PaperarXiv:2611.13758
Mixed Signals: Weight-Space Programming for EEG and ECG
arXiv preprint (cs.LG; cross-list eess.SP)
(opens in a new tab)Varam, Dara et al., "Mixed Signals: Weight-Space Programming for EEG and ECG", arXiv:2611.13758, CC BY 4.0
TL;DR
EPFL (Atienza lab): a frontier LLM reads and writes a small biosignal network's weights by reasoning, with no gradient steps. From 5 training subjects, Claude Opus 5.5 wrote all 2,730 weights of an EEG eyes-open/closed classifier: balanced accuracy 0.834 on 65 unseen subjects, vs 0.704 for gradient descent on the same 5 and 0.806 on 30. It also found planted faults; it could not improve already well-trained models.
Abstract
Weight-space programming allows language models to read a trained neural network's weights and edit them. This opens the door to training'' by reasoning rather than through gradient descent. Prior work has demonstrated its functionality and promise for game-playing networks and fixing specific failures in language models, even extending to diffusion models. In this work, we present its applicability to models trained on biosignal data, specifically EEG and ECG data. We show that frontier models are not only able to debug specific faults planted in biosignal-based models without being told what these faults are, but are even capable of writing all the weights of a network from scratch, given only the architecture, the task, and a minimal set of training subjects. In particular, in one set of experiments, a language model was able to match standard gradient-descent training using 6× fewer training subjects, through reasoning alone and without any gradient descent step. Networks written by Claude Opus 5.5 also scored higher than classic hand-crafted EEG pipelines (alpha band power, and common spatial patterns with linear discriminant analysis) when built from the same 5 subjects. We also highlight some of the limitations, which include the inability of language models to improve already well-trained classification models, and their incapacity to reliably recognize, from the weights alone, the task a network was trained on. Overall, our results indicate that weight-space programming can be useful for small networks, such as those deployed on edge devices and wearables, but the concept still requires further investigation before it can be used in practice.
Verbatim from arXiv (opens in a new tab) · arXiv metadata (abstracts are released under CC0)
Results on ChatPPG leaderboards
All leaderboardsNo results from this paper on ChatPPG leaderboards yet. Browse the leaderboards.
Tasks & signals
Signals
Keywords
Datasets used
How it works
Cite this paper
Varam, D., Najafi, T. A., & Atienza, D. (2026). Mixed Signals: Weight-Space Programming for EEG and ECG. arXiv preprint. https://arxiv.org/abs/2611.13758
Varam, Dara, Taraneh Aminosharieh Najafi, and David Atienza. 2026. "Mixed Signals: Weight-Space Programming for EEG and ECG." arXiv preprint. https://arxiv.org/abs/2611.13758.
Varam, D., Najafi, T. A. and Atienza, D. (2026) 'Mixed Signals: Weight-Space Programming for EEG and ECG', arXiv preprint. Available at: https://arxiv.org/abs/2611.13758
Varam, Dara, et al. "Mixed Signals: Weight-Space Programming for EEG and ECG." arXiv preprint, 2026. https://arxiv.org/abs/2611.13758.
D. Varam, T. A. Najafi, and D. Atienza, "Mixed Signals: Weight-Space Programming for EEG and ECG," arXiv preprint, 2026, [Online]. Available: https://arxiv.org/abs/2611.13758.
@misc{varam2026mixed,
title = {{Mixed Signals: Weight-Space Programming for EEG and ECG}},
author = {Varam, Dara and Najafi, Taraneh Aminosharieh and Atienza, David},
year = {2026},
howpublished = {arXiv preprint},
eprint = {2611.13758},
archivePrefix = {arXiv}
}Added to the ChatPPG research hub
