visual-cortex

NeurIPS | 1st place solution for Sensorium 2023 Competition

Simulation of cat V1 simple cell and receptive field.

[fMRI] Multisensory integration of metaphorically related audiovisual inputs in visual cortex

A lightweight, ready-to-run demo environment for GEM-pRF, featuring automatic GPU setup, path configuration, and multiple example pipelines.

Interactive Streamlit app mapping PyTorch CNN activation maps (VGG16) directly to biological visual cortex regions (V1 to IT). Features live feature map extraction via forward hooks and neural heatmaps.

Analysis code for dimensionality and connectivity across the macaque ventral visual stream (THINGS Ventral Stream Spiking Dataset).

Code for my BSc thesis: decoding orientation preference maps from spontaneous activity in human V1. Compares sleep against wakefulness and finds sleep-derived maps markedly more stable — a stimulus-free route to functional cortical mapping for visual prostheses.

This project was developed as part of the Biologically Inspired Artificial Intelligence course at the Silesian University of Technology. The goal is to automatically classify eye diseases (diabetic retinopathy, cataract, glaucoma, normal) from retinal images using a convolutional neural network (CNN) inspired by the human visual cortex.

Linear decoding of visual stimulus category across mouse V1/LM/AL/RL (MICrONS). LM leads on fine natural categories.

Layer-wise CNN encoding of mouse V1 responses (Allen Institute Neuropixels), with an adversarial audit that downgraded the headline claim.