A cryptography-based approach for movement decoding.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Brain decoders translate neural activity into user intent or actions.
• Traditional decoder training requires simultaneous neural recordings and motor output data (supervised learning).
• Obtaining such paired data is often challenging or infeasible for many applications.

Purpose of the Study:

• To develop a novel movement decoding algorithm that does not require simultaneously measured neural activity and motor outputs.
• To explore an unsupervised approach for brain-computer interface (BCI) training.
• To enable brain decoding in scenarios where supervised data collection is impractical.

Main Methods:

• A distribution-alignment decoding algorithm was developed, leveraging movement statistics.
• The method maps neural activity to motor variables by minimizing Kullback-Leibler divergence between decoder outputs and typical motor output distributions.
• The algorithm was tested on motor cortex neural recordings from non-human primates performing reaching and isometric force tasks.

Main Results:

• The distribution-alignment decoding algorithm achieved performance comparable to supervised decoding methods.
• The approach successfully decoded movement intentions without requiring paired neural and motor data.
• Demonstrated efficacy across different motor tasks (reaching, force production).

Conclusions:

• Distribution-alignment decoding offers a viable alternative to supervised methods, particularly when paired data is unavailable.
• This unsupervised approach significantly broadens the potential applications of brain decoding technologies.
• The findings pave the way for more accessible and versatile brain-computer interfaces.