Spotlight 2024 – Day3 – #1 | – Bryan DANIELS – Neural tuning for perceptual decisions

By Bryan Daniels (Arizona State University, Tempe, AZ). November 8th, 2024.

While drift-diffusion models are known to be good at explaining decision dynamics at a coarse scale, there is still no consensus about the detailed neural mechanisms that produce these dynamics. Data from macaque cortical neurons shows that a perceptual decision appears to occur in two phases: slow accumulation followed by fast consensus. We find that accumulation dynamics can be produced by critical slowing near a continuous transition (pitchfork bifurcation), with consensus arising naturally via an increase in effective interactions that pushes the system past the transition. We then examine the tuning necessary to utilize such a mechanism, showing how binary decision dynamics can be produced in random dynamical networks by adjusting two global parameters. Dynamical properties relevant to speed and accuracy of the decision are given by the network’s dominant eigenmode. These results suggest that dynamic tuning within a low-dimensional subspace that spans a critical transition could be an easily learned mechanism for implementing collective decisions.

Contact: Bryan Daniels, bryan.daniels.1@asu.edu