NETWORK INTELLIGENCE

Academic Year 2024/2025 - Teacher: Marco SIINO

Expected Learning Outcomes

Knowledge of basic and advanced theoretical aspects of Artificial Intelligence applied to the network domain. Complete knowledge of Machine Learning theory. Knowledge of the libraries and languages most used for Deep Learning.

Knowledge and understanding: Taking advantage of the knowledge acquired, the student will be able to apply the theoretical foundations inherent to the development of Deep Learning models for telecommunications and networks.

Applied knowledge and understanding: By taking advantage of the skills acquired in the course, students will be able to identify the main requirements for the specific application context and to identify and design the most appropriate solutions.

At the end of the course, students will be able to identify the best solutions for the design and development of Artificial Intelligence models to be applied to the networking domain.

Course Structure

Traditional/frontal teaching.

Required Prerequisites

Probability theory and random variables, theory of determinate and random signals, analog modulation schemes and the main aspects relating to digital modulations (ASK, FSK, PSK) and channel coding. Fundamentals of statistics and mathematical analysis.

Attendance of Lessons

Not mandatory.

Detailed Course Content

Topics: The competition is divided into three main areas of interest:

Fundamentals of Machine Learning:

Introduction to Machine Learning: definitions, techniques, and key algorithms.
Supervised and unsupervised learning: classification, regression, clustering.
Classic algorithms: decision trees, support vector machines, K-means, basic neural networks.

Deep Learning and deep neural networks:

Fundamentals of Deep Learning: multilayer neural networks, activation functions, and backpropagation.
Advanced architectures: convolutional neural networks (CNN), recurrent neural networks (RNN), Long Short-Term Memory (LSTM).
Training and optimization techniques: dropout, batch normalization, optimizers such as Adam and SGD.

Reinforcement Learning and applications:

Basics of Reinforcement Learning: agents, environments, rewards, and action policies.
Key algorithms: Q-learning, Deep Q-Networks (DQN), actor-critic policies (A2C, PPO).
Applications of RL in networking: network resource optimization, intelligent routing, traffic management.

Applications in the network domain: Participants will explore the application of the aforementioned techniques in real-world contexts related to communication networks. Some examples include:

Network performance optimization: using ML algorithms to predict and improve service quality.
Traffic management: applying DL and RL for dynamic network resource allocation.
Network security: detecting intrusions and cyber-attacks through predictive models.
Network automation: using Reinforcement Learning to automate the control of complex networks.

Textbook Information

"Pattern Recognition and Machine Learning", Christopher M. Bishop
"Deep Learning", Ian Goodfellow, Yoshua Bengio e Aaron Courville
"Understanding Machine Learning: From Theory to Algorithms", Shai Shalev-Shwartz e Shai Ben-David
"Reinforcement Learning: An Introduction", Richard S. Sutton e Andrew G. Barto

Course Planning

	Subjects	Text References
1	Machine Learning Fundamentals	1,3
2	Deep Learning	2
3	Reinforcement Learning and Applications	4
4	Networking Applications of AI

Learning Assessment

Examples of frequently asked questions and / or exercises

Explain the difference between supervised and unsupervised learning, providing examples of applications in the context of communication networks.

How does the Support Vector Machine (SVM) algorithm work, and how can it be applied to classify network traffic?

Describe the backpropagation process in a deep neural network and its importance in model training.

What are the main differences between Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN)? Provide an example of an application of each in the network domain.

How is the Q-learning algorithm used in Reinforcement Learning to manage network resource allocation?

Explain how dropout and batch normalization improve the training of a deep neural network.

How can Deep Learning be used to enhance network security and detect intrusions?

What role do optimizers like Adam or SGD play in the training of Deep Learning models, and how do they affect convergence?

Describe a practical application of Reinforcement Learning in network automation.

What challenges arise when applying Machine Learning to real-time network traffic management?

VERSIONE IN ITALIANO

Corso di laurea magistrale in

Communications Engineering