English

Model-Value Inconsistency as a Signal for Epistemic Uncertainty

Machine Learning 2022-07-01 v3 Artificial Intelligence

Abstract

Using a model of the environment and a value function, an agent can construct many estimates of a state's value, by unrolling the model for different lengths and bootstrapping with its value function. Our key insight is that one can treat this set of value estimates as a type of ensemble, which we call an \emph{implicit value ensemble} (IVE). Consequently, the discrepancy between these estimates can be used as a proxy for the agent's epistemic uncertainty; we term this signal \emph{model-value inconsistency} or \emph{self-inconsistency} for short. Unlike prior work which estimates uncertainty by training an ensemble of many models and/or value functions, this approach requires only the single model and value function which are already being learned in most model-based reinforcement learning algorithms. We provide empirical evidence in both tabular and function approximation settings from pixels that self-inconsistency is useful (i) as a signal for exploration, (ii) for acting safely under distribution shifts, and (iii) for robustifying value-based planning with a learned model.

Keywords

Cite

@article{arxiv.2112.04153,
  title  = {Model-Value Inconsistency as a Signal for Epistemic Uncertainty},
  author = {Angelos Filos and Eszter Vértes and Zita Marinho and Gregory Farquhar and Diana Borsa and Abram Friesen and Feryal Behbahani and Tom Schaul and André Barreto and Simon Osindero},
  journal= {arXiv preprint arXiv:2112.04153},
  year   = {2022}
}

Comments

The first three authors contributed equally. Accepted at ICML 2022