Foundations of Multi-objective Learning for Multi-modal Intelligence

Lisha Chen, Ph.D. candidate, Department of Electrical, Computer, and Systems Engineering, Rensselaer Polytechnic Institute

1400 Wegmans Hall

Abstract: Large-scale AI models such as ChatGPT-4 and Gemini have achieved remarkable progress recently. To unlock their potential and propel advancements in the next generation of AI, a key step is to empower them to proficiently handle multiple tasks, modalities, and metrics, which we generally term multi-objective learning (MOL). However, the integration of these models into real-world societal systems is hindered by the lack of interpretability and theoretical foundations.

In this talk, I will first discuss current computational and statistical challenges in MOL, including conflicts of objectives, and generalization abilities to new instances, tasks, and environments. Next, I will introduce our approach to tackling these challenges through the lens of a unified optimization and statistical learning theory. Then I will focus on how we can tailor these algorithms to solve specific machine vision and speech processing tasks, and integrate these processors into multi-modal AI systems. Finally, I will conclude by outlining future research directions, spanning user-friendly and modularized MOL optimizers, new applications in speech and image processing, and the integration of MOL into human-centered AI systems.

 
Bio: Lisha Chen (https://lisha-chen.github.io/) is a Ph.D. candidate in the Department of Electrical, Computer, and Systems Engineering at Rensselaer Polytechnic Institute (RPI). She received her M.S. in Electrical Engineering from RPI in 2021, and her B.S. from Huazhong University of Science and Technology in 2017. Her research focuses on the theoretical foundations of multi-objective learning and meta learning, as well as their applications to machine vision and speech tasks. For those topics, she has co-authored a monograph in Foundations and Trends in Signal Processing and will serve as a tutorial speaker at AAAI 2024 and ICASSP 2024.

Lisha Chen has received several awards, including the IEEE Signal Processing Society Scholarship, the Rensselaer’s Founders Award of Excellence, the Belsky Award for Computational Sciences and Engineering, the IBM-AIRC PhD Fellowship, and the National Scholarship from China. 

Refreshments will be provided.