Advith's Calendar

Zoom: https://illinois.zoom.us/j/86930636699?pwd=ln22dteObPj6jSsJ5qTaiUwPr7m8OB.1

Refreshments Provided.

Abstract: 
Kai-Wei Chang is an Associate Professor in the Department of Computer Science at UCLA and an Amazon Scholar at Amazon AGI. His research interests include designing trustworthy natural language processing systems and developing multimodal models for vision-language applications. Kai-Wei has published broadly in NLP, AI, and ML. His awards include IEEE AI's 10 to Watch (2024), the Sloan Fellow (2021), AAAI Senior Member (2023), CVPR Best Paper Finalist (2022), EMNLP Best Long Paper Award (2017), and KDD Best Paper Award (2010). Kai-Wei was elected as an officer of SIGDAT, the organizing body behind EMNLP, and will serve as President in 2026. He is an associate editor for journals such as JAIR, JMLR, TACL, and ARR and senior area chair for most ML and NLP conferences. He has delivered multiple tutorials on topics such as Fairness, Robustness, and Multimodal NLP at EMNLP (2019, 2021) and ACL (2023). Kai-Wei received his Ph.D. from the University of Illinois at Urbana-Champaign in 2015 and subsequently worked as a postdoctoral researcher at Microsoft Research in 2016. For more details, visit http://kwchang.net. 

Bio:
Math reasoning poses unique challenges when combined with visual information. While modern models have shown impressive progress in math problem solving, they still tend to make mistakes on seemingly simple visual math tasks. In this talk, I will first discuss how we benchmark and analyze the capabilities of multimodal large language models in solving visual math problems that require understanding geometric relationships, interpreting charts and graphs, and reasoning about spatial configurations in math contexts. I will then introduce techniques we have developed to strengthen grounding in mathematical concepts and highlight our OpenVLThinker framework, one of the first open-source vision-language models to demonstrate chain-of-thought reasoning through iterative self-improvement. I will conclude by outlining current limitations and future directions for building more reliable and interpretable multimodal reasoning systems.

Part of the Siebel School Speakers Series. Faculty Host: Heng Ji

Meeting ID: 869 3063 6699
Passcode: csillinois

If accommodation is required, please email <erink@illinois.edu> or <communications@cs.illinois.edu>. Someone from our staff will contact you to discuss your specific needs