AI Paper Atlas

Research map / Vision-language and multimodal / Cross-modal learning

PETRI: Learning Unified Cell Embeddings from Unpaired Modalities via Early-Fusion Joint Reconstruction

Ryan W Conrad, Ethan Weinberger, Saradha Venkatachalapathy, Yuwen Chen, Darshini Shah, Bay Johnson, Max R Salick, Vaishaali Natarajan, Emily Fox

ICLR 2026 · Vision-language and multimodal › Cross-modal learning · approach: Multi-modality image fusion

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 multimodal cell documents (early fusion) to unimodal transcriptomics modelPETRI · HepG2StringDB edge TPR@5%FPR: 0.144 -> 0.242Improved
add multimodal cell documents (early fusion) to unimodal image modelPETRI · HepG2StringDB edge TPR@5%FPR: 0.116 -> 0.242Improved
use PETRI instead of ViT MAEPETRI · HepG2StringDB ROC AUC: 0.549 -> 0.628Improved
use early-fusion PETRI instead of late-fusion max cosine aggregationPETRI · HepG2StringDB edge TPR@5%FPR: 0.219 -> 0.242Improved
use PETRI instead of CLIP early-fusion baselinePETRI · Perturb-MultiGuide Consistency: 0.000 -> 0.208Improved
group cells by perturbation instead of randomly (permuted data) (ablation)PETRI · Perturb-MultiGuide Consistency: 0.163 -> 0.208Improved
group cells by perturbation instead of randomly (permuted data) (ablation)PETRI · HepG2StringDB edge TPR@5%FPR: 0.255 -> 0.242Hurt
group cells by perturbation instead of randomly (permuted data) (ablation)PETRI · HepG2Guide Consistency: 0.274 -> 0.278No clear effect
use encoder outputs before the multimodal set transformer instead of after layer 4 (ablation)PETRI · HepG2Guide Consistency: 0.221 -> 0.260Improved
use encoder outputs before the multimodal set transformer instead of after layer 4 (ablation)PETRI · HepG2StringDB edge TPR@5%FPR: 0.186 -> 0.242Improved
apply robust center scaling and PCA whitening to embeddings (ablation)PETRI · HepG2StringDB edge TPR@5%FPR: 0.031 -> 0.242Improved
vary latent tokens per cell and set size (ablation)PETRI · HepG2StringDB and Guide ConsistencyNo clear effect

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