AI Paper Atlas

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

CLIP-FMoE: Scalable CLIP via Fused Mixture-of-Experts with Enforced Specialization

Luong Tran, Lan-Cuong Nguyen, Huynh Dang Nguyen, Dat Nguyen Cong, Dung D. Le, Van Nguyen

ICLR 2026 · Continual learning › Continual fine-tuning of LLMs · approach: Synthetic data and merging for CLIP

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
fine-tune CLIP with ICCL experts and Fusion Gate vs vanilla fine-tuningCLIP ViT-L/14 · zero-shot classification averagetop-1 accuracy: 66.08 -> 68.65Improved
replace CLIP-MoE with CLIP-FMoECLIP ViT-L/14 · zero-shot classification averagetop-1 accuracy: 66.47 -> 68.65Improved
replace CLIP-MoE with CLIP-FMoECLIP ViT-L/14 · standard retrieval averagerecall: 82.29 -> 82.68Improved
replace TULIP with CLIP-FMoECLIP ViT-L/14 · long-context retrieval averagerecall: 75.9 -> 77.2Improved
vanilla fine-tuning of CLIP on captionsCLIP ViT-L/14 · zero-shot classification averagetop-1 accuracy: 69.62 -> 66.08Hurt
unify experts in MoE instead of using single expert 1 (ablation)CLIP ViT-B/16 · Eurosataccuracy: 43.33 -> 55.47Improved
unify experts in MoE instead of using single expert 1 (ablation)CLIP ViT-B/16 · ImageNet-Aaccuracy: 40.64 -> 40.11Hurt
add Fusion Gate in both stages vs none (ablation)CLIP-FMoE · zero-shot classificationaverage top-1 accuracyImproved
add Fusion Gate in stage 1 as well as stage 2 (ablation)CLIP-FMoE · diverse retrievalaverage recallImproved
use CLIP-FMoE parallel training instead of CLIP-MoE sequentialCLIP ViT-L/14 · training timeminutes: 193.51 -> 53.24Improved
use CLIP-FMoE instead of CLIP-MoECLIP ViT-L/14 · inference costGFLOPs: 112.44 -> 128.22Hurt

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