AI Paper Atlas

Research map / Optimization / Optimizers

Provable Multi-instance Deep AUC Maximization with Stochastic Pooling

Dixian Zhu, Bokun Wang, Zhi Chen, Yaxing Wang, Milan Sonka, Xiaodong Wu, Tianbao Yang

ICML 2023 · Optimization › Optimizers · approach: Surrogate losses and robust solvers · 1 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 variance-reduced smoothed-max pooling instead of naive mini-batch smoothed-max poolingDAM on Breast Cancer · Breast CancerAUC: 0.725(0.184) -> 0.875(0.137)Improved
Use variance-reduced attention pooling instead of naive mini-batch attention poolingDAM on Breast Cancer · Breast CancerAUC: 0.875(0.112) -> 0.95(0.1)Improved
Use variance-reduced smoothed-max pooling instead of naive mini-batch smoothed-max poolingDAM on Colon Ade. · Colon Ade.AUC: 0.905(0.01) -> 0.91(0.02)Improved
Use variance-reduced attention pooling instead of naive mini-batch attention poolingDAM on Colon Ade. · Colon Ade.AUC: 0.882(0.029) -> 0.893(0.08)Improved
Use variance-reduced smoothed-max pooling instead of naive mini-batch smoothed-max poolingDAM on PDGM · PDGMAUC: 0.659(0.058) -> 0.669(0.032)Improved
Use variance-reduced attention pooling instead of naive mini-batch attention poolingDAM on PDGM · PDGMAUC: 0.624(0.112) -> 0.635(0.052)Improved
Use variance-reduced smoothed-max pooling instead of naive mini-batch smoothed-max poolingDAM on OCT · OCTAUC: 0.829(0.008) -> 0.848(0.01)Improved
Use variance-reduced attention pooling instead of naive mini-batch attention poolingDAM on OCT · OCTAUC: 0.842(0.013) -> 0.843(0.012)No clear effect
Increase instance-batch size B (ablation)MIDAM · MUSK2training convergence speedImproved
Use instance-batch size B=4 rather than all instances per bag (ablation)MIDAM · MUSK2training AUCNo clear effect
Allocate fixed total budget with hat B=4 and B=64 (ablation)MIDAM · Colon Ade.training convergence speedImproved
Use stochastic instance sampling rather than all-instance poolingMIDAM on tabular benchmarks · Testing performanceAUCMixed

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