NeurIPS 2019 Expo Workshop

Dec. 8, 2019

Expo 2019 Schedule »

Tensor Networks with TensorNetwork

Sponsor: Google X, The Moonshot Factory

Abstract:

Tensor networks are high-powered tools developed in physics and now finding new use in machine learning. Here we'll showcase TensorNetwork, a new open source library built on TensorFlow for doing tensor network computations. As tensor networks grow in popularity in the ML world, this library will be an invaluable resource for researchers and practitioners alike.

For more than fifteen years, physicists have been using tensor networks to study the properties of complex quantum systems. Fundamentally, these tensor networks are all about efficient storage and manipulation of correlations in a sparse representation of the data, built using nothing more complicated than matrix multiplication. The machine learning community is starting to catch on, and a growing community of researchers are exploring the applications of tensor networks to deep learning. A PhD in quantum physics is not required to understand these concepts.

Until now, there has not been a software library for tensor networks that has a low barrier to entry and is also powerful enough for state-of-the-art computations. That role is being filled by our new open-source library, TensorNetwork.

In this workshop, you'll:

  • Learn what a tensor network is.
  • See examples of a tensor network in action.
  • Get introduced to an easy-to-use open-source API for tensor networks using Python, with NumPy, TensorFlow, JAX, and PyTorch backends.
  • Leave excited and ready to use tensor networks for your own applications!