Research map / Reinforcement learning / Visual and representation RL
Spectral Bellman Method: Unifying Representation and Exploration in RL
Ofir Nabati, Bo Dai, Shie Mannor, Guy Tennenholtz
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 SBM representation learning with Thompson Sampling | DQN · Atari ALE | mean HNS: 1.62 -> 2.23 | Improved |
| add SBM representation learning with Thompson Sampling | DQN · Atari Explore | mean HNS: 0.24 -> 0.45 | Improved |
| add SBM with Thompson Sampling | R2D2 · Atari Explore | mean HNS: 0.40 -> 0.61 | Improved |
| add SBM with epsilon-greedy only (no TS) | R2D2 · Atari ALE | mean HNS: 3.21 -> 3.3 | No clear effect |
| add Thompson Sampling exploration on top of SBM features | SBM + DQN · Atari ALE | mean HNS: 1.80 -> 2.23 | Improved |
| use spectral SBM loss instead of naive MSE loss for features (ablation) | SBM + DQN (TS) · Atari ALE | mean HNS: 1.82 -> 2.23 | Improved |
| learn representation over a distribution of value functions instead of only the current policy (ablation) | SBM + DQN (TS) · Atari ALE | mean HNS: 1.73 -> 2.23 | Improved |
| add orthogonality regularizer (ablation) | SBM + DQN (TS) · Atari ALE | mean HNS: 2.11 -> 2.23 | Improved |
| use SBM + TS instead of Online PVN + TS | DQN · Atari Explore | mean HNS: 0.31 -> 0.45 | Improved |
| add SBM with TS | DQN · precision-control games e.g. Breakout | score | Hurt |
| add SBM | DQN · training compute | compute time per 1M steps: ~20% increase | Hurt |
Closest papers by meaning
- Inverting the Bellman Equation: From $Q$-Values to World Models (NeurIPS 2026 · similarity 0.90)
- Gradual Transition from Bellman Optimality Operator to Bellman Operator in Online Reinforcement Learning (ICML 2025 · similarity 0.89)
- Enhancing Value Function Estimation through First-Order State-Action Dynamics in Offline Reinforcement Learning (ICML 2024 · similarity 0.89)
- Computationally Efficient RL under Linear Bellman Completeness for Deterministic Dynamics (ICLR 2025 · similarity 0.88)
- Diffusion Spectral Representation for Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- Parameterized Projected Bellman Operator (AAAI 2024 · 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.
