Research map / Reinforcement learning / Policy gradient and actor-critic
Deep Reinforcement Learning with Double Q-Learning
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 |
|---|---|---|---|
| Replace the DQN max target with online-network action selection and target-network action evaluation | DQN on Atari · Atari 49 games, no-op starts, 5 minutes | median normalized score: ~93.5% -> 114.7% | Helped |
| Replace the DQN max target with online-network action selection and target-network action evaluation | DQN on Atari · Atari 49 games, no-op starts, 5 minutes | mean normalized score: 241.1% -> 330.3% | Helped |
| Replace the DQN max target with online-network action selection and target-network action evaluation | DQN on Atari · Atari 49 games, human starts, 30 minutes | median normalized score: ~47.5% -> ~88.4% | Helped |
| Replace the DQN max target with online-network action selection and target-network action evaluation | DQN on Atari · Atari 49 games, human starts, 30 minutes | mean normalized score: 122.0% -> 273.1% | Helped |
| Tune target-copy interval, exploration rates, and shared output-layer bias (ablation) | Double DQN · Atari 49 games, human starts, 30 minutes | median normalized score: ~88.4% -> 116.7% | Helped |
| Tune target-copy interval, exploration rates, and shared output-layer bias (ablation) | Double DQN · Atari 49 games, human starts, 30 minutes | mean normalized score: 273.1% -> 475.2% | Helped |
| Use a 30,000-frame rather than 10,000-frame target-network copy interval (ablation) | Double DQN (tuned) · Atari, human starts | performance | Helped |
| Reduce learning exploration from epsilon=0.1 to epsilon=0.01 (ablation) | Double DQN (tuned) · Atari, human starts | performance | Mixed |
| Reduce evaluation exploration to epsilon=0.001 (ablation) | Double DQN (tuned) · Atari, human starts | performance | Helped |
| Use a single shared bias for all action values in the top layer (ablation) | Double DQN (tuned) · Atari, human starts | performance | Helped |
Closest papers by meaning
- Deep Double Q-learning (NeurIPS 2026 · similarity 0.93)
- Convergent and Efficient Deep Q Learning Algorithm (ICLR 2022 · similarity 0.91)
- Confounding Robust Deep Reinforcement Learning: A Causal Approach (NeurIPS 2025 · similarity 0.90)
- ADDQ: Adaptive distributional double Q-learning (ICML 2025 · similarity 0.89)
- Adaptive $Q$-Network: On-the-fly Target Selection for Deep Reinforcement Learning (ICLR 2025 · similarity 0.88)
- Simplifying Deep Temporal Difference Learning (ICLR 2025 · 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.