Context Meta-Reinforcement Learning via Neuromodulation
| Autor: | Eseoghene Ben-Iwhiwhu, Jeffery Dick, Nicholas A. Ketz, Praveen K. Pilly, Andrea Soltoggio |
|---|---|
| Rok vydání: | 2021 |
| Předmět: |
FOS: Computer and information sciences
Computer Science - Machine Learning Computer Science - Artificial Intelligence Cognitive Neuroscience Guinea Pigs Computer Science - Neural and Evolutionary Computing Machine Learning (cs.LG) Artificial Intelligence (cs.AI) Artificial Intelligence Animals Learning Neural and Evolutionary Computing (cs.NE) Reinforcement Psychology Algorithms |
| DOI: | 10.48550/arxiv.2111.00134 |
| Popis: | Meta-reinforcement learning (meta-RL) algorithms enable agents to adapt quickly to tasks from few samples in dynamic environments. Such a feat is achieved through dynamic representations in an agent's policy network (obtained via reasoning about task context, model parameter updates, or both). However, obtaining rich dynamic representations for fast adaptation beyond simple benchmark problems is challenging due to the burden placed on the policy network to accommodate different policies. This paper addresses the challenge by introducing neuromodulation as a modular component to augment a standard policy network that regulates neuronal activities in order to produce efficient dynamic representations for task adaptation. The proposed extension to the policy network is evaluated across multiple discrete and continuous control environments of increasing complexity. To prove the generality and benefits of the extension in meta-RL, the neuromodulated network was applied to two state-of-the-art meta-RL algorithms (CAVIA and PEARL). The result demonstrates that meta-RL augmented with neuromodulation produces significantly better result and richer dynamic representations in comparison to the baselines. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|