Instructor: Francesco Delli Priscoli
Course web page: www.diag.uniroma1.it/~dellipri/ccen
Credits: 6
Infostud code: 1041429

Objectives

The course aims at applying advanced dynamic control methodologies to networks/systems by adopting a technology-independent abstract approach that copes with the network/system control problem, leaving out of consideration the specific network/system technologies. The students will be able to design network control actions suitable for communication, energy, transport, security, health systems/networks.

Program

The first part of the course details the following methods: Markov Decision Process, Dynamic Programming, Reinforcement Learning (in particular, TD learning, Sarsa, Q-learning), Machine Learning (k-means clustering, clustering). Besides the theoretical aspect, the course considers the practical use of such methods for the control of communication, transport, security, health systems/networks.
The second part of the course, held in constant synergy with the research projects funded by the European Union, (i) provides an overview of up-to-date control problems related to communication, energy, transport, security and health, (ii) details how the control methods considered in the first part of the course, as well as other control methods introduced in previous courses (e.g. Model Predictive Control) can be used to solve the above-mentioned problems.

Type of exam: Written question on theoretical aspects plus evaluation of a project.

Reference texts

• R.S. Sutton and A.G. Barto, "Reinforcement Learning: An Introduction," MIT Press, 1998.
• Cristopher M. Bishop, "Pattern Recognition and Machine Learning", Springer, 2006.
• Lecture notes mainly derived from Deliverables produced by up-to-date EU research projects. In particular, the projects ATENA (UE FP8), SESAME (UE FP8), 5G ALLSTAR (UE FP8), 5G SOLUTIONS (UE FP8), ARIES (ESA), VADUS (ESA).