Abstract: Reinforcement learning (RL) is usually introduced through games and simulations, but many real-world problems also involve sequential decisions under uncertainty. In this talk, I will present three applications of RL in practical machine learning systems. First, I will discuss how RL can be used for time-series anomaly detection, where an agent learns to balance missed anomalies and false alarms over time. Second, I will describe multimodal RL in the presence of adversarial noise, and how robustness issues arise when combining signals from different data sources. Finally, I will show how (deep) RL can be applied to group recommendation systems, where the goal is to optimize long-term engagement while accounting for diverse user preferences within a group. Throughout the talk, the focus will be on the high-level ideas, design choices, and lessons learned, rather than algorithmic details. I will highlight common challenges across these projects—such as reward design, stability, and evaluation—and discuss open questions for deploying RL in real-world settings.
Speaker(s): Banafsheh,
Virtual: https://events.vtools.ieee.org/m/561747
This event has passed.
