AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

Understanding Transfer Learning of RNA Foundation Models on Downstream Tasks

Yuan Li, Heng Yang, Renzhi Chen, Ke Li

ICML 2026 · Vision-language and multimodal › Human and embodied modeling · approach: Medical vision-language models

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 from pretrained weights instead of frozen last-layer probingRNA FMs · Archive2, SSItask metricImproved
use frozen pretrained features instead of one-hot encodingOmniGenome-52M, PlantRNA-FM · Archive2, bpRNAF1Improved
use frozen pretrained features instead of one-hot encodingRNA FMs · ncRNAaccuracyHurt
probe intermediate layers (L/4, L/2, 3L/4) instead of the final layerRNA FMs · downstream taskstask metricMixed
start from pretrained weights instead of random initialization (PT-FT vs RI-T) (ablation)RNA FMs · nine downstream taskstask metricImproved
pretrained init vs random init (ablation)RNA-BERT (0.5M params) · SSP tasksF1No clear effect
pretrained init vs random init (ablation)mRNA-FM (239M params) · all evaluated taskstask metricNo clear effect
pretrained init vs random init lowers SAM-style sharpness (ablation)RNA FMs · Archive2, ncRNA, mRNA-oodSAM-style flatness: lower in all 15 tested model-task pairsImproved
pretrained init vs random init speeds convergence (ablation)RNA FMs · seven tasks (training loss)training lossImproved

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