Scalar Reward is Not Enough: The Case for Multi-objective Reinforcement Learning
Abstract:
The 2021 paper "Reward is Enough" by Silver, Singh, Precup and Sutton posits that the concept of reward maximisation is sufficient to underpin all intelligence, both natural and artificial. This talk will explain why scalar rewards are insufficient to account for some aspects of both biological and computational intelligence, and argue in favour of explicitly multi-objective models of reward maximisation. This talk will also provide a brief introduction to the field of multi-objective reinforcement learning, an alternative approach to reinforcement learning based on vector rewards.
Date and Time
Location
Hosts
Registration
-
Add Event to Calendar
Loading virtual attendance info...
- Starts 13 November 2025 02:00 PM UTC
- Ends 10 December 2025 06:00 AM UTC
- No Admission Charge
Speakers
Peter of Federation University
Biography:
Peter Vamplew is a Professor of Information Technology in the Institute of Institute of Innovation, Science and Sustainability (IISS) at Federation University Australia. He currently also serves as the Associate Dean (Research) of the Institute (2023-current). His research focuses on the field of reinforcement learning, where he has been a pioneer in the extension of reinforcement learning to problems with multiple conflicting objectives. Since 2018 he has also been examining the potential application of these techniques within the areas of responsible AI and AI safety. Through this work he was appointed as member of the Future of Life Institute’s Existential AI Risk research community. He currently co-leads the Australian Responsible Autonomous Agents Collective, a research group with a dozen members across multiple Australian universities. He was awarded the 2021 Federation University Vice Chancellor’s Award for Excellence (Research Excellence) “Acknowledging world-leading research in the area of multi-objective reinforcement learning”.
Email:
Address:Victoria, Australia