Unverified50% confidenceFactExact time
Distributional reinforcement learning (DRL) learns the full distribution of returns rather than only their expectation, an idea first systematized by Bellemare et al. in the C51 algorithm.
1
Sources
50%
Confidence
Long-term
Relevance
10/22/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedDRaFT利用扩散模型去噪过程的可微性，将奖励函数梯度直接回传至模型参数，无需强化学习的策略梯度估计器，规避高方差问题69% similarUnverifiedDrex DLM's pointer head directly scores the answer options rather than decoding text, producing a calibrated probability distribution over the fixed options in a single forward pass.65% similarUnverifiedRLCD constructs positive and negative sample pairs for contrastive learning, combines reinforcement learning signals to distill model output distributions, and has advantages in scenarios that do not require manually annotated rewards.65% similarUnverifiedThe researchers proposed a credal-set-based method in which the learner's prediction output is no longer a single probability distribution but a set of possible distributions presented in the form of a credal set.63% similarUnverifiedIn RLHF and generative model fine-tuning, algorithm design should target the distribution-matching objective directly, exploiting the logarithmic convergence property of sampling regret to build a more efficient training pipeline.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1084257API
curl https://kongchang.com/api/v1/knowledge/claims/1084257MCP
get_claim(id=1084257)