Skip to content

exploring the spatiotemporal dynamics of face recognition by directly comparing meg recordings with activations from cnns trained on different tasks and using different face stimuli

Notifications You must be signed in to change notification settings

BabaSanfour/MFRS

Repository files navigation

GitHub languages GitHub stars License Last commit GitHub contributors GitHub commit activity

Project Overview

Welcome to the main project repository! This interdisciplinary venture started as my End of Study Project during my engineering diploma and continued as my master's Project at DIRO/Mila UdeM.

This project explores the intersection of artificial intelligence and neuroscience, focusing on face recognition dynamics. Inspired by artificial neural networks surpassing human performance, the study investigates whether Convolutional Neural Networks (CNNs) trained for face recognition mimic the neural dynamics of face recognition in the human brain. Utilizing Magnetoencephalography (MEG) and comparing activations across seven CNNs, the research leverages high temporal resolution and source reconstruction techniques to unveil spatio-temporal similarity patterns. The study contributes novel insights into the complex interplay between artificial and biological neural responses associated with face recognition.

Here's an overview of the key folders and their contents:

Images Data Processing

This folder showcases a collection of scripts meticulously crafted to prepare data for a variety of experiments within the project. The scripts are organized into distinct categories, including HDF5 and repartition scripts, aligning with the unique structures of our datasets. The goal is to ensure efficient and tailored data processing for the diverse experiments ahead.

MEG Data Processing

Containing scripts for processing Magnetoencephalography (MEG) data, this repository focuses on a multi-subject, multi-modal human neuroimaging dataset available on OpenNeuro. The dataset features simultaneous MEG/EEG recordings during a visual recognition task, providing valuable insights into neural responses. From data acquisition details to processing steps, this repository covers the intricate journey of handling complex neuroimaging data.

Models

Dive into the world of model architecture, training, and feature extraction in the Models subfolder. This collection includes scripts showcasing a variety of models, each selected for its unique approach and method. Emphasizing diverse building blocks, this subfolder offers a rich exploration of neural network models for your perusal.

Similarity Analysis

The Similarity Analysis folder is at the forefront of advanced neuroimaging analyses. Explore scripts dedicated to Representational Dissimilarity Matrices (RDMs), Representational Similarity Analyses (RSAs), and noise ceiling estimation. These analyses provide a nuanced understanding of neural representations, offering valuable insights into the brain's information organization.

Utils Folder

The Utils folder houses utility modules and resources indispensable for various aspects of the study. From handling image datasets to providing essential support functions, these utilities play a crucial role in the project's success.

Supervisors and Acknowledgments

I extend my gratitude to my esteemed supervisors for their guidance and support throughout this project:

  • Dr. Karim Jerbi
  • Dr. Shahab Bakhtiari

Project Evolution

What began as an End of Study Project during my engineering diploma has evolved into a comprehensive exploration at the intersection of engineering, neuroscience, and artificial intelligence. This repository serves as a journey of learning that I went through for almost three years, starting with random scripts of code to the project current state. In addition to the main goal of the project, we had an intern working on a side idea that involves comparing 1 model trained using different loss functions to the brain.

Papers and Presentations

Check out the list of papers and presentations associated with this project during its different stages:

Master thesis

The work done here was for my master thesis, which could be found here Link, a more cleaner version used for the thesis and paper publications could be found here

Funding

This project received funding from the UNIQUE (Unifying AI and Neuroscience) research center, Cerebrum, and CIRCA as well as generous contributions from my supervisor, Karim jerbi, research funds.

About

exploring the spatiotemporal dynamics of face recognition by directly comparing meg recordings with activations from cnns trained on different tasks and using different face stimuli

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •  

Languages