On June 26th, Dr. Arman Zharmagambetov, a Research Scientist at Meta Fundamental AI Lab, delivered a lecture on “Guardrailing AI Systems for Safety and Robustness: A Discrete Optimization Perspective” at the Institute of Smart Systems and Artificial Intelligence (ISSAI) of Nazarbayev University.
Dr. Zharmagambetov commenced his talk by discussing the crucial role of discrete optimization in creating safe AI systems. His presentation delved into deep generative models such as GANs and VAEs, the GenCO framework for generating compliant deep generative models, and AdvPrompter, a groundbreaking framework capable of producing human-readable adversarial prompts in seconds. He explored how these methods can be applied to red-team large language models (LLMs), contributing to the development of safe and robust models.
Dr. Zharmagambetov also shared his experiences at Meta’s FAIR lab, where he specializes in the intersection of machine learning and optimization. His recent work includes discrete optimization for foundational models, machine learning-guided optimization, and reinforcement learning.
The event attracted over 90 participants, both in person and online, who were highly interested in Dr. Zharmagambetov’s presentation. The session concluded with a Q&A session, where Dr. Zharmagambetov addressed questions from the audience and also provided advice to aspiring young researchers.