New people, next day: which part of a pretrained model should you update?

For scale: the core matrix's frozen LaBraM, on the same people and folds but read out by a closed-form ridge head on standardized features, scored 64.6% (core matrix release, experiments.json). The head-only arm here is a short five-epoch head on unscaled features — not the best a frozen encoder can do — and the two are not a paired comparison.

On macro F1 the gains are smaller, and LoRA helps fewer people over head only than it does on balanced accuracy: the metric matters.

What each rule costs

LoRA trains far fewer weights than the last block but took longer here: its adapters sit in all twelve blocks, so gradients flow back through most of the encoder, and this implementation is an effective-weight parametrization, not an optimized low-rank kernel. Fewer trainable weights do not by themselves mean less time or memory.

Promised on the roadmap and not delivered by this batch: peak memory (only lower-bound samples were recorded, so none is published) and a comparison at a fixed computation budget (every rule trained for the same five epochs, not the same compute).

Two recording days: run and replayed, held

BNCI2015-001: 12 people recorded on two days. Each person's model is trained on day A, calibrated with the first 0, 10, 20 or 40 labeled trials of day B under the same three update rules, and tested on the same later day-B trials. Unlike the result above, this design uses the test person's own labels: their day-A trials to train and, except at the zero budget, their first day-B trials to calibrate. It ran on 22 September and passed an independent replay. The source's editorial hold stands — its catalogue licence is CC BY-NC-ND and its description names no ethics approval — so no figure from it is published.

Same person, later sessions: status only

Stieger longitudinal BCI: one person's 11 sessions were acquired, not the full release. What is published is that the loader and a chronological split work on it — earlier sessions train, later sessions test. With one person, every score would be that person's, so none is published; the source is listed as status only in the release of 27 September.

Four questions kept apart

A new person, a new day, a new dataset and a new device are different questions, answered on different data. A task adapter trained on several people is not evidence that it transfers across devices.

A related question now has a measured answer, for two classical baselines on other data: the same person’s next session, with and without a few labeled trials from it. It is on How much calibration data does an EEG decoder need?.

Adapter size

The LaBraM base encoder has 5,819,936 parameters. Rank-4 updates to its twelve fused QKV weights add 38,400 trainable parameters (0.66%); with the 402-parameter head that is the 38,802 above. The adapter tensors take 150 KiB in float32 before metadata.

Engineering checks passed before any real-EEG run: exact adapter save and reload; exact merged-weight equivalence; finite, nonzero adapter gradients in every block; frozen encoder weights unchanged after updates; identical output at zero adapter initialization.

These are parameter counts — not accuracy, speed, memory-saving or privacy claims.

Not the first to ask

Other groups already benchmark EEG adapters: OpenEEGBench covers parameter-efficient fine-tuning of EEG foundation models; a systematic study of test-time adaptation reports inconsistent gains under distribution shift; a recent preprint compares self-supervised adaptation under fixed computation budgets. Our emphasis is the calibration cost, the cases that get worse, and whether recordings are compatible at all.

Data source: adaptation-update.json · schema bci-report-adaptation-update-v1 · generated 2026-10-01.

Data source: evidence-update.json · the adapter counts and checks, from the engineering check released on 22 September.