Talk page

Title:
Learning Relevant Features of Data Using Multi Scale Tensor Networks

Speaker:
Miles Stoudenmire

Abstract:
Tensor networks have been very successful in the field of many-body physics, since they underpin powerful algorithms and also give qualitative insights into different phases of matter. But more abstractly, tensor networks are simply a tool to compress tensors with a very large number of indices. Using them in the context of models for machine learning allows one to efficiently parameterize very expressive and interesting models. Previous applications include supervised learning and generative modeling of real-world data sets.

Link:
http://scgp.stonybrook.edu/video_portal/video.php?id=3490

Workshop:
Simons- Workshop: Tensor-Network Methods: Structure, Applications and Holography