Research map / Reinforcement learning / Visual and representation RL
GEP-PG: Decoupling Exploration and Exploitation in Deep Reinforcement Learning Algorithms
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 |
|---|---|---|---|
| Pre-fill DDPG's replay buffer with GEP-generated samples, then use parameter perturbation | DDPG on Half-Cheetah · Half-Cheetah | absolute performance: 5445 (1265) -> 6118 (634) | Helped |
| Pre-fill DDPG's replay buffer with GEP-generated samples, then use parameter perturbation | DDPG on Half-Cheetah · Half-Cheetah | final performance: 4961 (1288) -> 5740 (573) | Helped |
| Pre-fill DDPG's replay buffer with GEP-generated samples, then use action perturbation | DDPG on Half-Cheetah · Half-Cheetah | absolute performance: 3943 (1209) -> 5033 (833) | Helped |
| Pre-fill DDPG's replay buffer with GEP-generated samples, then use action perturbation | DDPG on Half-Cheetah · Half-Cheetah | final performance: 3596 (1102) -> 4521 (738) | Helped |
| Use parameter perturbation instead of action perturbation | DDPG · Half-Cheetah | absolute performance: 3943 (1209) -> 5445 (1265) | Helped |
| Use parameter perturbation instead of action perturbation | DDPG · Half-Cheetah | final performance: 3596 (1102) -> 4961 (1288) | Helped |
| Use GEP exploration before DDPG fine-tuning instead of using GEP alone | GEP on Half-Cheetah · Half-Cheetah | absolute performance: 2140 (881) -> 6118 (634) | Helped |
| Use GEP exploration before DDPG fine-tuning instead of using GEP alone | GEP on Continuous Mountain Car · Continuous Mountain Car | performance | Hurt |
| Use GEP exploration before DDPG fine-tuning instead of training DDPG from scratch | DDPG on Continuous Mountain Car · Continuous Mountain Car | performance | Helped |
| Use a linear policy instead of a (64,64) policy during GEP exploration (ablation) | GEP · Half-Cheetah | final performance | No clear effect |
| Use a linear policy instead of a (64,64) policy during GEP exploration (ablation) | GEP · Half-Cheetah | performance at 2.10^5 steps: 738 (395, 1080) | Helped |
| Increase initial OU action noise by linearly decreasing it from 0.6 to 0 (ablation) | DDPG · Continuous Mountain Car | performance | Helped |
Closest papers by meaning
- Improving Exploration in Evolution Strategies for Deep Reinforcement Learning via a Population of Novelty-Seeking Agents (NeurIPS 2018 · similarity 0.88)
- Exploration Conscious Reinforcement Learning Revisited (ICML 2019 · similarity 0.88)
- Decoupling Exploration and Policy Optimization: Uncertainty Guided Tree Search for Hard Exploration (NeurIPS 2026 · similarity 0.87)
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- Global Optimality for Constrained Exploration via Penalty Regularization (NeurIPS 2026 · similarity 0.86)
- Exploration from a Primal-Dual Lens: Value-Incentivized Actor-Critic Methods for Sample-Efficient Online RL (NeurIPS 2025 · 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.