CS 404/504 Special Topics: Adversarial Machine Learning
Link to the course materials from previous years: Fall 2020, Fall 2021
Course Syllabus
Course Description
The course introduces students to adversarial attacks on machine learning models and defenses against the attacks. The particular focus is on adversarial examples in deep learning models, due to their prevalence in modern machine learning applications. Covered topics include evasion attacks against white-box and black-box machine learning models, data poisoning attacks, privacy attacks, defense strategies against common adversarial attacks, and robust machine learning models. The course also provides an overview of adversarial attacks against machine learning models used in cybersecurity applications, including malware detection and classification, network intrusion detection, spam filtering, URL detection, cyber-physical systems, and biometric systems.
Course Objectives
The objective is that upon the completion of the course the students should demonstrate the ability to:
1. Outline the different categories of adversarial attacks against machine learning models.
2. Describe common defense against adversarial attacks approaches for improved robustness of machine learning models.
3. Understand the basics of adversarial privacy attacks and privacy-preserving defense methods.
4. Identify the unique characteristics of adversarial machine learning attacks in the cybersecurity domain.
5. Implement adversarial attacks and defenses against conventional machine learning models and deep learning models.
6. Implement adversarial attacks against anomaly detection systems for network intrusion detection, malware classifiers, and anti-spam filtering methods.
Course Materials
Textbook:
- There is no required textbook. The required readings for each week are listed in the Course Outline section of the Syllabus.
Topics
- Introduction to Adversarial Machine Learning (ppt, pdf)
- Deep Learning Overview (ppt, pdf)
- Mathematics for Machine Learning (ppt, pdf)
- Evasion Attacks against White-box Machine Learning Models (ppt, pdf)
- Evasion Attacks against Black-box Machine Learning Models (ppt, pdf)
- Generative Adversarial Networks for Adversarial ML (ppt, pdf)
- Physical-world Adversarial Examples (ppt, pdf)
- Defenses Against Evasion Attacks (ppt, pdf)
- Poisoning Attacks against Machine Learning Models (ppt, pdf)
- Defenses Against Poisoning Attacks (ppt, pdf)
- Adversarial ML in Cybersecurity: Network Intrusion Detection (ppt, pdf)
- Adversarial ML in Cybersecurity: Malware Detection and Classification (ppt, pdf)
- Adversarial ML in Cybersecurity: Spam Filtering, URL Detection, and Cyber-physical Systems (ppt, pdf)
- Privacy Attacks against Machine Learning Models (ppt, pdf)
- Defenses against Privacy Attacks (ppt, pdf)
- Adversarial Examples in Text and Audio Data (ppt, pdf)
Prerequisites
Instructor Permission Required
Students are expected to have basic knowledge of linear algebra, probability and statistics, and machine learning concepts. Knowledge of neural networks and deep learning is recommended, but not required. Programming in Python is required for completing the course assignments and the project. Additionally, it is preferred that the students are familiar with at least one of the following machine learning libraries: TensorFlow, Keras, or PyTorch.
Assignments
- Assignment 1 - white-box evasions attacks against deep learning-based classification models. Student solution.
- Assignment 2 - black-box evasions attacks against deep learning-based classification models. Student solution.
- Assignment 3 - adversarial defenses for white-box evasions attacks, and poisoning attacks against deep learning-based classification models. Student solution.
- Assignment 4 - adversarial attacks against machine learning models in cybersecurity applications.
- The datasets used in the assignments and data loading notebooks can be downloaded from here.
Evaluation Procedure
Homework Assignments (4) |
50 % |
Quizzes (3) |
30 % |
Lecture Presentation |
10 % |
Course Project |
10 % |