Research map / Reinforcement learning / Policy gradient and actor-critic
Learning Novel Policies For Tasks
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 |
|---|---|---|---|
| Add a novelty reward and bisector updates instead of using PPO with different random seeds | TNB · 4-Way Maze | Avg # Paths Found: 1.4 -> 4 | Helped |
| Add a novelty reward and bisector updates instead of using SAC | TNB · 4-Way Maze | Avg # Paths Found: 1 -> 4 | Helped |
| Add gradient projections when task and novelty gradients conflict (ablation) | TNB · 4-Way Maze | # Trials In Order: 3/5 -> 4/5 | Helped |
| Add novelty reward and bisector updates instead of using PPO with different random seeds | TNB · D-Maze | Avg # Succ Policies: 0.2 -> 1.2 | Helped |
| Add novelty reward and bisector updates instead of using SAC | TNB · D-Maze | Avg # Succ Policies: 0 -> 1.2 | Helped |
| Add gradient projections when task and novelty gradients conflict (ablation) | TNB · D-Maze | # Succ Trials: 3/5 -> 4/5 | Helped |
| Add novelty reward and bisector updates instead of using PPO with different random seeds | TNB · D-Reacher | Avg # Succ Policies: 0.2 -> 2.4 | Helped |
| Add novelty reward and bisector updates instead of using SAC | TNB · D-Reacher | Avg # Succ Policies: 0 -> 2.4 | Helped |
| Add gradient projections when task and novelty gradients conflict (ablation) | TNB · D-Reacher | Avg # Succ Policies: 2.6 -> 2.4 | Hurt |
| Add gradient projections when task and novelty gradients conflict (ablation) | TNB · D-Reacher | # Succ Trials: 4/5 -> 4/5 | No clear effect |
| Use a fixed weighted sum of task and novelty rewards instead of TNB | WSR · D-Maze | Avg # Succ Policies: 0.6 -> 1.2 | Hurt |
| Use a fixed weighted sum of task and novelty rewards instead of TNB | WSR · D-Reacher | Avg # Succ Policies: 1.2 -> 2.4 | Hurt |
Closest papers by meaning
- Going Beyond Heuristics by Imposing Policy Improvement as a Constraint (NeurIPS 2024 · similarity 0.90)
- Modular Multitask Reinforcement Learning with Policy Sketches (ICML 2017 · similarity 0.89)
- Learning Self-Imitating Diverse Policies (ICLR 2019 · similarity 0.88)
- Goal-Conditioned Reinforcement Learning with Imagined Subgoals (ICML 2021 · similarity 0.88)
- Efficient Language-instructed Skill Acquisition via Reward-Policy Co-Evolution (AAAI 2025 · similarity 0.88)
- Foundation Policies with Hilbert Representations (ICML 2024 · similarity 0.87)
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.