Which iterative models best explain recognition reaction times and neural dynamics

Determine whether iterative brain models that perform inference using generative models of visual inputs or iterative brain models based on discriminative computations better account for image-dependent recognition reaction times and associated neural dynamics in primate vision.

Background

Recognition of briefly presented images is rapid but varies across images and conditions, suggesting that timing depends on stimulus difficulty. Generative models typically involve iterative inference, potentially explaining image-dependent latencies, while purely feedforward discriminative models predict reaction times that are more stimulus-invariant.

The paper highlights that both iterative generative and iterative discriminative models could, in principle, account for the observed dynamics, leaving unresolved which class better explains recognition time variability and neural dynamics.