Uploaded October 2025 | Updated September 2026, 1 day ago
Large Language Models (LLMs) are redefining scientific research, transforming how we acquire knowledge and interpret data. These models offer the potential to create autonomous "AI scientists" capable of acquiring knowledge from virtual databases and experimental labs, unlocking new frontiers in discovery.
In this talk, Chang Ma outlines a vision for LLM-based autonomous agents in science, focusing on two key challenges: (1) enabling long-horizon planning to address complex tasks, and (2) building efficient, knowledge-intensive tools and environments for scientific exploration. Using examples from prior work, she demonstrates how these advancements can drive autonomous discovery. Finally, she discusses future directions, including lab-in-the-loop systems and solving open-ended problems, to realize AI-driven scientific breakthroughs.
Chang is a final-year Ph.D. candidate in Computer Science at the University of Hong Kong, advised by Lingpeng Kong and Tao Yu. Her research focuses on the intersection of language agents, AI4Science, and trustworthy systems, aiming to develop AI systems capable of autonomously driving scientific discoveries. Her work has been published in top-tier venues such as NeurIPS, ICML, ICLR, ACL, and EMNLP, with honors including the NeurIPS Oral Award. Chang has gained research experience at MILA, Microsoft, and Genentech, and is an active contributor to the NLP and AI4Science communities.
Large Language Models (LLMs) are redefining scientific research, transforming how we acquire knowledge and interpret data. These models offer the potential to create autonomous "AI scientists" capable of acquiring knowledge from virtual databases and experimental labs, unlocking new frontiers in discovery.
In this talk, Chang Ma outlines a vision for LLM-based autonomous agents in science, focusing on two key challenges: (1) enabling long-horizon planning to address complex tasks, and (2) building efficient, knowledge-intensive tools and environments for scientific exploration. Using examples from prior work, she demonstrates how these advancements can drive autonomous discovery. Finally, she discusses future directions, including lab-in-the-loop systems and solving open-ended problems, to realize AI-driven scientific breakthroughs.
Chang is a final-year Ph.D. candidate in Computer Science at the University of Hong Kong, advised by Lingpeng Kong and Tao Yu. Her research focuses on the intersection of language agents, AI4Science, and trustworthy systems, aiming to develop AI systems capable of autonomously driving scientific discoveries. Her work has been published in top-tier venues such as NeurIPS, ICML, ICLR, ACL, and EMNLP, with honors including the NeurIPS Oral Award. Chang has gained research experience at MILA, Microsoft, and Genentech, and is an active contributor to the NLP and AI4Science communities.
