Machine Learning for Computer Systems

Course Overview

This course covers applications of machine learning to computer systems, with a particular focus on security, performance analysis, and prediction problems in networking. Topics include data preparation, feature selection, model evaluation, fairness, interpretability, and ML testing/debugging.

Primary Focus: Networking applications, though principles apply broadly across systems.

Course Resources

Course Website: noise-lab.net/ml-systems
GitHub Repository: github.com/noise-lab/ml-systems
Board Notes (PDF): Dropbox
ML Readings: noise-lab.net/ml-systems/ml.html

Lecture Topics (22 Lectures)

Use Cases (Lectures 1-4)

Introduction with Packet Capture basics
Security applications (scanning)
Performance analysis (QoE inference)
Resource optimization

Data & Pipeline (Lectures 5-8)

Data acquisition and feature extraction
Data preparation and representation
Model training and evaluation

Supervised Learning (Lectures 9-14)

Naive Bayes
Linear and polynomial regression
Logistic regression and SVMs
Trees and ensembles
Deep learning
Traffic representation learning (nPrint)

Unsupervised Learning (Lectures 15-17)

Dimensionality reduction
Clustering
Autoencoders

Generative Models (Lectures 18-19)

Transformers and large language models
Diffusion and state-space models

Bonus Topics (Lectures 20-22)

Timeseries analysis
Reinforcement learning
Model performance and maintenance

Course Materials

Jupyter Notebooks

Interactive lectures available on GitHub covering each topic with hands-on examples.

Reading Materials

Textbook chapters (Ch. 1-8)
Research papers
Additional resources in ML readings list

Assignments

Programming assignments and projects focused on applying ML to networking problems.

Format

The course uses a combination of:

Lecture presentations
Interactive Jupyter notebooks
Hands-on programming assignments
Reading materials from textbooks and research papers