ARTICLE · 1095989
9月30日“软件新技术讲坛”学术报告 - 张梦晓 助理教授
9月30日“软件新技术讲坛”学术报告 - 张梦晓 助理教授 Toward Optimal Switching Regret for Multi-Armed Bandits with Oblivious Adversary Zoom会议号：862 4606 1765 密码：20260930 
张梦晓,助理教授Business Analytics Department University of Iowa In sequential decision-making with bandit feedback, a learner must adapt to changing conditions while observing only the loss of its chosen action. Switching regret measures performance against the best sequence of actions that changes a limited number of times. Although near-optimal guarantees are achievable when this number is known in advance, adapting to an unknown number of switches has remained a central challenge. In this talk, I will present an algorithm that achieves near-optimal expected switching regret without knowing the number of switches, simultaneously across all switch budgets. This resolves an open problem for oblivious adversaries and contrasts with the impossibility of such adaptation against adaptive adversaries. The key idea is to search for local opportunities to improve rather than explicitly detect switches. The algorithm combines a fixed-share learner with auxiliary learners operating across multiple time scales and uses their accumulated improvements to adapt its learning rate. I will discuss the main algorithmic ideas and explain how they enable adaptation under limited feedback. Mengxiao Zhang is an assistant professor at the Business Analytics Department at University of Iowa. He obtained a PhD degree in Computer Science at University of Southern California. His research is about designing robust and adaptive machine learning algorithms with strong theoretical guarantees, with a focus on general sequential learning problems, including online learning, bandit problems, game theory and various operational research and revenue management applications. He has published papers in top-tier machine learning conferences and learning theory conferences, including spotlight and oral presentations. He has interned with Microsoft Research and Amazon, and received a B.S. in the School of EECS from Peking University. 
欢迎参加
时间：2026年9月30日（星期三）10:15
摘 要
报告人简介