Neural Sympathy: Towards Interfaces Compatible with Neural Plasticity

Brain-machine interfaces (BMIs) enable a direct mapping from neural activity to interactions with the world. They have clinically validated applications for patients with severe loss of motor function, but also offer valuable scientific tools for probing the neural mechanisms of how actions are learned, planned, and executed. A major challenge for BMIs is the changeable environment in which they operate: as a result of ongoing experience, the brain is constantly reconfiguring its internal neural representations of tasks. To overcome the need for daily BMI recalibrations, we need to understand the interaction between BMI usage and the reorganisation of plastic neural populations. In this thesis, we describe the design of an optical BMI based on 2-photon imaging of CA1 hippocampus in mice. Populations in CA1 are highly plastic and are critical for spatial navigation, providing an excellent testbed for interpreting changes in neural activity induced by BMI usage. Closed-loop experiments with this BMI reveal multiple representations of the same environment, which are invoked depending on whether BMI or physical locomotion is used to control navigation. We show that BMI usage induces changes in hippocampal representations that persist even when animals stop using their BMI. These changes appear unique to circumstances in which the animal has a causal influence over its travel, and hint at the incorporation of navigational agency into hippocampal maps of space. We use representational drift, the ongoing changes in representation of familiar environments that take place even after tasks are mastered, to illustrate the eventual degradation of fixed decoder readouts. We show through modelling that, while a variety of drift formulations can degrade fixed decoders at similar rates, changes in representation that are statistically heavy-tailed are far easier for unsupervised adaptive decoders to detect and compensate for. Existing, in vivo observations of drift from both visual cortex and posterior parietal cortex, as well as our own recordings from CA1, exhibit a preference for such heavy-tailed changes in representation. We take this as a subtle hint that downstream readouts within the brain must solve similar design problems to BMI readouts from the brain. This fits into a mindset of ‘neural sympathy’: interface design stands to benefit from a more fundamental understanding of how and why representations change during BMI usage.

Name dc:type.qualificationname

Doctor of Philosophy (PhD)

Level dc:type.qualificationlevel

Doctoral

Grantor dc:publisher.institution

University of Cambridge

Year dc:date.issued

2025

Author dc:creator

• Micou, Charles

Advisor dc:contributor.advisor

• O'Leary, Timothy

• Neuroscience
• Engineering
• Representational Drift
• Spatial Navigation
• Hippocampus
• Brain-Machine Interfaces
• Brain-Computer Interfaces

Licence

• https://www.repository.cam.ac.uk/bitstreams/52109b81-3419-46b4-8eaa-ee35020c9c93/download
• http://purl.org/NET/rdflicense/allrightsreserved

Language dc:language

eng

DOI dc:identifier.doi

https://doi.org/10.17863/CAM.122063

OAI identifier oai:identifier

oai:www.repository.cam.ac.uk:1810/390510