Machine Learning in Python - Neural networks, data leakage, the train/test split

Overview: One of the most discussed and perhaps mysterious machine learning models is the neural network. Neural networks are a kind of machine learning model inspired by biological processes taking place in the brain. This lesson will demystify neural networks and provide you with a plain-English explanation of how they work. We will train a neural network to recognize handwritten digits; this is a classification task. We will also discuss deep learning and further explore the training step in the machine learning pipeline.

Learning Goal(s): By the end of the workshop, participants will be able to:

• Given a scaffolded environment and curated data set, follow a tutorial that trains a neural network to perform classification.
• Describe in plain English the structure of neural networks in general.
• Appreciate the use of backpropagation for training neural networks.
• Articulate the common pitfalls in training and validating machine learning models.

Prereqs: Participants should already have some familiarity with Python programming fundamentals, e.g. loops, conditional execution, importing modules, and calling functions. Furthermore, participants should ideally have attended the first lesson in the “Fundamentals of Machine Learning in Python” series, or they should already have some background on the general machine learning pipeline.

Approach: Our approach is primarily student-centered. Students will work in pairs and small groups on worksheets and Jupyter notebooks, interspersed with brief lecture and instructor-led live-coding segments.

Supporting Resources: We will refer to many of the materials used previously in our series “Computing Workshop” https://computing-workshop.com/

Deliverables: Our resources will be made available via our web site.

Resources required: Participants should have access to a laptop computer. Python should be already installed with Anaconda.

Location: HYBRID. Online via Zoom, or in-person at Burnside Hall room 1104 (11th floor).
Instructor: Jacob Errington, Faculty Lecturer in Computer Science at McGill University. Eric Mayhew, Computer Science professor at Dawson College.