Instructional goals
The course offers a comprehensive introduction to neural networks and deep learning. The main architectures (e.g. feedforward, convolutional and recurrent networks) and fundamental training techniques will be covered. Practical applications and experiential labs will complement the theoretical concepts.
Prerequisites
Computer programming skills (Python). Foundations of machine learning.
Intended learning outcomes
Knowledge and understanding: Students will understand the fundamental principles, challenges, and design choices underlying neural networks and deep learning models, analyzing the strengths and limitations of different architectures and training approaches.
Applying knowledge and understanding: Students will be able to develop innovative, data-driven solutions using modern neural network techniques, covering the full lifecycle from model training and validation to evaluation and deployment.
Making judgements: Students will be able to select the most appropriate neural network architecture for a given problem and critically assess the strengths and limitations of different architectures, training strategies, and evaluation methods presented in class.
Communication skills: Students will be able to master, with adequate terminological precision, the technical vocabulary relevant to neural networks and deep learning. Particular emphasis will be placed on oral presentations and pitches in project group work, as well as on writing technical reports and documentation.
Learning skills: Students will master tools and methods for designing, training, and validating neural networks, applying them to real-world problems that are typical of today’s data-driven companies.
Course Contents
Neural network foundations
Backpropagation and optimization
Deep feedforward networks
Convolutional neural networks
Sequence models (recurrent neural networks)
Training best practices
Python toolchain for deep learning
Reference Books
François Chollet, Matthew Watson, Deep Learning with Python, Manning, 3rd Edition
Andrew W. Trask, Deep learning, Manning, 2019
Lecture slides, research papers, notebooks and course material provided on the Luiss learning platform
Teaching Methods
Traditional lectures
Practical lab sessions
Industrial testimonials
Assessment Method
Midterm (30%)
Software project (40%)
Final exam (30%)
Thesis assignment criteria
Thesis assigned, upon specific request to the professor, to students who demonstrate a strong interest in the course topics and a genuine motivation to pursue research in related areas.
Week 1
Deep learning foundations. Core ideas: neurons, inputs, weights, predictions, and errors. Lab session.
Week 2
Forward propagation. Weighted sums, single- and multi-input models, layers, and forward propagation. Lab session.
Week 3
Gradient descent. Loss functions, gradient descent, learning rate, iterative weight updates, and convergence intuition. Lab session.
Week 4
Hidden layers, deep feedforward networks, composition of transformations, intuition behind backpropagation. Lab session.
Week 5
Backpropagation through multiple layers, chain rule, parameter updates, and training a complete feedforward network. Lab session.
Week 6
Generalization and regularization. Training versus validation error, overfitting, mini-batches, regularization, dropout. Lab session.
Week 7
Nonlinear neural networks. Activation functions, nonlinear representations, classification outputs. Lab session.
Week 8
Convolutional neural networks. Images as structured data, convolution, filters, feature maps. Lab session.
Week 9
Convolutional neural networks. CNN architectures, hierarchical feature extraction, pooling, and image classification. Lab session.
Week 10
Recurrent neural networks. Sequential data, recurrent connections, temporal dependencies, sequence classification, limitations. Lab session.
Week 11
Training practice and end-to-end deep learning. Model selection, evaluation, debugging, reproducibility, practical workflow. Lab session.
Week 12
Course wrap-up and industry perspectives.