← Founder Notes
Archive

The night is being scored by model

Yethikrishna ROriginal on Threads

the night is being scored by model: bcgnet learned from 596,000 hours of sleep recordings to stage sleep from a sensor under the pillow, oxford’s ai scores sleep disruption without labels in older adults, and beijing’s sleep interpreter reads which memories replay in eeg across strangers.

sleep is becoming a readable file.

Context

BCGNet is an AI model trained on about 600 thousand hours of sleep data for an under-pillow contactless ballistocardiogram device, per its medRxiv preprint. The paper frames polysomnography as costly and hard to scale.

Oxford engineering announced on Aug 12, 2026 an AI tool for automated sleep monitoring. A Neuron paper (Sep 2, 2026) describes a Sleep Interpreter that decodes spontaneous memory replay from sleep EEG using neural contrastive learning.

How it compares

The under-pillow BCGNet, the Oxford sleep tool and the Neuron memory replay paper match their sources. The exact 596,000 hours, 'older adults', 'without labels' and 'across strangers' are not seen in the sources read, so unsupported here, not refuted. The sources read give about 600 thousand hours.

Related work

Watch next

  • Check the Oxford release for the population the tool was tested on.

Sources

  1. medRxiv, BCGNet preprintmedrxiv.org
  2. Oxford Engineering, Aug 12, 2026eng.ox.ac.uk
  3. Neuron, Sep 2, 2026cell.com

Provenance

The note above is reproduced unedited from the original post, first published on Threads on 2 October 2026 at 11:19 IST. Sources are the papers and datasets the note draws on.

View the original post
Embed this note
<iframe src="https://founder.myndlabs.tech/notes/embed/the-night-is-being-scored-by-model-bcgnet-Dd-xZuNjamj" width="480" height="420" style="border:0;max-width:100%" loading="lazy" title="The night is being scored by model"></iframe>

More notes