Teachlink
APASS-Learning platform
Dr. David Khachaturov
- Supervisor and Lecturer at the Department of Computer Science
- St Catharine’s College, Cambridge
Dr. David Khachaturov
- Supervisor and Lecturer at the Department of Computer Science
- St Catharine’s College, Cambridge
Dr. David Khachaturov is a researcher in the Department of Computer Science and Technology at the University of Cambridge and a by-fellow and supervisor at St Catharine’s College, Cambridge. His research primarily focuses on Machine Learning Security, Adversarial Machine Learning, AI Safety, as well as the application and governance of AI technologies within policy and legal frameworks. He has supervised over 300 students, covering approximately one-third of the undergraduate computer science cohort at Cambridge.
01. Introduction to Algorithmic Thinking and Computational Complexity
This module builds students’ core algorithmic foundations in artificial intelligence, focusing on fundamental algorithms—including sorting (Merge Sort, Quick Sort), searching (Binary Search), and graph traversal (BFS/DFS)—alongside computational complexity analysis (Big-O notation). Students develop rigorous algorithmic thinking and the ability to evaluate efficiency and feasibility in intelligent systems, providing a theoretical foundation for later study in machine learning and system security.
02. Principles of the Architecture of Neural Network
This module introduces the core principles and architecture of neural networks, covering key concepts such as neurons, multilayer perceptrons, activation functions, backpropagation, and gradient descent. Students will develop a foundational understanding of how neural networks learn from data and how model performance is optimized.
From a cybersecurity perspective, the module also explores neural network vulnerabilities, including adversarial attacks and data poisoning, reflecting the theme of AI Construction and Breaching. Students will further examine the “black box” nature of deep learning and the associated challenges of transparency and accountability
03. Adversarial Machine Learning and System Vulnerability Research
This module examines the security boundaries of artificial intelligence, addressing a key question: why powerful AI models can also be fragile. Students will explore fundamental software security concepts and failure modes, followed by the principles of adversarial attacks—how small, imperceptible perturbations can cause significant model errors, revealing the mathematical vulnerabilities of deep learning.
The module also introduces basic adversarial defense strategies, helping students understand how to build more robust systems and develop a security mindset shaped by the ongoing interplay between attack and defense.
04. Algorithmic Transparency and Accountability - AI Risk and Legal Frontiers
This module examines a central question of the AI era: when intelligent systems make unexplainable decisions, who is accountable? Students will explore the black-box problem, algorithmic traceability, and liability challenges in high-risk applications such as autonomous driving.
The module also introduces the fundamentals of Explainable AI (XAI) and key global regulatory developments, including the EU AI Act risk classification framework and the GDPR’s right to explanation. Through this, students will develop a governance-aware perspective, learning to evaluate AI innovation alongside legal and ethical responsibilities.
05. The Era of Human-Machine Symbiosis - Critical Thinking and Cognitive Autonomy
This module explores how AI is reshaping human cognition as intelligent tools increasingly influence decision-making. Students will examine the risks of over-reliance on AI, including the outsourcing of critical thinking, skill degradation, and the impact of algorithmic bias on human judgment.
Through the research project The Erasure of Expertise, participants will analyze how AI affects learning, decision-making, and creativity, and reflect on how to balance efficiency with human autonomy in human-AI collaboration.
■ Topic 1: “The Unsafe Autopilot” - Technical Direction
This is a comprehensive project intersecting technology and law, highly challenging and practically significant. Students must fully experience the entire process of “Build → Break → Defend.” Below is the project breakdown:
Core Tasks of the Three Project Phases
Build
Task: Train a vision model.Technical / Legal Focus: Use a CNN to train a model to recognise traffic signs (e.g., distinguishing a stop sign from a speed limit sign).
Break
Task: Generate “noise” to attack the model.Technical / Legal Focus: Create adversarial samples using technical methods to trick the model into misidentifying a stop sign as a speed limit sign.
Defend
Task: Write a legal defence brief.Technical / Legal Focus: Simulate a legal defence following an autonomous driving accident—who should be held responsible? The manufacturer? The developer? The car owner?
■ Topic 2: “The Erasure of Expertise” - Discussion Direction
When large language models can instantly generate articles, code, and answers, will humans’ own writing ability, judgment, and creativity gradually deteriorate? This is an in-depth research project on the fate of human cognition in the AI era.
Sample Research Report Structure
Live in St Catharine's College, Cambridge
Cambridge Formal Dinner
Punting in River Cam
VI Oxford &LondonJourney
