Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster E in Poster Session E: Thursday, August 6, 10:30 am – 12:15 pm, Kimmel Center, Shorin & Rosenthal Rooms

Low-Rank Temporal Factorization of Drift-Diffusion Dynamics Reveals Staged Evidence Integration

JINGMING XUE¹; ¹Georgia Institute of Technology

Presenter: JINGMING XUE

Perceptual decision-making depends on integrating sensory evidence over time, but the resulting integration kernels vary substantially across individuals. Low-rank neural networks capture the principal component of this variability but lack a clear link to cognitive mechanism, whereas the Drift-Diffusion Model (DDM) provides an interpretable account of decision formation. To connect these frameworks, we parameterize the DDM’s starting point (y₀) and drift rate (a) with low-rank temporal kernels driven by the click sequence 𝐜_t in an auditory clicks task. Model fitting identifies a one-dimensional latent representation (d=1) for each parameter: drift rate weighs heavier on the middle and late clicks than earlier clicks, whereas starting point was set early and middle clicks and is not influenced by late clicks. Together, these results support a staged computation in which evidence accumulation may not be strictly synchonized with the stimulus in the auditory click task.

Topic Area: Decision-Making, Cognitive Control & Event Cognition