Research map / Reinforcement learning / Policy gradient and actor-critic
Fast deep reinforcement learning using online adjustments from the past
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 EVA to combine DQN values with replay-buffer trajectory planning | DQN · Gridworld, one coin | data efficiency | Helped |
| Use EVA to combine DQN values with replay-buffer trajectory planning | DQN · Atari, 55 games | learning speed | Helped |
| Use EVA to combine DQN values with replay-buffer trajectory planning | DQN · Atari, 55 games | final performance | Helped |
| Remove parametric evaluations of counter-factual actions, using n-step traces (ablation) | EVA with trajectory-centric planning · Atari, 55 games | human normalised score | Hurt |
| Replace trajectory-centric planning with parametrically augmented KBRL (ablation) | EVA · Atari, 55 games | human normalised score | No clear effect |
| Anneal the EVA mixing parameter to zero over two million steps (ablation) | EVA · Atari experiments | performance | No clear effect |
| Increase the number of neighbours used to estimate non-parametric values (ablation) | EVA · Atari, 55 games | performance | Hurt |
| Use EVA with DQN instead of Double DQN | DQN and Double DQN · Atari | human normalised score | Mixed |
Closest papers by meaning
- Neural Episodic Control (ICML 2017 · similarity 0.89)
- Sample-Efficient Deep Reinforcement Learning via Episodic Backward Update (NeurIPS 2019 · similarity 0.88)
- Reinforcement Learning with Unsupervised Auxiliary Tasks (ICLR 2017 · similarity 0.87)
- Turning Sand to Gold: Recycling Data to Bridge On-Policy and Off-Policy Learning via Causal Bound (NeurIPS 2025 · similarity 0.86)
- Partially Observable Reinforcement Learning with Memory Traces (ICML 2025 · similarity 0.85)
- Prioritized Generative Replay (ICLR 2025 · similarity 0.85)
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.