The Sachs Lab has prepared a 2-day workshop/tutorial on practical deep learning (DL) applied to intracranial neurophysiology. This repository contains documentation, code, and tutorial materials for the tutorial.
Attendees will gain familiarity with technologies commonly used in DL (e.g., keras/tensorflow on GPU, jupyter notebooks on the cloud), to understand DL programming paradigms, and to become proficient in building, training, and evaluating deep neural networks as applied to extracellular neurophysiology data.
Attendees will learn how to load and process electrophysiology datasets (ECoG, single-channel deep brain microelectrode recordings, and multichannel (~192) intracortical microelectrode array datasets). After an introduction to DL, attendees will learn how to apply several DL algorithms and architectures to these types of data, and finally they will explore different ways of using deep learning to advance neuroscientific endeavours. Some of the algorithms we cover include convolutional neural nets (CNN) and several flavours of recurrent neural nets (RNN).
Attendees are expected to have a basic understanding of machine learning concepts, have basic familiarity with Python syntax, and have an interest in applying deep learning to extracellular neurophysiology data. Most of the contents will apply just as well to other multi-channel timeseries data.
For each topic, attendees will work through prepared examples using real data and thus are expected to bring their own laptop. Their laptop should be configured prior to the tutorial. Instructions are below in the Getting started section.
The tutorial is opinionated in its selection of development environment, deep learning framework, and example datasets. There are many tools to choose from, but we choose to use the TensorFlow framework and Keras API, and we require a GPU (tensorflow-gpu). If your laptop does not have a GPU then you can do most of the deep learning on the cloud in Google Colab.
Follow the instructions in the BeforeTheWorkshop document.
When you have finished configuring your environment, you should be able to interact with the tutorial notebooks, either locally or remotely.
The lessons comprise a series of Jupyter Notebooks, a.k.a. interactive python notebooks with file extension .ipynb.
From the Jupyter website:
The Jupyter Notebook is an open-source web application that allows you to create and share documents that contain live code, equations, visualizations and narrative text. Uses include: data cleaning and transformation, numerical simulation, statistical modeling, data visualization, machine learning, and much more.
Chad's Opinion:
While Jupyter notebooks are well-suited for didactic purposes like this course, they are the wrong choice for real-time analysis (e.g., brain-computer interface) due to limited I/O and they are the wrong choice for the development of python packages due to the poor support for debugging imported packages. Even though we use Jupyter notebooks here, you should strongly consider whether or not they are the right choice for your intended application. Note: PyCharm professional >= 2019.2 (free for academics) has support for debugging jupyter notebooks so this may be a viable workflow for some.
We have curated two datasets specifically for this tutorial. Please see the README in a data subfolder for their descriptions.
Subsets of these datasets will be downloaded on demand as required by each lesson.
Larger datasets are not required but you may download them following the instructions in the data directory for your own exploration outside the tutorial. The data preprocessing and conversion scripts use NeuroPype. NeuroPype is free for academics but the publicly available academic version is currently too old (as of June 2019).
The tutorial comprises 8 Parts. Each Part presented by the Sachs Lab has slides and one or more Jupyter notebooks. The slides can be found in the repository slides folder (TODO: Check back here on the day of the tutorial). The notebooks can be found in the notebooks folder.
Start with the slides for Part 1. The slides will then instruct the student to open one of the notebooks. If using Google Colab, you can follow the link in the table found in the notebooks README.