Invited speaker Dr Yiannis Vlassopoulos will give a talk on Language as a poset and Tensor Networks, on Monday 8 February at 12.00.
Language can be described in algebraic terms as a partially ordered set (poset) where a text is smaller than another if and only if the former is included in the later. This poset structure underlies Latent Semantic Analysis as well as Neural Network language models like GPT.
We will explain how trying to compute this structure from raw text naturally leads to tensor network machine learning. As a result, we construct a poset map into the poset of positive definite operators on a vector space. The operator corresponding to a text gives the probability of the next word after that text.
We will also discuss why, from the algebraic point of view, the language poset which a priori describes language syntax would be expected to encode semantic information.
Yiannis Vlassopoulos earned a B.S in mathematics from the University of Athens, Greece (1991) and an M.A and a Ph.D in mathematics from Duke University, NC, USA (1998). Since then he has worked extensively at the Institut des Hautes Etudes Scientifiques, Paris, France, where he has had a long term collaboration with permanent professor Maxim Kontsevich. He has also worked at the University of Athens, the Max Planck Institute for Mathematics in Bonn, Germany, the University of Vienna, Austria, the Centre for Quantum Geometry of Moduli Spaces, Aarhus University, Denmark, the University of Miami, Coral Gables, Florida, USA and the Simons Center for Geometry and Physics, Stony Brook, NY, USA. His work was on mathematics related to String Theory and in particular on non-commutative algebra and geometry. Since 2016 he has worked on the mathematics of machine learning, in particular applied to natural language processing. Since 2017 he has been chief science officer at the company he co-founded (Tunnel Technologies) in NY, USA. The goal of the company is to develop machine learning for natural language aimed at applications such as abstractive summarization, logical entailment and general applications that require text comprehension.
Topic: SKEL Talk: Language as a poset and Tensor Networks
Time: Feb 8, 2021 12:00 Athens
Meeting ID: 898 9970 3863