Neural Information Processing Systems http://nips.cc/

Efficient exploration remains a major challenge for reinforcement learning (RL). Common dithering strategies for exploration, such as epsilon-greedy, do not carry out temporally-extended (or deep) exploration; this can lead to exponentially larger data requirements. However, most algorithms for statistically efficient RL are not computationally tractable in complex environments. Randomized value functions offer a promising approach to efficient exploration with generalization, but existing algorithms are not compatible with nonlinearly parameterized value functions. As a first step towards addressing such contexts we develop bootstrapped DQN. We demonstrate that bootstrapped DQN can combine deep exploration with deep neural networks for exponentially faster learning than any dithering strategy. In the Arcade Learning Environment bootstrapped DQN substantially improves learning speed and cumulative performance across most games.

Specifically,becauseaQfunctionis defined with respect toaparticular policy,constructingPˆQ requires selection ofareference policy or distribution over policies.

Efficient exploration remains a major challenge for reinforcement learning (RL). Common dithering strategies for exploration, such as epsilon-greedy, do not carry out temporally-extended (or deep) exploration; this can lead to exponentially larger data requirements. However, most algorithms for statistically efficient RL are not computationally tractable in complex environments. Randomized value functions offer a promising approach to efficient exploration with generalization, but existing algorithms are not compatible with nonlinearly parameterized value functions. As a first step towards addressing such contexts we develop bootstrapped DQN. We demonstrate that bootstrapped DQN can combine deep exploration with deep neural networks for exponentially faster learning than any dithering strategy. In the Arcade Learning Environment bootstrapped DQN substantially improves learning speed and cumulative performance across most games.

Randomised value functions (RVF) can be viewed as a promising approach to scaling PSRL.

This paper proposes using Bayesian linear regression to get a posterior over successor features as a way of representing uncertainty, from which they sample for exploration. I found the characterization of Randomised Policy Iteration to be strange, as it only seems to apply to UBE but not bootstrapped DQN, With bootstrapped DQN, each model in the ensemble is a value function pertaining to a different policy, thus there is no single reference policy. The ensemble is trying to represent a distribution of optimal value functions, rather than value functions for a single reference policy. Proposition 1: In the case of neural networks, and function approximation in general, it is very unlikely that we will get a factored distribution, so this claim does not seem applicable in general. In fact, in general there should be very high correlation between the q-values between nearby states. Is this claim a direct response to UBE? Also the analysis fixes the policy to consider the distribution of value functions, but this seems to not be how posterior sampling is normally considered, but rather only the way UBE considers it.

First, it would take months to re-produce these experiments (besides the hardware requirements). Second, with such complicated algorithms it's hard to know what exactly is leading to the improvement. For this reason I find this kind of paper a little unscientific, but maybe this is how things have to be. I wonder, do the authors plan to release their code? Overall I think this is an interesting idea, but the authors have not convinced me that this is a principled approach.

We uncover a surprising phenomenon in deep reinforcement learning: training a diverse ensemble of data-sharing agents - a well-established exploration strategy - can significantly impair the performance of the individual ensemble members when compared to standard single-agent training. Through careful analysis, we attribute the degradation in performance to the low proportion of self-generated data in the shared training data for each ensemble member, as well as the inefficiency of the individual ensemble members to learn from such highly off-policy data. We thus name this phenomenon the curse of diversity. We find that several intuitive solutions - such as a larger replay buffer or a smaller ensemble size - either fail to consistently mitigate the performance loss or undermine the advantages of ensembling. Finally, we demonstrate the potential of representation learning to counteract the curse of diversity with a novel method named Cross-Ensemble Representation Learning (CERL) in both discrete and continuous control domains. Our work offers valuable insights into an unexpected pitfall in ensemble-based exploration and raises important caveats for future applications of similar approaches. The potential benefits of a diverse ensemble are twofold. At training time, it enables concurrent exploration with multiple distinct policies without the need for additional samples. At test time, the learned policies can be aggregated into a robust ensemble policy, via aggregation methods such as majority voting (Osband et al., 2016) or averaging (Januszewski et al., 2021). Despite the generally positive perception of ensemble-based exploration, we argue that this approach has a potentially negative aspect that has been long overlooked. As shown in Figure 1, for each member in a data-sharing ensemble, only a small proportion of its training data comes from its own interaction with the environment. The majority of its training data is generated by other members of the ensemble, whose policies might be distinct from its own policy. This type of off-policy learning has been shown to be highly challenging in previous work (Ostrovski et al., 2021). We thus hypothesize that similar learning difficulties can also occur in ensemble-based exploration. We verify our hypothesis in the Arcade Learning Environment (Bellemare et al., 2012) with the Bootstrapped DQN (Osband et al., 2016) algorithm and the Gym MuJoCo benchmark (Towers et al., 2023) with an ensemble SAC (Haarnoja et al., 2018a) algorithm. We show that, in many environments, the individual members of a data-sharing ensemble significantly underperform their single-agent counterparts. Moreover, while aggregating the policies of all ensemble members via voting or averaging sometimes compensates for the degradation in individual members' performance, it is not always the case.

E cient exploration remains a major challenge for reinforcement learning (RL). Common dithering strategies for exploration, such as '-greedy, do not carry out temporally-extended (or deep) exploration; this can lead to exponentially larger data requirements. However, most algorithms for statistically e cient RL are not computationally tractable in complex environments. Randomized value functions o er a promising approach to e cient exploration with generalization, but existing algorithms are not compatible with nonlinearly parameterized value functions. As a first step towards addressing such contexts we develop bootstrapped DQN. We demonstrate that bootstrapped DQN can combine deep exploration with deep neural networks for exponentially faster learning than any dithering strategy. In the Arcade Learning Environment bootstrapped DQN substantially improves learning speed and cumulative performance across most games.

In Reinforcement Learning (RL), agents aim at maximizing cumulative rewards in a given environment. During the learning process, RL agents face the dilemma of exploitation and exploration: leveraging existing knowledge to acquire rewards or seeking potentially higher ones. Using uncertainty as a guiding principle provides an active and effective approach to solving this dilemma and ensemble-based methods are one of the prominent avenues for quantifying uncertainty. Nevertheless, conventional ensemble-based uncertainty estimation lacks an explicit prior, deviating from Bayesian principles. Besides, this method requires diversity among members to generate less biased uncertainty estimation results. To address the above problems, previous research has incorporated random functions as priors. Building upon these foundational efforts, our work introduces an innovative approach with delicately designed prior NNs, which can incorporate maximal diversity in the initial value functions of RL. Our method has demonstrated superior performance compared with the random prior approaches in solving classic control problems and general exploration tasks, significantly improving sample efficiency.