Jax's deep learning and scientific computing

         JAX frame ： Be able to use GPU Accelerated , Supporting automatic differentiation numpy.numpy It is very popular in the field of Scientific Computing , But in the field of deep learning .

         Jax yes Autograd and XLA To provide high-performance machine learning research . As Autograd Updated version of ,JAX Can automatically differentiate native Python and Numpy function . It can be done by loops、branches、recursion and closures, Differential , And be able to take the derivative of the derivative .

        2020 year ,DeepMind Declaration of use JAX To accelerate their research . More and more from Google brain （Google Brain） And other institutions' projects are also used JAX

        JAX It is a very promising project , And users have been growing steadily .JAX Already in-depth learning 、 robot / Control system 、 Bayesian method and scientific simulation have been widely used in many fields

AX Positioning scientific computing （Scientific Computing） And function conversion （Function Transformations） Cross fusion of , Have a range of capabilities beyond the training depth learning model , Includes the following ：

• Just in time compilation （Just-in-Time Compilation）
• Automatic parallelization （Automatic Parallelization）
• Automatic vectorization （Automatic Vectorization）
• Automatic differentiation （Automatic Differentiation）

JAX Faster than NumPy fast N An order of magnitude ,JAX It uses TPU,NumPy Used CPU, To emphasize JAX The upper speed limit is much higher than NumPy

JAX It includes an extensible system to realize such function conversion , There are four typical ways ：

        Grad() Perform automatic differentiation ;

        Vmap() Automatic vectorization ;

        Pmap() Parallel computing ;

        Jit() Convert functions to just in time versions

Scientific Computing

  Deep learning

Introduction to deep learning

Reference resources ：

Why You Should (or Shouldn't) be Using Google's JAX in 2022 (assemblyai.com)