Real-World AI: Risk and Robustness in Reinforcement Learning and Kalman Filtering

Join a technical lecture exploring the challenges and solutions in applying artificial intelligence to real-world scenarios, focusing on risk management in Reinforcement Learning (RL) and robustness in Kalman Filtering (KF). Discover why optimizing for risk measures instead of expected returns in RL applications creates unexpected challenges, including compromised data efficiency and biased policy gradients, and learn proposed solutions to these issues. Examine how model misspecification affects Kalman filtering systems and understand a straightforward approach to enhance robustness against assumption violations. Led by Ido Greenberg, a PhD candidate at Technion and Nvidia research intern, whose research spans risk-aversion in RL, conversational planning at Google, and NP-hard routing problems at Nvidia, along with medical and biological forecasting applications.