Junxing Shi

This paper uses a variational autoencoder to model and decode fMRI activity in the visual cortex, finding lower prediction accuracy compared to CNNs but successful decoding of latent variables.

This paper introduces a deep recurrent neural network to model process memory in visual processing and shows it better predicts cortical responses to natural movies than a standard CNN.

The paper uses a deep residual network to map cortical representations of 64,000 visual objects, revealing a nested hierarchy organized by category specificity and spatial scale.

This study develops methods to transfer and generalize deep-learning-based encoding models across subjects using Bayesian inference and incremental data training, applied to fMRI data from naturalistic video watching.