Zhuokun Ding

This paper incorporates inhibition into deep neural networks for neural prediction, showing it improves surround suppression learning but not cross-orientation inhibition.

NEURD automates proofreading and feature extraction from connectomics data to improve accuracy and accessibility of neural morphology analyses.

Functionally similar neurons in mouse visual cortex connect preferentially across layers and areas, including feedback, and feature tuning, not receptive-field position, predicts synapses beyond axon-dendrite proximity.

This paper trains a foundation model on neural activity data to predict responses to new stimuli and anatomical features, showing strong generalization across mice and tasks.

The paper identifies a novel bipartite invariance in mouse V1 neurons, where receptive fields encode shift-invariant high-frequency textures and fixed low-frequency patterns, contributing to segmentation.

The paper uses CNNs to analyze contextual modulation in visual cortex by synthesizing surround patterns, finding that pattern completion enhances responses while disruption suppresses them, consistent across mice and macaques.

The paper synthesizes neural metamers for the mouse visual system using deep encoding models and shows they reproduce neural activity across different animals.

This paper introduces a flow-based generative model to capture neural population responses to natural images, outperforming existing models in predicting variability and learning latent factors related to behavior and brain area.