AI Paper Atlas

Research map / Continual learning / Continual fine-tuning of LLMs

Mixing Expertise with Confidence: A Mixture of Experts Framework for Robust Multi-Modal Continual Learning

Md Abdullah Al Forhad, Yuansheng Zhu, Abhinab Acharya, Xumin Liu, Qi Yu, Weishi Shi

ICML 2026 · Continual learning › Continual fine-tuning of LLMs · approach: Mixture-of-experts for multimodal continual tuning

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 adaptive threshold temperature scaling to independently trained experts (ablation)Expert mixture · CIFAR-100 10 stepstask ID accuracyImproved
add Mahalanobis-distance weighting only (no temperature) (ablation)Expert mixture · CIFAR-100 10 stepstask ID accuracyImproved
combine temperature scaling and MD weighting (ablation)Expert mixture · CIFAR-100 10 stepstask ID accuracyImproved
apply temperature scaling to task-ID detection (ablation)Expert mixture · CIFAR-100 10 stepsper-task detectionMixed
scale temperature on shortcut-activated samples (ablation)Expert · CIFAR-100 10 stepsmean predictive variance: 0.007-0.010 range -> 0.0005-0.0015 rangeImproved
remove shared parameters and use confidence-based expert communicationOurs vs MoE-Adapters · CIFAR-100 10 stepslast accuracy: 77.52 -> 80.6Improved
use multimodal expert mixture instead of RAIL on subpopulation shiftOurs vs RAIL · BREEDS Entity-13 (5 steps)All accuracy: 81.06 -> 87.08Improved
use same mixture on 50-step CIFAR-100Ours vs RAIL · CIFAR-100 50 stepaverage accuracy: 85.11 -> 85.68Improved

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