Research map / Reinforcement learning / Policy gradient and actor-critic
Pitfall of Optimism: Distributional Reinforcement Learning by Randomizing Risk Criterion
Taehyun Cho, Seungyub Han, Heesoo Lee, Kyungjae Lee, Jungwoo Lee
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace optimism with a randomized risk criterion (p-DLTV vs DLTV) in the stochastic N-Chain (ablation) | DLTV · N-Chain (8,10) | total count of optimal action: 9997 -> 14344 | Improved |
| replace optimism with randomized criterion at high intrinsic uncertainty (ablation) | DLTV · N-Chain (1,17) | total count of optimal action: 7257 -> 13404 | Improved |
| use PQR randomized criterion instead of epsilon-greedy | QR-DQN · N-Chain (1,17) | total count of optimal action: 11619 -> 14354 | Improved |
| increase variance of the suboptimal reward (distance between the two Gaussians) (ablation) | DLTV · N-Chain | total count of optimal action: 9997 -> 7257 | Hurt |
| add perturbation exploration to QR-DQN on Atari (PQR) under no-ops | QR-DQN-zoo · Atari 55 | mean human-normalized score: 559 -> 1121 | Improved |
| PQR versus IQN under no-ops | IQN-zoo · Atari 55 | mean human-normalized score: 902 -> 1121 | Improved |
| PQR versus Rainbow under sticky actions | RAINBOW-dopamine · Atari 55 | mean human-normalized score: 965 -> 962 | No clear effect |
| PQR versus Rainbow under no-ops | RAINBOW-zoo · Atari 55 | mean human-normalized score: 1160 -> 1121 | Hurt |
| PQR versus QUOTA and DLTV | PQR · Atari 55 | average human-normalized score: 603.66 -> 1078.00 | Improved |
| ablate schedule: fixed-size ambiguity set (1/t^0) instead of decaying (ablation) | PQR · Pong-v4 | score | Hurt |
| ablate: force optimistic output vector (OPT) instead of random (ablation) | PQR · Pong-v4 | score | Hurt |
| ablate: use sqrt(log t / t) schedule of p-DLTV (ablation) | PQR · Pong-v4 | score | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- QUOTA — QUOTA: The Quantile Option Architecture for Reinforcement Learning (AAAI 2019) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Distributional Reinforcement Learning With Quantile Regression (AAAI 2018 · similarity 0.90)
- Bellman Unbiasedness: Toward Provably Efficient Distributional Reinforcement Learning with General Value Function Approximation (ICML 2025 · similarity 0.88)
- Intrinsic Benefits of Categorical Distributional Loss: Uncertainty-aware Regularized Exploration in Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- Near-Minimax-Optimal Distributional Reinforcement Learning with a Generative Model (NeurIPS 2024 · similarity 0.87)
- More Benefits of Being Distributional: Second-Order Bounds for Reinforcement Learning (ICML 2024 · similarity 0.87)
- Provable Risk-Sensitive Distributional Reinforcement Learning with General Function Approximation (ICML 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
