MAGIK: Mapping to Analogous Goals via Imagination-enabled Knowledge Transfer
This addresses the challenge of inefficient retraining in RL for tasks with structural similarities, offering a novel transfer mechanism that could reduce training costs, though it is incremental as it builds on existing analogy and imagination concepts.
The paper tackles the problem of reinforcement learning agents requiring extensive retraining for new tasks by proposing MAGIK, a framework that enables zero-shot knowledge transfer to analogous tasks without interacting with the target environment, achieving effective transfer using only a small number of human-labelled examples in custom MiniGrid and MuJoCo tasks.
Humans excel at analogical reasoning - applying knowledge from one task to a related one with minimal relearning. In contrast, reinforcement learning (RL) agents typically require extensive retraining even when new tasks share structural similarities with previously learned ones. In this work, we propose MAGIK, a novel framework that enables RL agents to transfer knowledge to analogous tasks without interacting with the target environment. Our approach leverages an imagination mechanism to map entities in the target task to their analogues in the source domain, allowing the agent to reuse its original policy. Experiments on custom MiniGrid and MuJoCo tasks show that MAGIK achieves effective zero-shot transfer using only a small number of human-labelled examples. We compare our approach to related baselines and highlight how it offers a novel and effective mechanism for knowledge transfer via imagination-based analogy mapping.