Course Content

Note

This is the inaugural run of this Machine Learning course. With zero historical data (i.e., no prior offerings), predicting the optimal level of detail for each topic is, ironically, an ill-posed ML task. There’s no training set, no validation split, and certainly no ground truth to benchmark against. So consider this syllabus a fluid model — one that will adapt, update, and re-train as we go.

💪 Despite all this, stick with me and in just 4 months you'll learn to navigate the depths of deep learning — and maybe even unbox that mysterious black box everyone keeps warning you about.

1. Introduction

• Overview of machine learning and its applications. Types of learning: supervised, unsupervised, reinforcement.

2. Preliminaries

• 2.0 GitHub & Conda Setup
• 2.1 Data Manipulation
• 2.2 Data Preprocessing
• 2.3 Quick Refreshers: Linear Algebra, Calculus and Automatic Differentiation, Probability and Statistics
• 2.4 PyTorch Overview

3. Linear Neural Networks for Regression

4. Linear Neural Networks for Classification

5. Multilayer Perceptrons (MLPs)

• Building and training feedforward networks.

6. Optimization Algorithms

• Gradient descent and its variants. Understanding how learning happens.

7. Building a Neural Network

• Modular design, managing parameters and hyperparameters, initialization strategies, and tuning methods.

8. Convolutional Neural Networks (CNNs)

• For image data and spatial hierarchies.

9. Recurrent Neural Networks (RNNs)

• For sequence modeling and temporal data.

10. Selected Topics in Unsupervised Learning

• Autoencoders and an introduction to Generative Adversarial Networks (GANs).

11. Interpretability and Explainability (XAI and IAI)

• Exploring methods like SHAP and other tools from scikit-learn to understand model decisions.

12. Writing About ML Techniques in Scientific Papers

• Best practices for describing machine learning models and results in academic writing, with examples from real research in the field.