Matthew G. Perich

This paper introduces CURBD, a method using data-constrained recurrent neural networks to infer brain-wide interactions from neural recordings across species.

The paper shows that neural dynamics during movement are conserved across mice, monkeys, and humans, despite phylogenetic differences.

This paper introduces the BAND model to extract neural dynamics for feedback-driven motor corrections during reaching, finding that corrections are encoded in small neural variability in M1 cortex and are transient and sensory-driven.

This study shows that feedback-based motor control and motor adaptation share a common implementation in M1 circuits, as demonstrated through a recurrent neural network trained with error feedback.

This study shows that reinforcement learning produces neural dynamics in artificial networks that match biological motor cortex activity, suggesting RL underlies behavioral control development.

The paper shows that motor cortex activity encodes spatial and temporal arm movement parameters independently through low-dimensional neural trajectories.

This paper introduces GLUE, a framework linking neural activity geometry to computational efficiency, showing task-relevant representations untangle across various domains.

The paper shows that neural population dynamics are preserved across individuals of the same species performing similar behaviours, even when the brain structures differ.

Across monkey, mouse and human motor cortex and mouse striatum, population activity is reported to occupy curved rather than flat low-dimensional spaces, most clearly in complex tasks.

The paper shows that neural activity structure from prior learning influences adaptation, with more structured networks adapting better under specific conditions.

The paper uses a network model to show that small, correlated changes in synaptic connectivity can explain motor learning without disrupting neural covariance, aligning with experimental observations.

The paper examines the relationship between local field potentials (LFPs) and latent dynamics in primate sensorimotor cortex, finding frequency-dependent and region-specific correlations.

The paper shows that oscillatory trajectories in dimensionality reduction of neural activity data are not evidence of latent dynamics but result from continuous temporal variation in the data.

Recordings from four monkeys handling objects place input-driven activity in one putative somatosensory-feedback subspace of motor cortex, prominent around error correction and probed by stimulating ascending tracts.

The paper introduces a robotic platform for monitoring sensorimotor signals during natural reaching and grasping tasks in monkeys, demonstrating its use in kinematic studies, neural recordings, and brain-computer interface development.

The paper shows that latent neural dynamics in the cortex remain stable over two years, enabling consistent behavior decoding despite neuronal turnover.

The paper demonstrates a neural population mechanism in motor cortices that enables rapid behavioral adaptation without altering functional connectivity.

The paper investigates how motor cortical populations generate varied behaviors by analyzing neural modes across different motor tasks.

The paper introduces Linear-Nonlinear-Time-Warp-Poisson models to account for shared temporal variability in neural activity, showing improved predictions in macaque premotor cortex recordings.

The paper shows that dorsal premotor cortex encodes probability distributions of upcoming movements based on body state, while primary motor cortex does not.

This paper introduces a new decoding method called distribution alignment decoding (DAD) that uses movement statistics to map neural activity to motor variables, achieving comparable or better performance than traditional supervised decoders in some cases.