Syllabus
Unit 1
Single Layer Perceptron, Multi-Layer Perceptron (MLP), Activation functions – Sigmoid, Tanh, ReLU, Leaky ReLU, GELU, Softmax. Loss functions – Mean Squared Error, Mean Absolute Error, Cross-Entropy. Backpropagation – chain rule, computation graphs, gradient flow. Practical issues in neural network training – Vanishing and exploding gradient problems, Difficulties in convergence, The problem of overfitting – Bias-variance tradeoff, Regularization techniques: L1, L2, Dropout, Early stopping. Parameter initialization strategies Zero/Constant, Random, Xavier, He initialization
Unit 2
Introduction to deep learning, Deep feed-forward networks, Training deep models. Optimization techniques – Gradient Descent (GD), Stochastic GD (SGD), Mini-batch GD. Adaptive optimizers – RMSProp, Adam, AdamW. Learning rate schedules – step decay, cosine annealing, warm restarts. Batch Normalization, Layer Normalization. Dataset augmentation, Hyperparameter tuning strategies – grid search, random search, Bayesian optimization. Practical training pipeline – data preprocessing, train/validation/test splits, model checkpointing
Unit 3
Convolutional Neural Networks (CNNs) – convolution operation, stride, padding, pooling (max, average, global), variants of convolution functions (1D/2D/3D/Stride/ Dilated/Transposed/ Depthwise Separable). Transfer learning and fine-tuning. Recurrent Neural Networks (RNN), Attention mechanisms – self-attention, multi-head attention, scaled dot-product attention. Transformer architecture – positional encoding, encoder-decoder stack. Pre-trained language models – BERT, GPT family, Phishing and spam detection – NLP-based models, BERT fine-tuning for URL and email classification. Vision Transformers (ViT). Application of CNNs and ViT to malware image classification. Applications of transformers to time-series, text, and log data.
Unit 4
Graph Neural Networks (GNNs). GNN-based financial fraud detection. Autoencoders – vanilla, sparse, denoising, variational autoencoders (VAEs), latent space representation. Generative Adversarial Networks (GANs) – architecture, training dynamics, mode collapse. Applications of generative models in synthetic attack traffic generation, data augmentation for rare attack classes. Diffusion models – Latent Diffusion Models (LDMs), Denoising Diffusion Probabilistic Models (DDPM) basics. Large Language Models (LLMs) for cybersecurity – vulnerability detection, automated CVE triage, threat report generation, sentiment analysis, MITRE ATT&CK mapping. Retrieval Augmented Generation (RAG) for cybersecurity knowledge bases. Autonomous security agents – LLM-driven penetration testing assistants
Text Books / References
- Ian Goodfellow, Yoshua Bengio, Aaron Courville. Deep Learning. MIT Press, 2016.
- Aggarwal, C. C. Neural Networks and Deep Learning: A Textbook (2nd ed.). Springer Nature Switzerland AG, 2023.
- Christopher M. Bishop. Pattern Recognition and Machine Learning. Springer, 2006.
- Aurélien Géron. Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow. 3rd Edition. O’Reilly Media, 2022.
Introduction
This course introduces the fundamentals and advanced concepts of deep learning with a focus on modern AI-driven cybersecurity applications. It covers neural networks, optimization techniques, CNNs, RNNs, Transformers, GNNs, generative models, and Large Language Models (LLMs), along with practical aspects of model training and evaluation. The course emphasizes the application of deep learning techniques to real-world problems such as phishing detection, malware classification, fraud detection, threat intelligence, and autonomous security systems.
Objectives and Outcomes
Course Objectives:
- To build a solid foundational understanding of artificial neural networks, deep learning architectures, and training methodologies from first principles.
- To enable students to design, train, and evaluate deep learning models including CNNs, RNNs, Transformers, and Generative Models for real-world tasks.
- To apply advanced deep learning techniques to cybersecurity problems
Course Outcomes:
| COs |
Description |
| CO1 |
Learn and implement foundational deep learning concepts including perceptron, MLPs, activation functions, loss functions, backpropagation, and optimization algorithms. |
| CO2 |
Design and train advanced deep learning architectures such as CNNs, RNNs, Autoencoders, and Generative Adversarial Networks. |
| CO3 |
Analyze and apply Transformer-based models, and Graph Neural Networks |
| CO4 |
Apply deep learning and Large Language Models to cybersecurity tasks including fraud detection, vulnerability analysis, and threat intelligence. |
Prerequisites:
Generative AI with Python, Applied Machine Learning for Cybersecurity
Evaluation Pattern
CO-PO Mapping
| COs |
Description |
PO1 |
PO2 |
PO3 |
PSO1 |
PSO2 |
PSO3 |
PSO4 |
| CO1 |
Learn and implement foundational deep learning concepts including perceptron, MLPs, activation functions, loss functions, backpropagation, and optimization algorithms. |
2 |
– |
3 |
– |
– |
– |
2 |
| CO2 |
Design and train advanced deep learning architectures such as CNNs, RNNs, Autoencoders, and Generative Adversarial Networks. |
3 |
2 |
3 |
– |
2 |
– |
3 |
| CO3 |
Analyze and apply Transformer-based models, and Graph Neural Networks |
3 |
2 |
3 |
2 |
2 |
– |
3 |
| CO4 |
Apply deep learning and Large Language Models to cybersecurity tasks including fraud detection, vulnerability analysis, and threat intelligence. |
3 |
3 |
3 |
3 |
3 |
2 |
3 |
Evaluation Pattern – 60:40
- · Midterm Exam – 20%
- Quizzes – 20%
- Lab Assignments – 20%
- End Semester Exam – 40% (Theory and Practical)
