AI Paper Atlas

Research map / Learning theory and algorithms / Submodular and game algorithms

Auxiliary Learning as an Asymmetric Bargaining Game

Aviv Shamsian, Aviv Navon, Neta Glazer, Kenji Kawaguchi, Gal Chechik, Ethan Fetaya

ICML 2023 · Learning theory and algorithms › Submodular and game algorithms · approach: Equilibrium solvers for adversarial games · 0 citations (OpenAlex)

Read the paper

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 asymmetric bargaining with dynamically learned task preferencesAuxiNash · NYUv2mIoU (semantic segmentation): 38.30 -> 40.79Improved
use asymmetric bargaining with dynamically learned task preferencesAuxiNash · CityscapesmIoU (semantic segmentation): 48.64 -> 52.52Improved
use asymmetric bargaining with dynamically learned task preferencesAuxiNash · CIFAR10-SSL-5Ktest accuracy: 83.17 -> 87.01Improved
use asymmetric bargaining with dynamically learned task preferencesAuxiNash · Speech Commands SC-1000test accuracy: 97.0 ± 0.1 -> 97.2 ± 0.1Improved
optimize preferences using a 10% auxiliary set rather than the entire training set (ablation)AuxiNash · CityscapesmIoU: 52.52 -> 51.81Hurt
optimize preferences using a 10% auxiliary set rather than the entire training set (ablation)AuxiNash · NYUv2mIoU: 40.79 -> 38.94Hurt
fix the main-task preference at 0.7 rather than learn it dynamically (ablation)AuxiNash · NYUv2mIoU: 40.79 -> 41.36Improved
fix the main-task preference at 0.8 rather than learn it dynamically (ablation)AuxiNash · NYUv2mIoU: 40.79 -> 41.27Improved
fix the main-task preference at 0.9 rather than learn it dynamically (ablation)AuxiNash · NYUv2mIoU: 40.79 -> 41.35Improved
change outer preference learning rate from 5e-3 to 1e-2 (ablation)AuxiNash · NYUv2mIoU: 40.79 -> 40.75No clear effect
change outer preference learning rate from 5e-3 to 1e-3 (ablation)AuxiNash · NYUv2mIoU: 40.79 -> 41.03No 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