Machine Learning WG updates - Sep 2021

Welcome to another update from the Machine Learning WG! This time we are glad to announce many improvements to our on-going efforts and welcome new projects into the fold!

Nx updates

Nx is a multi-dimensional tensors library for Elixir with multi-staged compilation to the CPU/GPU.

• Refactored Nx’ core to support TPUs (https://github.com/elixir-nx/nx/pull/423)

• Many new index-based operations were added, such as Nx.take/3, Nx.take_along_axis/3, Nx.gather/2, and Nx.scatter_add/3

• The EXLA backend (which uses Google’s XLA) has been updated to Tensorflow 2.6 and now supports NVIDIA’s Jeton and macOS aarch64

• We now precompile Google’s XLA to multiple OSes and architectures. This brings the getting started experience down from several minutes to a couple seconds

Explorer updates

We are glad to welcome Explorer, led by Chris Grainger, into the Machine Learning WG and into the elixir-nx organization. Explorer is a dataframe library for Elixir. It is an API for data manipulation with high-level features such as:

• Simply typed series: float, integer, boolean, string, date, and datetime

• A powerful but constrained and opinionated API, so you spend less time looking for the right function and more time doing data manipulation

• Pluggable backends, providing a uniform API whether you’re working in-memory or (forthcoming) on remote databases or even Spark dataframes

• The first (and default) backend is based on NIF bindings to the blazing-fast polars library

Learn more about it in its README or in its introductory notebook.

Livebook updates

Livebook is a collaborative and interactive code notebook in Elixir developed by Jonatan Kłosko.

• Add an explore section with many introductory notebooks on Livebook, Elixir, VM instrumentation, etc (https://github.com/livebook-dev/livebook/pull/310)

• Introduce the Kino library: a Livebook client for building interactive charts, tables and more (https://github.com/livebook-dev/livebook/pull/306)

• Make notebooks more interactive with the addition of inputs (https://github.com/livebook-dev/livebook/pull/328), which can be made reactive (https://github.com/livebook-dev/livebook/pull/389), and allow for multiple input types (colors, text area, urls, passwords, etc)

• Introduce branching sections to perform asynchronous work (https://github.com/livebook-dev/livebook/pull/449)

• Support relative links within notebooks (https://github.com/livebook-dev/livebook/pull/441)

• Hovering modules and function calls in the Elixir code editor now shows their documentation (https://github.com/livebook-dev/livebook/pull/453)

• Allow notebooks to be exported to Elixir code (https://github.com/livebook-dev/livebook/pull/476) and Live Markdown, optionally with output (https://github.com/livebook-dev/livebook/pull/483)

• Support for pluggable file-systems with S3 support out of the box: (https://github.com/livebook-dev/livebook/pull/492)

Livebook has been moved to its own organization and it will launch their own website soon. The organization also hosts Nerves Livebook, led by Frank Hunleth, which provides a straightforward way to run Livebook on embedded devices and ships with many additions to teach Nerves to new developers through Livebook.

Axon updates

Axon is a library led by Sean Moriarity that brings deep learning to Elixir. It is powered by Nx and runs on the CPU/GPU.

• Axon now global pooling layers (https://github.com/elixir-nx/axon/pull/86)

• Axon handles mixed precision policies (https://github.com/elixir-nx/axon/pull/74)

• Axon supports callbacks (https://github.com/elixir-nx/axon/pull/71) and metrics (https://github.com/elixir-nx/axon/pull/67)

• Added constant, bilinear, and adaptive_lp_pool (adaptive power average pool) layers

Other bits

• Benjamin Moss wrote an article on Predicting fuel efficiency with Elixir, Nx, and Axon: a gentle introduction to Machine Learning

• Dmitry Rubinstein has implemented a Jupyter backend for Elixir during this year’s Spawnfest

• Robert Bates has shared a presentation about AI packages for Elixir currently available on Hex.pm

Are you also interested in Machine Learning and the Erlang Ecosystem? Join us and come chat on our Slack.