AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Drop-Bottleneck: Learning Discrete Compressed Representation for Noise-Robust Exploration

Jaekyeom Kim, Minjung Kim, Dongyeon Woo, Gunhee Kim

ICLR 2021 · Label-efficient and robust learning › Recognition and long-tail · approach: Contrastive representation theory · 8 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
Use Drop-Bottleneck rather than VIB in the exploration methodPPO + Ours · VizDoom MyWayHome sparse and very sparse tasks under noiseaverage episodic sum of rewardsImproved
Use Drop-Bottleneck rather than VIB in episodic explorationPPO + Ours · DMLab, all six Sparse and Very Sparse noise settingsaverage episodic sum of rewards: 28.8 vs 23.5; 29.1 vs 25.4; 26.9 vs 22.3Improved
Add Drop-Bottleneck feature dropping (ablation)PPO + Ours exploration · DMLab Very Sparse, Image Actionaverage episodic sum of rewards: 14.9 ± 1.17 -> 28.8 ± 2.1Improved
Add Drop-Bottleneck feature dropping (ablation)PPO + Ours exploration · DMLab Very Sparse, Noiseaverage episodic sum of rewards: 11.7 ± 1.3 -> 29.1 ± 2.0Improved
Add Drop-Bottleneck feature dropping (ablation)PPO + Ours exploration · DMLab Very Sparse, Noise Actionaverage episodic sum of rewards: 10.3 ± 1.3 -> 26.9 ± 1.7Improved
Add Drop-Bottleneck feature dropping (ablation)PPO + Ours exploration · DMLab Very Sparse, Originalaverage episodic sum of rewards: 33.0 ± 1.8 -> 42.7 ± 2.3Improved
Use DB with β=0.01 instead of VIBInception-ResNet-v2 · ImageNet targeted ℓ2 attackattack success rate (%): 100.0 -> 18.5Improved
Use DB with β=0.003162 instead of VIBInception-ResNet-v2 · ImageNet targeted ℓ∞ attackattack success rate (%): 97.5 -> 1.5Improved
Use deterministic DB feature selection instead of univariate MI feature selectionInception-ResNet-v2 · ImageNet targeted ℓ2 attack with 8 featuresattack success rate (%): 99.5 -> 20.0Improved
Use deterministic DB feature selection instead of univariate MI feature selectionInception-ResNet-v2 · ImageNet targeted ℓ∞ attack with 8 featuresattack success rate (%): 99.5 -> 2.0Improved
Increase DB compression strength βDB image classifier · ImageNet validationclassification accuracyMixed
Use the deterministic DB representation instead of its stochastic representationDrop-Bottleneck · ImageNet validationclassification accuracyNo clear effect

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