Language Models Write EEG and ECG Network Weights Through Reasoning
The paper “Mixed Signals” applies weight-space programming to neural networks trained on EEG and ECG data, enabling language models to read, debug, and rewrite network weights through reasoning rather than gradient descent. In one experiment, a model matched standard training with six times fewer subjects, while networks written by Claude Opus 5.5 outperformed traditional EEG pipelines built from the same five subjects. The authors also report that the approach cannot reliably improve already well-trained classifiers or infer a network's task from weights alone.
The results suggest that language models may directly construct small neural networks for biosignal tasks, including edge and wearable applications, but the findings come from one preprint and require further validation.
Computer Science > Machine Learning
arXiv:2611.13758 (cs)
[Submitted on 10 Nov 2026]
Title:Mixed Signals: Weight-Space Programming for EEG and ECG
Authors:Dara Varam, Taraneh Aminosharieh Najafi, David Atienza
View a PDF of the paper titled Mixed Signals: Weight-Space Programming for EEG and ECG, by Dara Varam and 2 other authors