Machine Learning for Neuroscience

Deepen your understanding of aspects of machine learning as applied to neuroscience. This course will cover supervised and unsupervised methods as well as classic machine learning methods and deep learning.

What will you do in this course?

This course will explore machine learning methods with a discussion on their use in neuroscience. We will cover several aspects of machine learning, from classical methods to deep learning.

By the end, you will be able to:

• ILO 1: Critique and contrast machine learning techniques and analyse the problem settings within which they are particularly useful for.
• ILO 2: Develop machine learning models by applying common machine learning libraries and tools.
• ILO 3: Apply practical computational and machine learning techniques to analyse data from a variety of sources with a particular focus on published research, and interpret the results.
• ILO 4: Plan and implement neural networks to develop end-to-end solutions for data analysis problems.
• ILO 5: Justify the choices of machine learning models based on their appropriateness for the problems, generalisation and limitations.

Prerequisites

This course has no specific prerequisites, but knowledge of linear algebra, probability theory and calculus will give you the best chance of success.

Reading List

A good place to start is Kevin Murphy's book on Probabilistic Machine Learning. Before starting this course you might find it helpful to read:

• The Introduction
• Foundations: Probability: Univariate Models
• Foundations: Linear Algebra

If you get through these and want some more reading, please contact us and we can recommend some more!

What are you being assessed on?

We want to examine your ability to apply machine learning methods learnt during the lectures to novel situations. Questions will test theoretical machine learning concepts as well as your ability to apply these methods in Python.

How will this be assessed?

Through this course, you will complete three pieces of coursework during lab sessions, and a final project.

: Slides

: Notes

: Download lab

: Open lab in Colab

♦: Guest Lecture: Introduction to Pytorch, Alex Capstick

♣: Guest Lecture: Large Language Models for Electronic Healthcare Records (EHR) Data Analysis, Louise Rigny (Data Scientist, Great Ormond Street Hospital for Children)

March/April 2024

In March/April 2024 there will be an optional series on Generative AI and Large Language Models (LLMs) covering the topics:

• Variational Auto-encoders
• Transformers and Large Language Models
• Diffusion Models

Last year's slides and recordings are available here.

The slides are by Payam Barnaghi, whilst the notes are written by Nan Fletcher-Lloyd.

Content to get you started:

• Preface: Preface.pdf
• Notations: Mathematical Notations.pdf
• Basics of Matrix Algebra: Matrix Algebra.pdf

The lecture slides are available here:

• Lecture 1: Introduction to Machine Learning: Slides | Notes
• Lecture 2: Linear Models: Slides | Notes
• Lecture 3: Probability and Information Theory: Slides | Notes
• Lecture 4: Bayesian Models: Slides | Notes
• Lecture 5: Ensemble Models and Kernel Based Models: Slides | Notes
• Lecture 6: Neural Networks: Slides | Notes
• Lecture 7: Convolutional Neural Networks (CNNs): Slides | Notes
• Lecture 8: Applicaitions and Neuroscience Inspired Machine Learning: Slides | Notes
• Lecture 9: Ethical Considerations and Responsible Machine Learning: Slides | Notes
• Lecture 10: Review and Summary: Slides

Labs and assessments are produced by Payam Barnaghi, Alex Capstick, Nan Fletcher-Lloyd, Yu Chen, Tianyu Cui, Marirena Bafaloukou, Ruxandra Mihai, Francesca Palermo, and Anastasia Gailly de Taurines. The entire github repository can be downloaded as a zip file using this link. This includes the labs and a copy of this website, which can be opened locally.

Miscellaneous Content:

• Linear Algebra Tutorial (Adapted from Kevin Murphy's linalg.ipynb Tutorial): Linear Algebra Tutorial.ipynb
• Normalisation, Scaling, and Imputation: Normalisation Scaling Imputation.ipynb
• Pytorch Tutorial: Pytorch Tutorial.ipynb

The following links are to the individual labs, already run and stored in GitHub. Clicking any of the "Open in Colab" buttons will open that lab in Google Colab, which allows you to run code from your browser with a GPU if selected.

Lab Notebooks:

• Lab 1: Python Tutorial Preliminary Skills and Concepts:
  • Description of Tutorial Notebooks: Notes
  • Python for Beginners: Lab Notebook
  • Machine Learning for Beginners: Lab Notebook
  • Machine Learning for Beginners Assessment (Not Marked): Lab Notebook
• Lab 2: Linear Models - Regression and Classification Models: Lab Notebook
• Lab 3: Probability and Information Theory: Lab Notebook
• Lab 4: Bayesian Models: Lab Notebook
• Lab 5: Ensemble Models and Kernel Based Models: Lab Notebook
• Lab 6: Neural Networks: Lab Notebook
• Lab 7: Convolutional Neural Networks (CNNs):
  • Convolutional Neural Networks (CNNs): Lab Notebook
  • CNN Edge Detection Sample: CNN Edge Detection Sample.ipynb
  • CNN LeNet-5 Sample: CNN LeNet-5 Sample.ipynb

The following are the steps to set up your environment if you will be running the labs locally:

1. Install Miniconda here.
2. On Windows open Anaconda Prompt and on Mac/Linux open Terminal.
3. To create an environment for the course, run:

   conda create -n ml4ns python=3.11.5
           

4. Download the the requirements.txt file (from here) and place it in your root directory.
5. Activate your environment:

   conda activate ml4ns
           

6. Install the requirements:

   pip install -r requirements.txt
           

7. If you're on Mac or Windows (without a GPU), you can install Pytorch using the below. If you have a GPU, you can install Pytorch using the command on the Pytorch website.

   pip3 install torch torchvision torchaudio
           

8. Install Transformers using:

   pip install transformers
           

9. Recommended: Install VScode here.
10. Open VSCode and open the Jupyter Notebook files, or open Jupyter Lab using the below and then open the Jupyter Notebook files:

    jupyter lab
            

11. If you are using VSCode, you need to install the Jupyter Notebook extension.
12. In VScode, select the Python interpreter for your environment (top right, usually!).
13. If everything has worked then in the Python environment you should be able to run:

    
                  import torch
                

    Then:

    
                  torch.cuda.is_available()
                

    And if you have a GPU, you should see True, otherwise False.

Each session:

• If you're using Jupyter Lab, every time you go to work on this project you'll need to activate your environment by running conda activate ml4ns using either the Terminal (Mac) or Anaconda Prompt (Windows). Then you may enter the command jupyter lab to open Jupyter Lab.

• If you're using VSCode, you'll need to select the Python environment for each new file you open.