MMRL4H @ EurIPS 2025

About

Clinical decision making often draws on multiple diverse data sources including, but not limited to, images, text reports, electronic health records (EHR), continuous physiological signals and genomic profiles and yet most AI systems deployed in healthcare remain unimodal. However, integrating diverse medical modalities presents significant challenges including data heterogeneity, data scarcity, missing or asynchronous modalities, variable data quality and the lack of standardized frameworks for aligned representation learning. This workshop aims to bring the EurIPS community together to tackle the problem of fusing these heterogeneous inputs into interpretable, coherent patient representations, aiming to mirror the holistic reasoning of clinicians. We aim to bring together machine learning researchers, clinicians and industry partners dedicated to the theory, methods, and translation of multimodal learning in healthcare.

Goals

In this workshop, we aim to:

Advance methods for learning joint representations from images, text, signals, and genomics.
Investigate foundation-model pretraining at scale on naturally paired modalities.
Address robustness, fairness, and missing-modality issues unique to healthcare fusion.
Foster clinician–ML collaboration and outline translational paths to deployment.

Key Info

Important Dates

Our Call for Papers is now open!

Please note all deadlines are Anywhere on Earth (AOE).

Submission Deadline: October 15, 2025
Acceptance Notification: October 31, 2025
Workshop Date: December 6/7, 2025 (TBC)

Call for Papers

Authors are invited to submit 4-page abstracts on topics relevant to multimodal representation learning in healthcare. These include, but are not limited to, vision-language models for radiology, temporal alignment of multimodal ICU streams, graph and transformer architectures for patient data fusion, cross-modal self-supervised objectives, and multimodal benchmarks with fairness and bias analysis.

Submission

Submission site: via OpenReview
Format: NeurIPS 2025 template
Length: max 4 pages excluding references
Review: Double-blind
Anonymization: Required, ensure that there are no names or affiliations in all parts of the submission including any code.

All accepted papers will be published on the website. Please note that there will be no workshop proceedings (non-archival).

Preliminary Schedule

Time	Session
09:00–09:15	Opening Remarks
09:15–10:00	Speaker Session I
10:00–10:20	Spotlight Session I (3x7mins)
10:20–11:00	Coffee & Poster Session I
11:00–11:45	Speaker Session II
11:45–12:30	Panel - "Translating Multimodal ML to the Bedside"
12:30–14:00	Lunch & Networking
14:00-14:45	Speaker Session III
14:45-15:15	Spotlight Session II (3x7mins)
15:15–15:45	Coffee & Poster Session II
15:45–16:30	Round-table Discussion
16:30-16:50	Open Q&A
16:50-17:00	Closing Remarks & Awards