MPI-DS Seminar: Chaos-Based Reinforcement Learning with TD3
MPI-DS Seminar
- Datum: 11.03.2024
- Uhrzeit: 13:15 - 14:15
- Vortragende(r): Prof. Toshitaka Matsuki
- National Defense Academy of Japan, Yokosuka, Japan
- Ort: Max-Planck-Institut für Dynamik und Selbstorganisation (MPIDS)
- Raum: Lise Meitner Seminar room (0.77)
- Gastgeber: MPI-DS
- Kontakt: ulrich.parlitz@ds.mpg.de
Chaos-based reinforcement learning (CBRL) is a reinforcement learning approach that ex-ploits the internal chaotic dynamics of the learning system for exploration. This approach addresses the question of how the brain achieves the variability of behavior necessary for exploratory learning from the reinforcement learning aspect.
CBRL agents also have the learning advantage of being able to switch between exploration and exploitation autonomously. We expect that improving CBRL can achieve models that learn how to explore or make decisions as transient dynamics. This presentation will report on our recent work introducing deep reinforcement learning techniques to CBRL.
CBRL agents also have the learning advantage of being able to switch between exploration and exploitation autonomously. We expect that improving CBRL can achieve models that learn how to explore or make decisions as transient dynamics. This presentation will report on our recent work introducing deep reinforcement learning techniques to CBRL.