Research map / Architectures and efficiency / RNNs and normalization
Ignorance is Bliss: Robust Control via Information Gating
Manan Tomar, Riashat Islam, Matthew E. Taylor, Sergey Levine, Philip Bachman
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 learned InfoGating to multi-step inverse pretraining | Inv · cheetah-run expert overall | return: 24.7 -> 106.0 | Improved |
| use random masks instead of learned masks (ablation) | Inv w/ Rand · cheetah-run expert overall | return: 106.0 -> 10.8 | Hurt |
| add dropout / VIB / RCAD regularization instead of InfoGating (ablation) | Inv · cheetah-run expert overall | return: 30.3 / 56.6 / 15.4 vs 24.7 | Mixed |
| add InfoGating to inverse model on medium-quality data | Inv · cheetah-run medium overall | return: 19.4 -> 76.0 | Improved |
| add InfoGating to inverse model on medium-expert data | Inv · cheetah-run medium-expert overall | return: 14.3 -> 95.3 | Improved |
| InfoGating on hard distractors with medium data | Inv · cheetah-run medium hard | return: 10.7 vs 3.0 | No clear effect |
| add InfoGating to DrQ | DrQ · cheetah-run expert overall | return: 12.9 -> 40.2 | Improved |
| add InfoGating to DrQ on hard distractors | DrQ · cheetah-run expert hard | return: 3.4 -> 5.3 | No clear effect |
| add InfoGating to forward dynamics model | fwd · cheetah-run expert overall | return: 5.9 -> 47.1 | Improved |
| fine-tune ImageNet features with InfoGating for BC | BC · Kitchen-v3 | normalized success rate: 12.8 -> 24.1 | Improved |
| fine-tune CLIP features with InfoGating for BC | BC · Kitchen-v3 | normalized success rate: 14.8 -> 24.3 | Improved |
| gate features instead of pixels (feature InfoGating) (ablation) | InfoGating · cheetah-run expert overall | return: 106.0 -> 69.0 | Hurt |
Closest papers by meaning
- Forget Sharpness: Perturbed Forgetting of Model Biases Within SAM Dynamics (ICML 2024 · similarity 0.84)
- InfoSAM: Fine-Tuning the Segment Anything Model from An Information-Theoretic Perspective (ICML 2025 · similarity 0.84)
- Focus On What Matters: Separated Models For Visual-Based RL Generalization (NeurIPS 2024 · similarity 0.83)
- Pathways of Visual Information Flow in Vision-Language Models (NeurIPS 2026 · similarity 0.83)
- Seeing through Uncertainty: Robust Task-Oriented Optimization in Visual Navigation (NeurIPS 2025 · similarity 0.83)
- DMC-VB: A Benchmark for Representation Learning for Control with Visual Distractors (NeurIPS 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.
