Week 7: Semi-Supervised and Reinforcement Learning
Dates: Feb 23-27 · Reading: Handout 6: Semi-Supervised and Reinforcement Learning
This week completes the four styles of machine learning. Monday: semi-supervised learning (a few labels + lots of unlabeled data) and reinforcement-learning intuition (learning from rewards), plus security as a defender-attacker game. Wednesday: applying both in security - malware/intrusion detection and RL-driven automated defense - with the risks of automation. Lab 6 released this week.
Learning Objectives
- Explain semi-supervised learning as learning from a few labels plus lots of unlabeled data
- Describe reinforcement learning as learning from rewards and penalties
- Identify security use cases for semi-supervised and reinforcement learning
- Relate reinforcement learning to defender-attacker game theory
Lecture Slides
⬇ Monday: Semi-Supervised & Reinforcement Learning (PPTX) ⬇ Wednesday: SSL & RL in Security (PPTX)
Monday Session
Semi-supervised learning: why labels are scarce, how a few labels plus lots of unlabeled data beat labels alone, and the -1-for-unlabeled workflow in scikit-learn. Reinforcement-learning intuition: agent, environment, reward, and policy; exploration vs. exploitation; everyday analogies. Closes with security framed as a defender-attacker game.
Wednesday Session
Applications: semi-supervised learning for malware classification and intrusion detection; reinforcement learning for defender-attacker games and automated response (SOAR), including honeypots and moving-target defense; and the real risks of automating security decisions. Ends with the full four-paradigm map and how real pipelines combine them.
Lab
Lab 6: Semi-Supervised Classification. Hide most labels, train a supervised baseline on the few that remain, then a semi-supervised model that also uses the unlabeled files - and measure how much the unlabeled data helps.
⬇ Lab 6 starter notebook (.ipynb)
Quiz / This Week
Quiz 6. Semi-supervised learning; reinforcement learning; the four learning paradigms; security as a defender-attacker game.
In-Class Activity
When to Request Manual Labels — CISA-style incident-response tabletop (small groups). Graded, 20 points.
⬇ Activity slides (PPTX) 📄 Student worksheet (PDF)