Research map / Reinforcement learning / Policy gradient and actor-critic
f-Policy Gradients: A General Framework for Goal-Conditioned RL using f-Divergences
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use fkl-PG instead of the compared methods | Goal-conditioned RL · PointMazeMedium and PointMazeLarge | task performance | Helped |
| Use fkl-PG instead of the compared methods | Goal-conditioned RL · PointMazeU | task performance | No clear effect |
| Use fkl-PG rather than discriminator-based methods | Goal-conditioned RL · PointMaze with separated initial-state and goal distributions | task performance | Helped |
| Use fkl-PG rather than discriminator-based methods | Goal-conditioned RL · PointMaze with separated initial-state and goal distributions | stability | Helped |
| Use fkl-PG rather than GAIL | Goal-conditioned RL · Gridworld | exploration | Helped |
| Replace fkl-PG with rkl-PG (ablation) | f-PG · PointMazeU, PointMazeMedium, and PointMazeLarge | task performance | Hurt |
| Replace fkl-PG with JS-PG (ablation) | f-PG · PointMazeU, PointMazeMedium, and PointMazeLarge | task performance | Hurt |
| Replace fkl-PG with chi^2-PG (ablation) | f-PG · PointMazeU, PointMazeMedium, and PointMazeLarge | task performance | Hurt |
| Use the fkl learning signal as a shaping reward rather than optimizing fkl-PG directly (ablation) | fkl-PG · PointMaze | task performance | Hurt |
| Add shaping rewards rather than using only the task reward (ablation) | PPO · PointMaze | task performance | Mixed |
| Use fkl-PG rather than AIM or GAIL | Goal-conditioned RL · Gridworld | exploration | Mixed |
Closest papers by meaning
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 2025 · similarity 0.91)
- Last-Iterate Global Convergence of Policy Gradients for Constrained Reinforcement Learning (NeurIPS 2024 · similarity 0.90)
- Global Reinforcement Learning : Beyond Linear and Convex Rewards via Submodular Semi-gradient Methods (ICML 2024 · similarity 0.90)
- Exploration-Driven Policy Optimization in RLHF: Theoretical Insights on Efficient Data Utilization (ICML 2024 · similarity 0.90)
- Global Optimality for Constrained Exploration via Penalty Regularization (NeurIPS 2026 · similarity 0.89)
- Occupancy-based Policy Gradient: Estimation, Convergence, and Optimality (NeurIPS 2024 · similarity 0.89)
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.