Flipping Coins to Estimate Pseudocounts for Exploration in Reinforcement Learning
This addresses exploration bottlenecks in reinforcement learning for complex environments, though it is an incremental improvement over prior count-based methods.
The paper tackled the problem of count-based exploration in high-dimensional state spaces by proposing a method that estimates visitation counts using coin flips, resulting in outperforming existing approaches on most of 9 challenging tasks, including Montezuma's Revenge.
We propose a new method for count-based exploration in high-dimensional state spaces. Unlike previous work which relies on density models, we show that counts can be derived by averaging samples from the Rademacher distribution (or coin flips). This insight is used to set up a simple supervised learning objective which, when optimized, yields a state's visitation count. We show that our method is significantly more effective at deducing ground-truth visitation counts than previous work; when used as an exploration bonus for a model-free reinforcement learning algorithm, it outperforms existing approaches on most of 9 challenging exploration tasks, including the Atari game Montezuma's Revenge.