Connections between Weighted Automata, Spectral Learning and Tensor Networks - Guillaume Rabusseau

Abstract: In this talk, I will present fundamental connections between weighted automata, spectral learning and tensor networks. Tensor network methods have been a key ingredient of advances in condensed matter physics and have recently sparked interest in the machine learning community for their ability to compactly represent very high-dimensional objects. I will show how weighted automata are equivalent to the so-called uniform matrix product state models from quantum physics and the tensor train decomposition used in machine learning and numerical analysis. I will also briefly present how this equivalence extends to connections with recurrent neural networks and context free grammars. Building upon these connections, I will introduce a novel view of the spectral learning algorithm for weighted automata using the tensor network perspective and discuss promising research directions leveraging these connections for learning with structured data.

Guillaume Rabusseau is an assistant professor at Univeristé de Montréal and holds a Canada CIFAR AI chair at the Mila research institute. Prior to joining Mila, he was an IVADO postdoctoral research fellow in the Reasoning and Learning Lab at McGill University, where he worked with Prakash Panangaden, Joelle Pineau and Doina Precup. He obtained his PhD in computer science in 2016 at Aix-Marseille University under the supervision of François Denis and Hachem Kadri. His research interests lie at the intersection of theoretical computer science and machine learning, and his work revolves around exploring inter-connections between tensors and machine learning to develop efficient learning methods for structured data relying on linear and multilinear algebra.

[gayo]

Thanks for presenting the ideas in such an intuitive way.