A few-shot LLM planner for ALFRED that relabels suboptimal trajectories with hindsight prompts reaches 25.51 SR on Test Seen, approaching or beating the full-shot HLSM baseline.
Hindsight Value Function for Variance Reduction in Stochastic Dynamic Environment
1 Pith paper cite this work. Polarity classification is still indexing.
abstract
Policy gradient methods are appealing in deep reinforcement learning but suffer from high variance of gradient estimate. To reduce the variance, the state value function is applied commonly. However, the effect of the state value function becomes limited in stochastic dynamic environments, where the unexpected state dynamics and rewards will increase the variance. In this paper, we propose to replace the state value function with a novel hindsight value function, which leverages the information from the future to reduce the variance of the gradient estimate for stochastic dynamic environments. Particularly, to obtain an ideally unbiased gradient estimate, we propose an information-theoretic approach, which optimizes the embeddings of the future to be independent of previous actions. In our experiments, we apply the proposed hindsight value function in stochastic dynamic environments, including discrete-action environments and continuous-action environments. Compared with the standard state value function, the proposed hindsight value function consistently reduces the variance, stabilizes the training, and improves the eventual policy.
fields
cs.AI 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Hindsight Planner: A Closed-Loop Few-Shot Planner for Embodied Instruction Following
A few-shot LLM planner for ALFRED that relabels suboptimal trajectories with hindsight prompts reaches 25.51 SR on Test Seen, approaching or beating the full-shot HLSM baseline.