Research map / Reinforcement learning / Policy gradient and actor-critic
Policy gradients for special settings
Extend policy gradients to continuous time, performative, bounded-action, and diffusion-policy settings.
45 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in Policy gradient and actor-critic, keyword cluster “policy gradient · gradients · variance”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
11 papers in 2023–24 and 16 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 3.2% to 3.5%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2017 · Scalable Multitask Policy Gradient Reinforcement Learning (AAAI 2017 · 18 citations)
- 2018 · Credit Assignment For Collective Multiagent RL With Global Rewards (NeurIPS 2018 · 59 citations)
- 2019 · Large-Scale Interactive Recommendation with Tree-Structured Policy Gradient (AAAI 2019 · 134 citations)
- 2020 · Gradient Surgery for Multi-Task Learning (NeurIPS 2020 · 99 citations)
- 2021 · Queue-Learning: A Reinforcement Learning Approach for Providing Quality of Service (AAAI 2021 · 27 citations)
- 2022 · Continuous MDP Homomorphisms and Homomorphic Policy Gradient (NeurIPS 2022 · 1 citation)
- 2023 · Policy Optimization for Continuous Reinforcement Learning (NeurIPS 2023 · 6 citations)
- 2024 · Excluding the Irrelevant: Focusing Reinforcement Learning through Continuous Action Masking (NeurIPS 2024 · 4 citations)
- 2025 · Reinforcement Learning with Random Time Horizons (ICML 2025 · 0 citations)
- 2026 · Rethinking the Design Space of Reinforcement Learning for Diffusion Models: On the Importance of Likelihood Estimation Beyond Loss Design (ICML 2026 · 0 citations)
Earliest papers here
- Scalable Multitask Policy Gradient Reinforcement Learning (AAAI 2017 · 18 citations)
- Credit Assignment For Collective Multiagent RL With Global Rewards (NeurIPS 2018 · 59 citations)
- Learning Temporal Point Processes via Reinforcement Learning (NeurIPS 2018 · 50 citations)
What has been tried
31 papers here report 244 measured changes (a component added, removed or swapped, compared with a baseline): helped 67%, no clear effect 7%, hurt 20%, mixed 6%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| use 12 rather than 36 generators for the generator mask | Walker2D · episode return | Helped | Excluding the Irrelevant: Focusing Reinforcement Learning through Continuous Action Masking (NeurIPS 2024 · 4 citations) | |
| Reset all optimizer state at target changes instead of retaining momentum and resetting only the timestep | Atari-10 · human-normalized performance | 65.7% vs. 23.5% | Hurt | Adam on Local Time: Addressing Nonstationarity in RL with Relative Adam Timesteps (NeurIPS 2024 · 3 citations) |
| Increase batch size | DDiffPG hyperparameter ablation · sample efficiency | Helped | Learning Multimodal Behaviors from Scratch with Diffusion Policy Gradient (NeurIPS 2024 · 3 citations) | |
| Include the expected stopping-time factor in the policy gradient instead of omitting it | Modified continuous mountain car · convergence speed | Helped | Reinforcement Learning with Random Time Horizons (ICML 2025 · 0 citations) | |
| Use a regularized (SimNorm, Mish) world model for FoG policy learning instead of ReLU model | Ant world model regularization ablation · policy reward | Helped | PWM: Policy Learning with Multi-Task World Models (ICLR 2025 · 0 citations) | |
| Use an ensemble of critics instead of a single critic | MuJoCo locomotion tasks · policy return | Helped | Augmented Bayesian Policy Search (ICLR 2024) |
What reviewers will expect
Datasets and baselines named in the 6 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Baselines
- SAC · 2 of 6 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 27 papers; from OpenAlex and the papers' affiliations).
- Technische Universität Darmstadt · 2 papers
- TU Darmstadt · 2 papers
- Google DeepMind · 2 papers
- Swiss Federal Institute of Technology · 2 papers
Use it
Download BibTeX of 36 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Variance-reduced policy gradients (79 papers)
- Trajectory reuse in policy gradients (68 papers)
- Alternative policy update rules (50 papers)
Elsewhere in the topic
- Model-based RL with guarantees (245 papers)
- Regularized scalable deep value learning (216 papers)
- Temporal-difference and discounting theory (143 papers)
- Tree search and policy optimization (109 papers)
- Safety layers and shielding (85 papers)
- Actor-critic theory (76 papers)
- Off-policy actor-critic with expressive policies (71 papers)
- Distributional value learning (60 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).