MAGIK: Mapping to Analogous Goals via Imagination-enabled Knowledge Transfer

2025-06-02

AI Summary

Reinforcement learning (RL) agents typically require extensive retraining when deployed on structurally similar but novel tasks, whereas humans achieve zero-shot knowledge transfer via analogical reasoning. Method: We propose the first zero-shot analogical transfer framework for RL that operates without any interaction with the target environment. Our approach leverages symbolic imagination modeling to establish cross-domain entity alignment between source and target tasks, and introduces a human-annotated guidance mechanism for analogical mapping to enable semantic-level policy reuse and adaptation. Contribution/Results: The method overcomes fundamental limitations of conventional transfer RL—namely, dependence on target-environment sampling and parameter fine-tuning. Evaluated on customized MiniGrid and MuJoCo benchmarks, it significantly outperforms existing baselines, achieving efficient transfer with only minimal human annotations. This work establishes a new paradigm for structured, interpretable knowledge reuse in RL.

Technology Category

• Cognitive Modeling & Cognitive Systems: Analogy
• Machine Learning: Imitation Learning & Inverse Reinforcement Learning
• Computer Vision: Visual Reasoning & Symbolic Representations

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

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.

Problem

• Enables RL agents to transfer knowledge without retraining
• Maps target task entities to source domain via imagination
• Achieves zero-shot transfer with minimal human examples

Innovation

• Imagination mechanism maps target to source entities
• Zero-shot transfer with minimal human-labeled examples
• Reuses original policy via analogy mapping