This talk will be held in RIKEN AIP Open Space (only for AIP members) and online.
Title: Tensor Networks for Machine Learning: Models, Applications, and Learning Methods
Abstract:
Tensor networks (TNs) provide efficient and structured representations of high-dimensional objects and have become increasingly relevant in machine learning, with applications ranging from data compression to the design of expressive models. In this talk, I will present an overview of different approaches to integrating TNs into machine learning, focusing on their use as standalone models, their connections to deep learning, and the methods used to learn these representations.
I will discuss some of my recent work exploring the advantages of TN models, particularly for privacy, adversarial robustness, and interpretability, where their mathematical structure and tractability offer opportunities beyond conventional neural networks. I will then consider how tensorization methods can be used to compress and transform pretrained models, as well as the challenges of scaling TN representations to more complex architectures.
Finally, I will discuss different approaches to learning TN representations, from direct optimization to tensor decompositions of high-dimensional functions, and ongoing work towards extending these methods beyond matrix product states to more general tensor-network geometries.
Affiliation: Department of Artificial Intelligence, Polytechnic University of Madrid, Spain.
Short bio:
José Ramón Pareja Monturiol is an Assistant Professor in the Department of Artificial Intelligence at the Polytechnic University of Madrid (UPM), Spain. He holds a B.Sc. in Mathematics and an M.Sc. in Computational Statistics from the Complutense University of Madrid (UCM), and a PhD in Mathematics from UCM and the Institute of Mathematical Sciences (ICMAT), where he worked in the Mathematics and Quantum Information group (mathQI). During his PhD, he was also a visiting researcher at the Perimeter Institute for Theoretical Physics in Canada. His research focuses on tensor networks for machine learning, particularly their applications to privacy, interpretability, and robustness, as well as tensorization methods and low-rank decompositions.
