Sharing Lifelong Reinforcement Learning Knowledge via Modulating Masks
This work addresses the challenge of scalable and efficient knowledge sharing among lifelong learning agents in dynamic, decentralized environments, representing an incremental advance in distributed lifelong reinforcement learning.
The paper tackles the problem of enabling multiple lifelong reinforcement learning agents to share knowledge in a distributed, asynchronous system by using modulating masks for on-demand transfer of specific task knowledge, resulting in robust performance and rapid learning compared to distributed RL baselines like DD-PPO, IMPALA, and PPO+EWC.
Lifelong learning agents aim to learn multiple tasks sequentially over a lifetime. This involves the ability to exploit previous knowledge when learning new tasks and to avoid forgetting. Modulating masks, a specific type of parameter isolation approach, have recently shown promise in both supervised and reinforcement learning. While lifelong learning algorithms have been investigated mainly within a single-agent approach, a question remains on how multiple agents can share lifelong learning knowledge with each other. We show that the parameter isolation mechanism used by modulating masks is particularly suitable for exchanging knowledge among agents in a distributed and decentralized system of lifelong learners. The key idea is that the isolation of specific task knowledge to specific masks allows agents to transfer only specific knowledge on-demand, resulting in robust and effective distributed lifelong learning. We assume fully distributed and asynchronous scenarios with dynamic agent numbers and connectivity. An on-demand communication protocol ensures agents query their peers for specific masks to be transferred and integrated into their policies when facing each task. Experiments indicate that on-demand mask communication is an effective way to implement distributed lifelong reinforcement learning and provides a lifelong learning benefit with respect to distributed RL baselines such as DD-PPO, IMPALA, and PPO+EWC. The system is particularly robust to connection drops and demonstrates rapid learning due to knowledge exchange.