Research map / Reinforcement learning / Policy gradient and actor-critic
Inferring DQN structure for high-dimensional continuous control
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 |
|---|---|---|---|
| Inferred module composition (Q-value or sigma uncertainty based) instead of parallel/sequential baselines | DDQN action-module architecture · MuJoCo Hopper, Walker2d, HalfCheetah, Humanoid | maximum total reward (avg 30 games): e.g. Humanoid 3990 (BDDQN) to 4224 (Ours [Q]) | Helped |
| Inferred module composition instead of parallel/sequential baselines | DDQN action-module architecture · AAA sailing game | score: 2.2 vs 1.6 (BDDQN) and 1.3 (SDDQN) | Helped |
| Inferred module composition with Noisy DDQN sigma uncertainty vs Q-value uncertainty | Inferred structure · MuJoCo Hopper and HalfCheetah | maximum total reward: identical scores (3495, 9222) | No clear effect |
| Sigma-based (Noisy DDQN) structure inference instead of Q-value-based | Inferred structure · Walker2d and Humanoid | maximum total reward: Walker2d 4855 vs 4552 (sigma higher); Humanoid 4074 vs 4224 (sigma lower) | Mixed |
| Module ordering (inferred or real skeleton structure) vs inversed structure (ablation) | Action module composition · MuJoCo locomotion | learning curve performance: qualitative | Helped |
| Stopping threshold k for structure inference (1.0 to 2.0 times initial score) (ablation) | Uncertainty-based structure inference · MuJoCo Hopper, Walker2d, HalfCheetah, Humanoid | structure recovered matches full-training structure (5 runs): Hopper succeeds at all k; Humanoid needs k=1.5 or higher | Mixed |
| Number of averaged runs n for uncertainty estimation (1 to 5) (ablation) | Uncertainty-based structure inference · MuJoCo Hopper, Walker2d, HalfCheetah, Humanoid | structure recovered matches full-training structure (5 runs): Humanoid needs n=3 to succeed; Hopper succeeds at n=1 | Helped |
Closest papers by meaning
- Universal Approximation Theorem of Deep Q-Networks (ICML 2025 · similarity 0.87)
- Convergent and Efficient Deep Q Learning Algorithm (ICLR 2022 · similarity 0.87)
- Action Branching Architectures for Deep Reinforcement Learning (AAAI 2018 · similarity 0.87)
- Confounding Robust Deep Reinforcement Learning: A Causal Approach (NeurIPS 2025 · similarity 0.84)
- Adaptive $Q$-Network: On-the-fly Target Selection for Deep Reinforcement Learning (ICLR 2025 · similarity 0.84)
- Coarse-to-fine Q-Network with Action Sequence for Data-Efficient Reinforcement Learning (NeurIPS 2025 · similarity 0.83)
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.