Title: Bridging Cognitive Foundation and Efficient Specialization: Visual Reasoning in Cultural Contexts

Abstract: This talk presents a case study on visual reasoning in culturally grounded multimedia tasks, emphasizing the complementary roles of foundation models and specialized models. We examine how vision-language models (VLMs) perform on multiple-choice visual questions involving low-resource data, such as Southeast Asian culture, where correct image selection must be supported by segmentation-based evidence. This setting reveals both the reasoning capabilities of foundation models and their limitations in terms of cultural understanding, visual grounding, and latency.

Bio: Zhixin MA is currently a Research Scientist at Singapore Management University. He obtained his Ph.D. in Computer Science from Singapore Management University in 2024, following the completion of his Bachelor's degree in Computer Science at Shandong University in 2019. His research interests encompass multimedia computing and human-in-the-loop interactive systems. His current work focuses on video retrieval, as well as the reasoning and grounding capabilities of multimodal large language models.

Title: Large Language Models for Drug Discovery:  From Molecule Generation to Optimization

Abstract: Molecular design is fundamental to driving innovation across diverse domains, including drug discovery, materials science, energy, and sustainability. While large language models (LLMs)—such as ChatGPT and Grok—have demonstrated superhuman capabilities in natural language tasks, they often fall short when applied to chemical and biological sequence problems. In this talk, I will present recent progress in leveraging LLMs for drug discovery and molecular design. I will discuss our efforts to adapt and enhance GPT models to address the unique challenges posed by the distinct "languages" of chemistry and biology. These challenges include effectively balancing exploration and exploitation in lead discovery, optimizing lead compounds, improving model controllability and safety, and advancing multimodal representations of molecules and proteins through LLMs.

Bio: Xuefeng Liu is a Ph.D. candidate in Computer Science at the University of Chicago and research associate in Argonne National Laboratory. His primary research interests include Generative AI for molecular design, reinforcement learning, active learning, and computational biology. His work has been selected for oral and spotlightXuefeng Liu is a Ph.D. candidate in Computer Science at the University of Chicago and research associate in Argonne National Laboratory. His primary research interests include Generative AI for molecular design, reinforcement learning, active learning, and computational biology. His work has been selected for oral and spotlight presentations at top conferences and workshops, including NeurIPS and ICLR. He has been awarded the Crerar Fellowship. He organized the 2025 1st Midwest AI for Drug Discovery and Development workshop. presentations at top conferences and workshops, including NeurIPS and ICLR. He has been awarded the Crerar Fellowship. He organized the 2025 1st Midwest AI for Drug Discovery and Development workshop.