AI Paper Atlas

Research map / Architectures and efficiency / Mixture of experts

LiME: Lightweight Mixture of Experts for Efficient Multimodal Multi-task Learning

Md Kowsher, Haris Mansoor, Nusrat Jahan Prottasha, Ozlem Garibay, Victor Zhu, Zhengping Ji, Chen Chen

ICML 2026 · Architectures and efficiency › Mixture of experts · approach: Mixture-of-experts LoRA · 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 separate per-expert adapters with a shared PEFT module plus lightweight expert vectorsLiME-LoRA vs MoELoRA · GLUE efficiencytrainable parameters: 1.97M -> 0.52MImproved
replace separate per-expert adapters with a shared PEFT module plus expert vectorsLiME-DoRA vs MoE-DoRA · GLUE efficiencytraining time (min): 50.2 -> 35.5Improved
add LiME modulation and zero-parameter routing to base PEFT methodsLoRA, DoRA, LoRA-FA, SliceFine, Prompt Tuning · MMT-47 categoriesaverage score: Vision Benchmark LoRA 77.02 -> LiME-LoRA 78.01Improved
use zero-parameter routing instead of a learned router (ablation)LiME · text and vision tasksaccuracyNo clear effect
choose routing feature dimensions (leading, central, trailing, random E) (ablation)LiME routing · Figure 4aaccuracyNo clear effect
mix frozen and PEFT outputs for routing (gamma_r around 0.7) instead of either alone (ablation)LiME routing · text and vision tasksaccuracyImproved
use Auto Top-K relative threshold theta 0.7 instead of fixed top-k (ablation)LiME · GLUE SST-2accuracy: 92.08 -> 92.43 (k=3 vs theta=0.7)Improved
activate all four experts (fixed k = 4) instead of k = 3 (ablation)LiME fixed Top-K · GLUE SST-2accuracy: 92.08 -> 91.94Hurt
set relative threshold too strict (theta 0.8 instead of 0.7) (ablation)LiME Auto Top-K · GLUE STS-Baccuracy: 81.87 -> 81.54Hurt
add load balancing losses (coefficient about 1e-2 to 0.1) versus none (ablation)LiME · GLUEaccuracyImproved
over-strong load balancing (coefficient above 1.0) (ablation)LiME · GLUEaccuracyHurt
increase number of experts from 1 to 4-5, then beyond 6 (ablation)LiME · GLUEaccuracyMixed

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