Research map / Reinforcement learning / Policy gradient and actor-critic
Source Traces for Temporal Difference Learning
Silviu Pitis
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 |
|---|---|---|---|
| Combine source traces with experience replay instead of using one-step TD backups during replay (ablation) | TD(0) w/ ER and TD Source-SR w/ ER · 3D Gridworld | thousands of steps to target error: 58 -> 41; 161 -> 100; 319 -> 189 | Improved |
| Add experience replay to TD(0) (ablation) | TD(0) · 3D Gridworld | thousands of steps to target error: 154 -> 58; 465 -> 161; 979 -> 319 | Improved |
| Use source backups instead of one-step TD backups (ablation) | TD Source-SR vs TD(0) · 3D Gridworld | thousands of steps to target error: 154 -> 58; 465 -> 158; 979 -> 335 | Improved |
| Combine row-wise TD SR updates with column-wise TD Source updates (ablation) | TD Source-SR · Random MRP and 3D Gridworld | source-map approximation error | Improved |
| Anneal lambda linearly from 0.5 to 1 over 25,000 steps (ablation) | TD Source · Random MRP | source-map approximation error | Hurt |
| Increase partial source-trace lookahead through larger n or lambda (ablation) | Partial source learning · Random MRP and 3D Gridworld | value estimation error | Mixed |
| Use annealed rather than fixed learning rates (ablation) | TD(0), TD(lambda), partial source learning, and full source learning · Random MRP and 3D Gridworld | value estimation error | Improved |
| Use triple model learning instead of ideal source traces (ablation) | Source-trace value learning · 3D Gridworld | final error: 1.61 -> 1.45 | Improved |
| Use SR-reward decomposition rather than direct value updates (ablation) | Source-map-based value learning · 3D Gridworld | value estimation error | Mixed |
Closest papers by meaning
- Discerning Temporal Difference Learning (AAAI 2024 · similarity 0.85)
- Temporal Difference Flows (ICML 2025 · similarity 0.83)
- Contrastive Difference Predictive Coding (ICLR 2024 · similarity 0.83)
- Learning Temporal Distances: Contrastive Successor Features Can Provide a Metric Structure for Decision-Making (ICML 2024 · similarity 0.82)
- CaRiNG: Learning Temporal Causal Representation under Non-Invertible Generation Process (ICML 2024 · similarity 0.82)
- TRACE: Contrastive learning for multi-trial time series data in neuroscience (NeurIPS 2025 · similarity 0.82)
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.
