B8313 - BIG DATA FOR COMMUNICATIONS M

Learning outcomes

In the first part of the course students are introduced to the fundamental concepts of Big Data, machine learning, and neural networks. The course also covers key aspects of probability theory and graph theory, which provide the foundation for understanding probabilistic graphical models (e.g., Bayesian networks). These models are widely used in industrial applications, including information validation, fault isolation, and reliability analessi, all of which are becoming increasingly relevant in modern contexts. The second part of the course focuses on more advanced topics in artificial intelligence, such as energy based models (e.g., Boltzmann machines). Students also gain practical experience in applying machine learning and other AI techniques to enhance the performance of communication systems by leveraging the vast amount of data available in contemporary wireless communication environments.

Course contents

Please discard the text contained in the box "Learning Outcomes", as it is outdated. The program has changed with respect to the previous years and is organised as follows.

The course introduces the fundamental techniques used to analyse big application and network datasets, for the sake of advanced service provision and network management/optimisation, respectively, with reference to 5G/6G mobile radio networks. The data analysis techniques introduced, belong to the field of numerical optimisation, learning and AI (Artificial Intelligence).

Half of the course will be dedicated to hands-on and lab activities. Real world data will be used.

The course is split in two modules.

The first module will first describe the protocol and network architecture of 5G, then it will introduce optimisation, learning and neural network basics, with the goal to analyse data from a real world industrial 5G robotic application. Lab and hands-on sessions are included (about 10 hours).

The second module will introduce to 5G Key Performance Indicators (KPIs) and MDT data; then, more advanced AI techniques will be described and all practical activities will be geared towards the use of measured KPIs and MDT data to identify features in network data and exploit them to optimise some 5G RAN parameters. Lab and hands-on sessions are included, using real world KPI/MDT data from a 5G network (about 20 hours).

More details follow.

First Module.

1.1 Introduction to 5G.

1.2 Introduction to industrial AI based 5G/6G use cases.

1.3 Data-driven AI: training and test, supervised (regression, classification), unsupervised.

1.4 Machine Learning: linear/logistic regression, decision tree, random forests Deep Learning: artificial neuron, neural networks, back propagation, CNN, AutoEncode.

1.5 Data Analysis & Engineering: data vis, feature creation, missing/anomalous values, imbalanced data.

1.6 Description of an Industrial IoT Case Study.

1.7 Lab IIoT 1: Data Analysis and Engineering for IoT.

1.8 Lab IIoT 2: Deep Learning for fault prediction.

1.9 Hands-On Session: 1D-CNNs for IoT / mobile networks.

1.10 Hands-On Session: AutoEncoders for IoT / mobile networks.

Second Module.

2.1 Introduction to 5G KPIs and MDT Data.

2.2 Lab 5G 1: MDT & KPIs Data Analysis and Engineering.

2.3 Lab 5G 2: Radio KPIs regression.

2.4 Lab 5G 3: MDT to predict Radio KPIs

2.5 Fingerprinting Positioning, wKNN.

2.6 Lab 5G 4: MDT for Positioning.

2.7 Hands-On Session: RayTracing, Sionna, 6G Digital Twins.

2.8 Hands-On Sessions: 3D-CNNs.

2.9 Seminars on future uses of AI in 5G/6G networks.

Readings/Bibliography

Slidesets, databases and software are provided by the instructor.

Teaching methods

About 30 hours: frontal lectures.

About 20 hours: laboratory activities; students will be assigned tasks and will perform them on their laptop under supervision by the instructor. They will use datasets and software libraries provided by the instructor. It is expected that students are familiar with the use of Python coding.

About 10 hours: hands-on sessions: students will be exposed to software developments made and discussed by the instructor.

Assessment methods

At the end of each module, a questionnaire with multiple choices plus the assignment of the task to analyse a dataset will be proposed.

Teaching tools

datasets and software libraries

Office hours

See the website of Roberto Verdone