Tutorial 1: Deep Learning for Cybersecurity

Prerequisites

Mathematical maturity; Basics of Security including security goals, mechanisms, threat analysis and attacks such as malware, intrusion and phishing; Basics of Machine Learning including linear regression, supervised learning including basic loss functions especially categorical cross entropy, classification methods especially multi-layer perceptron model, and unsupervised learning especially clustering; Familiarity with basic text processing, natural language processing and understanding concepts; python programming.

Outline

Module 1: Introduction to Deep Leaning (2 hours 30 minutes lecture, 45 minutes of exercises)
– Feedforward Networks (FFN)
– Convolutional Neural Networks (CNN)
– Long Short-term Memory (LSTM) model
– Attention, Transformers, LLMs
– Autoencoders
– Generative Models
– Parameter-efficient machine learning
– Adversarial Machine Learning including attacks and defenses
– Examples and exercises will include: Python notebooks for FFN, CNN, Transformers, BERT, and an open source LLM

Module 2: Applications to cybersecurity challenges (2 hours lecture, 40 minutes of exercises)
– Deceptive attacks including social engineering attacks, business email compromise, fake news, and romance/job scams
– Intrusion detection
– Malware detection
– Adversarial robustness of deep learning models for cybersecurity
– Techniques for explainable ML
– Key takeaways and directions for future research
– Examples and exercise will include deep learning models for deceptive attacks and adversarial robustness, explainability techniques

References

1. Deep Learning book by Ian Goodfellow et al. Available online
2. Cybersecurity Analytics by Rakesh Verma and David Marchette, CRC Press, 2019. (recommended for prerequisite knowledge)
3. AI for Cyber security by Alessandro Parisi, Packt Press, 2019.
4. AI/ML in Cybersecurity, Malini Rao, 2023.