Research map / Reinforcement learning / Visual and representation RL
Deep Q-learning From Demonstrations
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 demonstration pretraining and retain demonstration data during replay | PDD DQN · 42 Atari games, first million steps | game score: 41 of 42 games | Helped |
| Use DQfD instead of PDD DQN | Atari games · 42 Atari games | steps until PDD DQN catches up: 83 million steps | Helped |
| Combine demonstration learning with reinforcement learning rather than imitation alone | pure imitation learning · 42 Atari games | mean game score: 39 of 42 games | Helped |
| Use DQfD with human demonstrations | Atari agents · 42 Atari games | game score versus best demonstration: 14 of 42 games | Helped |
| Use DQfD with human demonstrations | Atari agents · 42 Atari games | game score versus prior best deep RL: 11 of 42 games | Helped |
| Add supervised large-margin classification loss during pretraining (ablation) | DQfD · Montezuma's Revenge and Q-Bert | online reward | Helped |
| Add n-step TD loss during pretraining (ablation) | DQfD · Montezuma's Revenge and Q-Bert | online reward | Helped |
| Replace the large-margin supervised loss with cross-entropy loss (ablation) | DQfD (ADET variant) · Montezuma's Revenge and Q-Bert | online reward | Hurt |
| Use replay-buffer initialization with demonstrations but no DQfD pretraining and supervised loss (ablation) | DQfD (RBS variant) · Montezuma's Revenge and Q-Bert | online reward | Hurt |
| Use mixed demonstration and agent replay without DQfD pretraining and supervised loss (ablation) | DQfD (HER variant) · Montezuma's Revenge and Q-Bert | online reward | Hurt |
Closest papers by meaning
- Deep Reinforcement Learning with Double Q-Learning (AAAI 2016 · similarity 0.88)
- Confounding Robust Deep Reinforcement Learning: A Causal Approach (NeurIPS 2025 · similarity 0.88)
- Learning from Suboptimal Data in Continuous Control via Auto-Regressive Soft Q-Network (ICML 2025 · similarity 0.87)
- Simplifying Deep Temporal Difference Learning (ICLR 2025 · similarity 0.87)
- Deep Double Q-learning (NeurIPS 2026 · similarity 0.87)
- Adaptive $Q$-Network: On-the-fly Target Selection for Deep Reinforcement Learning (ICLR 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.