MoPET method uses mixture-of-experts for medical image classification

Researchers have developed MoPET, a novel parameter-efficient fine-tuning (PEFT) method that utilizes a mixture-of-experts (MoE) approach for medical image classification. MoPET addresses the issue of negative transfer in PEFT by employing a learned sparse router to direct inputs through a subset of low-rank experts within a frozen foundation model. This allows for shared capacity across datasets while mitigating gradient conflicts. Evaluations on the MedMNIST benchmark demonstrated MoPET's effectiveness, showing improved accuracy over isolated PEFT adapters and enhanced performance on data-constrained clinical targets when co-trained with auxiliary datasets. AI

IMPACT Introduces a novel approach to improve efficiency and accuracy in medical AI applications.

RANK_REASON Research paper detailing a new method for medical image classification. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• Hugging Face
• medical image classification
• MedMNIST
• mixture-of-experts
• parameter-efficient fine-tuning
• Sebastian Doerrich

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →