Research map / Reinforcement learning / Visual and representation RL
Interaction-Grounded Learning with Action-Inclusive Feedback
Tengyang Xie, Akanksha Saran, Dylan J Foster, Lekan P Molu, Ida Momennejad, Nan Jiang, Paul Mineiro, John Langford
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 |
|---|---|---|---|
| Use action-specific contrastive reward decoding and symmetry breaking instead of the full-conditional-independence IGL method | IGL · MNIST (action-inclusive feedback) | policy accuracy (%): 9.18 ± 2.10 -> 83.63 ± 1.25 | Improved |
| Use action-specific contrastive reward decoding and symmetry breaking instead of the full-conditional-independence IGL method | IGL · Simulated BCI (1% noise) | policy accuracy: 32.60±0.24 -> 89.10 ± 3.58 | Improved |
| Use action-specific contrastive reward decoding and symmetry breaking instead of the full-conditional-independence IGL method | IGL · Simulated BCI (5% noise) | policy accuracy: 33.75±0.29 -> 76.60 ± 4.09 | Improved |
| Use action-specific contrastive reward decoding and symmetry breaking instead of the full-conditional-independence IGL method | IGL · Simulated BCI (10% noise) | policy accuracy: 33.33 ± 0.29 -> 64.25 ± 4.16 | Improved |
| Use action-specific contrastive reward decoding and symmetry breaking instead of the full-conditional-independence IGL method | IGL · OpenML CC-18 (K≥3) | policy accuracy (%): 15.65±2.30 -> 35.74±1.45 | Improved |
| Use action-specific contrastive reward decoding and symmetry breaking instead of the full-conditional-independence IGL method | IGL · OpenML CC-18 (K≥3, N≥70000) | policy accuracy (%): 11.91±1.80 -> 50.11±2.98 | Improved |
| Use AI-IGL rather than IGL (full CI) | IGL · MNIST (action-exclusive feedback) | policy accuracy (%): 86.13 ± 0.52 -> 86.14 ± 0.94 | No clear effect |
| Increase simulated BCI feedback noise from 1% to 10% | AI-IGL · Simulated BCI | policy accuracy: 89.10 ± 3.58 -> 64.25 ± 4.16 | Hurt |
| Use AI-IGL instead of contextual bandits with exact reward | Contextual bandits · OpenML CC-18 (K≥3) | policy accuracy (%): 57.98±5.65 -> 35.74±1.45 | Hurt |
Closest papers by meaning
- An Information Theoretic Approach to Interaction-Grounded Learning (ICML 2024 · similarity 0.88)
- Provably Efficient Interactive-Grounded Learning with Personalized Reward (NeurIPS 2024 · similarity 0.86)
- Towards Better Robot Learners: Leveraging Implicit and Explicit Human Feedback Together in Human Robot Interactions (AAAI 2025 · similarity 0.84)
- Ground-Compose-Reinforce: Grounding Language in Agentic Behaviours using Limited Data (NeurIPS 2025 · similarity 0.84)
- When Does Interaction Help? Representational Tradeoffs in Value-Based Imitation Learning (NeurIPS 2026 · similarity 0.84)
- Provable Interactive Learning with Hindsight Instruction Feedback (ICML 2024 · 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.
