has been released, the first release of the high-level deep learning framework to support Tensorflow 2.0.
spaCy v2.2 has been released, with retrained natural language models and a new data augmentation system.
greta: an R package to fit complex Bayesian models using Tensorflow as the optimization engine.
Julia 1.2.0 has been released. This minor release of the scientific computing language includes no breaking changes.
Industry News
Facebook and Microsoft launch a research program to detect deepfakes, and Google releases a library of deepfake videos for research.
Facebook open sources Hydra, a framework for elegantly configuring complex Python applications.
Hyperparameter tuning has been added to Facebook's fastText text classifier library.
Google contributes MLIR, the compiler framework for Tensorflow graphs, to the LLVM Foundation.
Google open-sources differential-privacy, a C++ library of algorithms to produce aggregate statistics from sensitive datasets.
Microsoft News
The Vision AI Developer Kit, a Qualcomm IoT camera coupled with an Azure toolkit for AI vision applications, is now generally available.
PyTorch 1.2 is now supported in Azure: Azure ML Service, Azure Notebooks and Data Science Virtual Machine.
Azure Video Indexer adds support for detecting animated characters and multilingual identification and transcription.
Jupyter Notebooks are now generally available in Azure Cosmos DB, supporting all data models including Cassandra, MongoDB, SQL, Gremlin and Spark.
Twelve Cognitive Services are now included in Azure Free, including Custom Vision, Text Analytics, Translator and Personalizer.
Machine Learning Notebooks for Azure ML Service, including a suite of examples of automated machine learning.
Visual Studio expands testing and debugging support for Python.
Python support in Azure Functions is now generally available.
Learning resources
Python for Beginners, a series of instructional videos and associated code samples from Microsoft.
Natural Language Processing Best Practices & Examples, a collection of Jupyter notebooks and utility functions for text classification, entity recognition and more, from Microsoft.
The CodeSearchNet Corpus, an open database of six million code samples released by Github, with the aim of improving semantic analysis of code and documentation.
Excavating AI, a philosophical essay on the issues related to image selection and labelling in training datasets for computer vision.
Video: How Backpropagation Works, part of Brendan Rohrer's End-to-End Machine Learning School series.
Recorded talks from the useR!2019 conference are available to view on the R Consortium channel.
A paper describes how AI agents may learn "superstitions" from reinforcement learning, recalling Skinner's superstitious pigeons.
Reference implementation for deploying ONNX models with Azure IoT edge, as demonstrated in this episode of the IoT Show.
Applications
Write with Transformer, a handy website from Hugging Face where you can generate text from powerful natural language models.
A new paper describes a method to generate a 3-D "Ken Burns effect" zoom from a single image.
An OpenAI project finds virtual agents learning unexpected strategies in a game of hide-and-seek.
CTRL: the Control Transformer Language model can be used to attribute the source of text, or generate synthetic text on a variety of themes. This new language model from Salesforce claims to be the largest released to date.
Startup Airdoc is using computer vision to detect chronic diseases including diabetes and hypertension from retinal scans of patients in China.
Fashion service Stitch Fix provides a machine-learning based service to recommend items that coordinate with a piece of clothing.
Find previous editions of the AI roundup here.
