Thursday / CS Seminar

October 01 / 1:00PM

Hybrid / Mudd 3514

Speaker

Cem Subakan, Laval University

Talk Title

Faithful and Grounded Audio Language Models

Abstract

Large Audio Language Models has been a recent research focus in the audio community. These models exhibit impressive capabilities for audio understanding tasks such as question answering. Moreover they reveal a reasoning chain for their decisions, which is promising for explainability of these models. It is however unclear whether these reasoning chains truly reveal the inner decision mechanisms. I will talk about measuring the faithfulness of these reasoning chains, to assess if they can be used as trustworthy explanations. I will also later talk about our recent work grounding the reasoning chains in the audio input, in order to make the model more closely listen to the audio.

Biography

Cem Subakan is a tenure-track assistant professor in Laval University, and Mila-Quebec AI Institute. He has a PhD in Computer Science from UIUC. His research interests are in machine learning for speech / audio and more recently in explainability and AI safety. He's a current member of the IEEE Machine Learning for Signal Processing Technical Committee, and was the lead general chair for IEEE MLSP 2025. He has published in venues such as ICASSP, Interspeech, NeurIPS, ICML, TMLR, TASLP.

Research Interests/Areas: Explainability, Interpretability, Machine Learning for Speech and Audio, Audio Language Models, Audio Source Separation