Sebastian Goldt

Noisy models of the ventral stream reveal the impact of recurrence and learned representations on information processing timescales

The paper investigates how network architecture influences representation timescales in a noisy, recurrent model of the ventral visual stream, showing that timescales are set by architecture rather than training.

Diverse perceptual biases emerge from Hebbian plasticity in a recurrent neural network model

This study shows that contractive and repulsive perceptual biases emerge from Hebbian plasticity in a recurrent neural network model across multiple tasks and datasets.