Instructor: Luca Iocchi
Course web page: www.diag.uniroma1.it/~iocchi/Didattica/LAS/index.html
Credits: 6
Infostud code: 1044393

Objectives

The goal of the course is to present techniques and tools for machine learning in complex dynamic systems and autonomous agents. In particular, the course will describe probabilistic models for representing dynamic systems and autonomous agents, reinforcement learning techniques, learning in graphical models, state estimation techniques. The course will also present many examples of successful application of Machine Learning algorithms in different application scenarios. At the end of the course the student will be able to use the addressed techniques and tools in modeling and solving learning problems for complex dynamic systems. Students will gain the capability of solving complex learning problems with dynamic systems, by devising a proper formulation of the problem, performing adequate design and implementation choices, designing and executing effective experiments to evaluate the results obtained.

Program

• Introduction Typical problems for robotic applications. Basics of probabilities and linear algebra.
• Models of dynamic systems General concepts. Model taxonomy. Markov Decision Processes. Hidden Markov Models. Dynamic Bayesian Networks. Partially Observable Markov Decision Processes. Probabilistic Graphical Models.
• Reinforcement Learning Q-Learning algorithm. Non-deterministic algorithms. Inverse Reinforcement Learning. RL in plan space.
• Bayes filtering in DBN Discrete filters (forward). Particle filters.
• Learning in Probabilistic Graphical Models Learning in HMM (Baum-Welch). Learning in DBN: Estimating CPD from supervised data sets.
• Multi-Agent Learning Multi-source multi-object tracking. Multi-agent learning.

Type of exam: Written test, Project

Reference texts

• Slides and material provided by the instructor.