Universal data load for neural networks

Project title: Universal data load for neural networks

Leader:

• Ning Mei (@nmningmei)

Collaborators:

Topic: Deep learning, Input/Output

Project description: There exist data loader pipeline for deep neural network project using Tensorflow or Pytorch, but they mainly focus on computer visions and natural language processing. When neuroscientists implement deep neural network models on neuroimaging or neurosignal data, a customized pipeline must be written from scratch every time. This project aims to provide a universal data loader, based on either Tensorflow or Pytorch data loader pipelines, so that it is easier for neuroscientists to implement neural network modeling in the next steps.

Data to use: Data is to be determined but one M/EEG and one fMRI data will be selected for testing.

Link to project repository:

Goals for Brainhack Donostia 2021:

1. A data loader for Numpy arrays

2. A data loader for converting EEG Epochs data to batches3. A data loader for converting fMRI data to batches

First tasks:

1. https://github.com/nmningmei/BOLD5000_autoencoder/blob/master/scripts/2.2.extract%20volumes.py

2. https://github.com/nmningmei/BOLD5000_autoencoder/blob/master/scripts/3.1.simple%20autoencoder%202D.py#L25:L39

Communication channels:

Video channel: Jitsi

Number of collaborators: 3

Credit to collaborators: N/A

Type of project: Data management, Documentation, Pipeline development

Development status: Basic structure

Programming languages: Python

Necessary Programming skills level for the project: Intermediate

Necessary git skills level for the project: Familiar

Modality: EEG, fMRI

Software suites: MNE, Nipype

Email: nmei@bcbl.eu

What will participants learn:

1. Object oriented programming in Python
2. Knowledge about arrays and tensors
3. Basic I/O knowledge of M/EEG (MNE-python) and fMRI (nibabel)