AI Paper Atlas

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

NeurIPS 2023 · Architectures and efficiency › RNNs and normalization · approach: Layer-wise analysis of language models · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
add learned InfoGating to multi-step inverse pretrainingInv · cheetah-run expert overallreturn: 24.7 -> 106.0Improved
use random masks instead of learned masks (ablation)Inv w/ Rand · cheetah-run expert overallreturn: 106.0 -> 10.8Hurt
add dropout / VIB / RCAD regularization instead of InfoGating (ablation)Inv · cheetah-run expert overallreturn: 30.3 / 56.6 / 15.4 vs 24.7Mixed
add InfoGating to inverse model on medium-quality dataInv · cheetah-run medium overallreturn: 19.4 -> 76.0Improved
add InfoGating to inverse model on medium-expert dataInv · cheetah-run medium-expert overallreturn: 14.3 -> 95.3Improved
InfoGating on hard distractors with medium dataInv · cheetah-run medium hardreturn: 10.7 vs 3.0No clear effect
add InfoGating to DrQDrQ · cheetah-run expert overallreturn: 12.9 -> 40.2Improved
add InfoGating to DrQ on hard distractorsDrQ · cheetah-run expert hardreturn: 3.4 -> 5.3No clear effect
add InfoGating to forward dynamics modelfwd · cheetah-run expert overallreturn: 5.9 -> 47.1Improved
fine-tune ImageNet features with InfoGating for BCBC · Kitchen-v3normalized success rate: 12.8 -> 24.1Improved
fine-tune CLIP features with InfoGating for BCBC · Kitchen-v3normalized success rate: 14.8 -> 24.3Improved
gate features instead of pixels (feature InfoGating) (ablation)InfoGating · cheetah-run expert overallreturn: 106.0 -> 69.0Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy