AI Paper Atlas

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

When More Experts Hurt: Underfitting in Multi-Expert Learning to Defer

Shuqi Liu, Yuzhou Cao, Lei Feng, Bo An, Luke Ong

ICML 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Group-robust classification · 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
replace vanilla CE surrogate with PiCCE-CE, 20 dog expertsmulti-expert CE L2D · ImageNet Dog Expert, 20 expertssystem error: 48.00 -> 41.32Improved
replace vanilla OvA surrogate with PiCCE-OvA, 20 dog expertsmulti-expert OvA L2D · ImageNet Dog Expert, 20 expertssystem error: 66.56 -> 42.25Improved
replace vanilla CE with PiCCE-CE, 20 animal expertsmulti-expert CE L2D · CIFAR-100 Animal Expert, 20 expertssystem error: 21.13 -> 18.11Improved
replace vanilla OvA with PiCCE-OvA, 20 animal expertsmulti-expert OvA L2D · CIFAR-100 Animal Expert, 20 expertssystem error: 22.91 -> 18.63Improved
PiCCE-OvA coverage with 20 animal expertsmulti-expert OvA L2D · CIFAR-100 Animal Expert, 20 expertscoverage: 77.71 -> 86.69Improved
PiCCE-CE with 8 MiceBone expertsmulti-expert CE L2D · MiceBone, 8 expertssystem error: 16.72 -> 13.35Improved
PiCCE-OvA with 8 Chaoyang expertsmulti-expert OvA L2D · Chaoyang, 8 expertssystem error: 2.48 -> 1.02Improved
PiCCE-CE with 2 MiceBone expertsmulti-expert CE L2D · MiceBone, 2 expertssystem error: 15.17 -> 15.23No clear effect
add more experts to vanilla CEmulti-expert CE L2D · CIFAR-100 Animal Expertsystem error: 18.48 -> 21.13Hurt
add more experts to PiCCE-CEPiCCE-CE · CIFAR-100 Animal Expertsystem error: 18.32 -> 18.11Improved
use a hard argmax pick of the confident expert instead of PiCCE (ablation)Pick the Confident Expert loss · multi-expert L2Dsurrogate continuityHurt

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