Intracortical motor decoder adapts across sessions with fixed network weights

Association profile conditioning in a set-temporal transformer for cross-session intracortical motor decoding

Artificial Intelligence

Summary

Intracortical devices that decode brain signals to control movements usually lose accuracy over time because the neurons they record from change or behave differently. The authors present a method called APST that adjusts to these changes without retraining the whole model, by analyzing how each neuron’s activity relates to movement during a few short calibration periods. This approach uses a transformer model that can handle varying numbers and orders of neurons and improves accuracy in predicting movement velocity across multiple sessions. APST outperforms methods that rely on neural data alone or require fine-tuning with the same calibration data.

What this means in practice

  • For brain computer interface teams: Improve decoding accuracy in intracortical motor prosthetics across recording sessions without retraining the entire model.$Commercial implications: Enables more reliable and user-friendly brain-controlled devices for people with motor disabilities by reducing calibration time and maintaining performance.
  • For neural signal processing engineers: Deploy streaming decoding models that flexibly adapt to different neuron sets using association profiles derived from limited calibration data.

Authors

Xinyuan Zhang, Handong Mo, Pengfei Wen, Shuang Liang, Jichang Yang, Yan Zeng, Zhongrui Wang, Han Wang

Abstract

Intracortical motor decoders degrade across sessions because the set of recorded units changes and persisting units can alter how their firing relates to behavior. Most existing methods update network weights on each new session or rely on unlabeled activity, which does not directly reveal such changes. We present APST, an Association Profile-conditioned Set-Temporal transformer that adapts to new sessions with all network weights frozen. From a few labeled calibration trials, APST summarizes how each unit's firing relates to behavior in a four-dimensional association profile computed in closed form. The profiles condition a set-attention encoder that accepts any number and order of units, followed by a causal transformer for streaming decoding. On held-out DANDI688 sessions from two monkeys, APST reaches velocity $R^2$ of $0.78$ and $0.81$, versus $0.40$ and $0.58$ for a variant that uses neural activity alone, and matches or exceeds an RNN fine-tuned on the same trials. On FALCON private held-out evaluation, it attains $R^2$ of $0.65$, $0.42$, and $0.44$ on M1, M2, and H1.
View PDFOpen arXiv