ResearchNovember 11, 2026
-26
š§
CARMEN reads ten vital signals at once
The CARMEN encoder handles ten cardiorespiratory signals and outperforms specialized models.
#foundation models#healthcare#physiological signals#pretraining#arXiv

š„ What happened
A Korean team unveiled CARMEN, a foundation model pretrained on ten cardiorespiratory waveforms ā 2.27 million modality-hours, no labels. It handles any combination of co-recorded signals from 10 seconds to 5 minutes.
š” Why it matters
Prior physiological models leaned almost entirely on ECG and PPG. CARMEN beat supervised models on 12 of 15 tasks and three other foundation models on 14 to 17 of 19 ā none outperformed it. Without its LSCNorm conditioning, intraoperative hypotension AUROC dropped from 0.894 to 0.780.
ā” Our take
One encoder for all vital signs could reshape OR and ICU monitoring. Anyone still building ECG-only models is either brave or badly informed.
The title, summary and analysis of this item were produced automatically by an AI system and have not been editorially reviewed. They may contain errors, bias or omissions ā when in doubt, read the linked original source.
Deep Dives & Similar Intelligence

Quantized Agents Get a Teacher
ResearchNovember 10, 2026

Stanford's 31x Faster Text Generation
ResearchNovember 09, 2026

Rewriting boosts LLMs by 27 percent
ResearchNovember 11, 2026

AI Agent Cracks Hex Mesh Puzzle
ResearchNovember 07, 2026

Diffusion LLMs learn from their own mistakes
ResearchNovember 07, 2026