Keywords: World models, Object centric, Causality
Abstract: Taking inspiration from the Independent Causal Mechanisms principle in the context of world models, we present COmpetitive Mechanisms for Efficient Transfer (COMET), a modular world model which leverages reusable, independent mechanisms across different environments. COMET is trained on multiple environments with varying dynamics via a two-step process: competition and composition. This enables the model to recognise and learn transferable mechanisms. Specifically, in the competition phase, COMET is trained with a winner-takes-all gradient allocation, encouraging the emergence of independent mechanisms. These are then re-used in the composition phase, where COMET learns to re-compose learnt mechanisms in ways that capture the dynamics of new environments. In so doing, COMET explicitly reuses prior knowledge, enabling efficient and interpretable adaptation. In contrast to competitive baselines, we demonstrate that COMET captures recognisable mechanisms without supervision and is able to adapt to new environments.
Submission Number: 7
Loading