My work sits between systems neuroscience and machine learning. I am interested identifying represented structure among neural populations. These include the geometry of population activity, what that geometry preserves across sensory and motor contexts, and what that implies for decoding. In the near term this means speech and movement decoding from intracortical recordings; in the longer term, methods robust enough to decode declarative memory. I draw from both the theoretical side of neuroscience (Cog-Neuro) and the computational-application side (Brain-Computer Interface) in order to achieve this goal.
Testing whether the premotor and primary motor cortices encode bimanual cooperative context independently of kinematics — that is, whether the cortex represents the two hands are working together on one task as something separable from what each hand is doing.
Analysed intracortical population activity from a participant with C4 spinal cord injury across ten multisensory conditions, combining motor control, proprioceptive feedback and vision. Compared spike-band power against threshold crossings and found spike-band power significantly more reliable; several sensory pairings proved indistinguishable at the population level.
Built a Unity environment allowing participants with tetraplegia to control a soft robotic arm from neural signals in real time. Refactored a 1,600-line codebase into two modular files in three weeks and expanded the degrees of freedom available for assistive movement.
Validating Shepard’s Universal Law using human fMRI responses to the THINGS image set, and comparing the representational geometry of that cortical data against deep neural network embeddings. Manuscript in preparation.