iiinova — where your digital self keeps evolving

Raeed Chowdhury @raeedcho.com · 15/04/2026

At the same time, we found that for trying to detect external perturbations, the orthogonal and opposite geometries could succeed, but the aligned geometry could not. 9/

Bump classification performance for orthogonal, aligned, and opposite geometry of CD and FB signals

110

Raeed Chowdhury @raeedcho.com · 15/04/2026

What benefit might there be to keeping these signals separate and orthogonal? We simulated various geometries to find out. We found that both orthogonal and aligned geometries could combine the two signals for better state estimation, but an opposed geometry could not. 8/

State estimation performance in models using orthogonal, aligned, and opposite geometry of CD and FB signals

120

Raeed Chowdhury @raeedcho.com · 15/04/2026

CD and FB signals in area 2 also seem to be orthogonal to each other in the full neural state space. This echos similar findings in other cortical areas, showing that, for example, motor preparation and execution lie in orthogonal neural subspaces. 7/

Histogram of Monte Carlo analysis on principal angles between corollary discharge and feedback signal subspaces

133

Raeed Chowdhury @raeedcho.com · 15/04/2026

However, for passive trials (unseen by the decoding procedure), this relationship is reversed--FB dimensions come on first, and CD signals come on later, likely to coordinate the active return reach. 6/

Example corollary discharge and feedback signal dimensions during passive perturbation trials

110

Raeed Chowdhury @raeedcho.com · 15/04/2026

Using an iterative decoding procedure in each of these active trial epochs, we isolated CD and FB signals. For active trials, as we would expect, we found that CD signals start before movement, and then reverse direction for the return reach. FB signals look similar, but delayed. 5/

Example corollary discharge and feedback signal dimensions during active reaching trials

101

Raeed Chowdhury @raeedcho.com · 15/04/2026

In active reaching trials, we would expect corollary discharge (CD) signals, but not feedback (FB) signals, to be present in neural activity just before the reach. Likewise, just before the end of the reach, we would expect primarily feedback signals. 4/

Diagram showing the epochs where corollary discharge and feedback signals can be isolated during active reaching

111

Raeed Chowdhury @raeedcho.com · 15/04/2026

To look for neural evidence of this integration, we recorded from area 2 of somatosensory cortex as monkeys actively reached to targets and were passively perturbed. 3/

Task diagram showing time course and hand trajectories of an active and passive task

121

Raeed Chowdhury @raeedcho.com · 15/04/2026

Most of us interact with the world without thinking about how. A big part of this ability is sensory feedback about body movement, but because of pathway delays, we can't rely only on feedback. Instead, we integrate feedback with motor signals ("corollary discharge") into an estimated body state. 2/

Block diagram showing integration of sensory feedback and corollary discharge for motor control

112

Raeed Chowdhury @raeedcho.com · 15/04/2026

New preprint! 🧵🧪🧠 doi.org/10.64898/202... Upshot: somatosensory cortex contains both feedback *and* intended movement signals (a.k.a. corollary discharge). These two signals exist orthogonally in neural space, flexibly allowing both fast state estimation and external perturbation detection. 1/

Neural trajectory plot showing neural activity moving through orthogonal corollary discharge and feedback signal subspaces

34619

Raeed Chowdhury @raeedcho.com · 15/04/2026

At the same time, we found that for trying to detect external perturbations, the orthogonal and opposite geometries could succeed, but the aligned geometry could not. 9/

000

Raeed Chowdhury @raeedcho.com · 15/04/2026

100

Raeed Chowdhury @raeedcho.com · 15/04/2026

100

Raeed Chowdhury @raeedcho.com · 15/04/2026

100

Raeed Chowdhury @raeedcho.com · 15/04/2026

100

Raeed Chowdhury @raeedcho.com · 15/04/2026

100

Raeed Chowdhury @raeedcho.com · 15/04/2026

To look for neural evidence of this integration, we recorded from area 2 of somatosensory cortex as monkeys actively reached to targets and were passively perturbed. 3/

100

Raeed Chowdhury @raeedcho.com · 15/04/2026

100