Linxing Preston Jiang

This study shows that data heterogeneity limits the scaling benefits of pretraining neural data transformers in neuroscience datasets, suggesting that careful session selection can improve performance.

This paper introduces a hierarchical model of sequence learning and prediction in the neocortex, showing how it develops temporal response hierarchies and supports memory recall.

This paper introduces a dynamic predictive coding model using hypernetworks to predict spatiotemporal patterns in natural image sequences, showing it learns visual cortex-like receptive fields.

This paper introduces a method using effective connectivity analysis of fMRI data to distinguish between competing computational models of the Stroop task, showing that functional connectivity patterns can differentiate models with identical behavioral predictions.