Published in


Google Introduces Flax: A Neural Network Library for JAX

In optimization theory, a loss or cost function measures the distance between the fitting or predicted values and real values. For the majority of machine learning models, improving performance means minimizing the loss function.

But for deep neural networks, performing gradient descent to minimize the loss function for every parameter can be prohibitively resource-consuming. Traditional approaches include manually deriving and coding, or implementing the neural model using syntactic and semantic constraints of a machine learning framework like TensorFlow.

But what if it were possible to simply write down loss functions using a NumPy library and have the work done automatically? That’s a job for JAX — the Just-in-time compiler Google introduced in 2018 that uses Autograd and XLA (Accelerated Linear Algebra) and can automatically differentiate native Python and NumPy code through a large subset of Python features such as ifs, loops, recursion and closures. JAX also allows for fast scientific computing by automatically parallelising code across multiple accelerators such as GPUs and TPUs.

Taking this one step further, Google recently introduce Flax — a neural network library for JAX that is designed for flexibility. Flax can train neural networks by forking an example from its official GitHub repository. When it comes to modifying models, developers need no longer add features to the framework, they can simply modify the training loop (such as train_step setting) to achieve the same result. At its core, Flax is built around parameterised functions called Modules, which override apply and can be used as normal functions.

Flax code used to define a learned linear transformation.

The Flax release has created a buzz on social media. Director of Machine Learning research at NVIDIA Anima Anandkumar tweeted the Flax GitHub link, adding: “We used CGD for training GANs and for constrained problems in RL. This library will be very useful.” Google Brain Research Scientist David Ha (twitter name hardmaru) also endorsed the new repository.

For those interested in trying Flax, there are currently three examples available for testing: MNIST, a database of handwritten digits that is mainly used as handwritten digits recognition task; ResNet, a deep residual learning architecture for image recognition that is trained in ImageNet and mostly used to measure large-scale cluster computing capability; And 1 Billion Word Language Model Benchmark, a standard training and test setup for language modeling experiments.

The Flax team is also calling on developers to help to build additional end-to-end examples, such as Translation, Semantic Segmentation, GAN , VAE etc.

The Google Research: Flax repository is on GitHub.

Author: Hecate He | Editor: Michael Sarazen

Thinking of contributing to Synced Review? Sharing My Research welcomes scholars to share their own research breakthroughs with global AI enthusiasts.

We know you don’t want to miss any story. Subscribe to our popular Synced Global AI Weekly to get weekly AI updates.

Need a comprehensive review of the past, present and future of modern AI research development? Trends of AI Technology Development Report is out!

2018 Fortune Global 500 Public Company AI Adaptivity Report is out!
Purchase a Kindle-formatted report on Amazon.
Apply for Insight Partner Program to get a complimentary full PDF report.



We produce professional, authoritative, and thought-provoking content relating to artificial intelligence, machine intelligence, emerging technologies and industrial insights.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store

AI Technology & Industry Review — | Newsletter: | Share My Research | Twitter: @Synced_Global