MEA.Seddik

MEA.Seddik

Researcher at TII.

Talks and presentations

A Random Matrix Theory Analysis of Linear Generative Models

November 23, 2023

Talk, Mathematics Seminar, Khalifa University, Abu Dhabi

In this talk, we will delve into the asymptotic study of simple linear generative models when both the sample size and data dimension grow to infinity. In this high-dimensional regime, random matrix theory (RMT) appears to be a natural tool to assess the model’s performance by examining its asymptotic learned conditional probabilities, its associated fluctuations, and the model’s generalization error. This analytical approach not only enhances our comprehension of generative language models but might also offer novel insights into their refinement through the lens of high-dimensional statistics and RMT. [slides]

Spike Recovery from Large Random Tensors

April 19, 2023

Talk, Tensor Journal Club, Online

In this talk, we will present an asymptotic analysis of large asymmetric spiked tensor models, relying on tools from random matrix theory. In the first part of the talk, we will provide the analysis of a rank-one model along with some algorithmic implications in terms of possible signal recovery with a polynomial time algorithm. The second part of the talk will present a generalization to a low-rank spiked model with non-independent components, by studying two types of deflation procedures and proposing an improved algorithm. [slides] [video]

Deciphering Asymmetric Spiked Tensor Models via Random Matrix Theory

February 15, 2023

Talk, New York University Abu Dhabi, Abu Dhabi, UAE

In this talk, we will present an asymptotic analysis of large asymmetric spiked tensor models, relying on tools from random matrix theory. In the first part of the talk, we will provide the analysis of a rank-one model along with some algorithmic implications in terms of possible signal recovery with a polynomial time algorithm. The second part of the talk will present a generalization to a low-rank spiked model with non-independent components, by studying two types of deflation procedures and proposing an improved algorithm. [slides]

Asymptotic Analysis of Asymmetric Spiked Tensor Models with Random Matrix Theory

November 24, 2022

Talk, Lagrange Center (Huawei Technologies France), Paris, France

In this talk, we will present an asymptotic analysis of large asymmetric spiked tensor models, relying on tools from random matrix theory. In the first part of the talk, we will provide the analysis of a rank-one model along with some algorithmic implications in terms of possible signal recovery with a polynomial time algorithm. The second part of the talk will present a generalization to a low-rank spiked model with correlated components, by studying a Hotteling-type deflation procedure. [slides]