DATA DRIVEN CONTROL DESIGN

International Teaching DATA DRIVEN CONTROL DESIGN

0622700085
DIPARTIMENTO DI INGEGNERIA DELL'INFORMAZIONE ED ELETTRICA E MATEMATICA APPLICATA
EQF7
COMPUTER ENGINEERING
2021/2022

YEAR OF COURSE 2
YEAR OF DIDACTIC SYSTEM 2017
PRIMO SEMESTRE
CFUHOURSACTIVITY
324LESSONS
324LAB
Objectives
THE COURSE WILL BE FOCUSED ON THE DESIGN OF AUTONOMOUS CLOSED-LOOP SYSTEMS FROM DATA. AT THE END OF THE COURSE, THE STUDENT WILL BE ABLE TO DESIGN AUTONOMOUS AGENTS CAPABLE TO MAKE OPTIMAL DECISIONS (I.E. TO SYNTHESIZE CONTROL POLICIES) PURELY FROM DATA. THESE POLICIES WILL BE ABLE TO GUARANTEE KEY DESIRED PROPERTIES FOR THE CLOSED-LOOP SYSTEM, SUCH AS SAFETY AND PERFORMANCE. THE STUDENT WILL BE INTRODUCED TO KEY STATE-OF-THE-ART ALGORITHMS AND WILL MAKE HANDS-ON EXPERIENCE ON MODERN DEVELOPMENT TOOLS AND ENVIRONMENTS. THROUGHOUT THE MODULE, THE METHODOLOGICAL ASPECTS WILL BE COMPLEMENTED WITH THE IN-CLASS DEVELOPMENT OF INTERESTING CASE STUDIES IN THE FIELD OF AUTONOMY.

KNOWLEDGE AND UNDERSTANDING
- FORMULATION OF DATA-DRIVEN CONTROL PROBLEMS
- MAIN ALGORITHMS FOR DATA-DRIVEN CONTROL
- PROPERTIES OF THE ALGORITHMS

APPLYING KNOWLEDGE AND UNDERSTANDING
- EVALUATION OF DESIGN TRADE-OFFS FOR DATA-DRIVEN CONTROL ALGORITHMS
- IMPLEMENTATION OF DATA-DRIVEN CONTROL ALGORITHMS
- VALIDATION VIA SIMULATION ENVIRONMENTS
Prerequisites
FOR THE SUCCESSFUL ACHIEVEMENT OF THE COURSE GOALS, KNOWLEDGE ON CLOSED-LOOP CONTROL SYSTEMS IS REQUIRED. THIS KNOWLEDGE CAN BE ACQUIRED IN THE COURSES: AUTOMAZIONE.
Contents
INTRODUCTION TO DATA-DRIVEN CONTROL (LECTURE/PRACTICE/LABORATORY HOURS 8/4/2)
BASIC NOTIONS OF MARKOV DECISION PROCESSES, BAYESIAN STATISTICS AND OPTIMIZATION ALGORITHMS, EXAMPLES

THE FUNDAMENTAL PROBLEM OF DECISION MAKING (LECTURE/PRACTICE/LABORATORY HOURS 6/4/2)
FORMULATION AND ITS ALGORITHMIC SOLUTION TECHNIQUES, A COMMON FRAMEWORK COVERING MODEL-BASED AND MODEL-FREE CONTROL, EXAMPLES OF LINKS WITH POPULAR CONTROL METHODS SUCH AS MODEL PREDICTIVE CONTROL (MPC) AND LINEAR QUADRATIC REGULATOR (LQR), APPLICATIONS

FULLY DATA-DRIVEN DESIGN (LECTURE/PRACTICE/LABORATORY HOURS 8/4/2)
THE BAYESIAN CONTROL FRAMEWORK, CONTROLLERS AS DATA GENERATORS, OPTIMAL CONTROL ALGORITHMS, CONTROL FROM DEMONSTRATIONS, MULTI-AGENT PERSPECTIVE, IMPLEMENTATION STRATEGIES, APPLICATIONS

DESIGN PERSPECTIVES (LECTURE/PRACTICE/LABORATORY HOURS 2/4/2)
GENERATION OF DATA, DESIGNING AN AGENT USING OFF-THE-SHELF TOOLS

TOTAL LECTURE/PRACTICE/LABORATORY HOURS 24/16/8


EACH OF THE ABOVE TOPICS WILL HAVE A STRONG COMPUTATIONAL AND HANDS-ON EXPERIENCE COMPONENT. IN PARTICULAR, STUDENTS WILL MAKE HANDS-ON EXPERIENCE WITH MODERN COMPUTATIONAL TOOLS AND ENVIRONMENTS. THE ALGORITHMS WILL BE ILLUSTRATED THROUGH DESIGN CASE STUDIES FROM E.G. INDUSTRY 4.0, BIOMEDICAL SYSTEMS AND EMBEDDED SYSTEMS. STUDENTS WILL ACTIVELY CONTRIBUTE TO THE DESIGN PROCESS AS PART OF THEIR LEARNING.
Teaching Methods
•LECTURES SUPPORTED BY PROBLEM-SOLVING TUTORIALS WITH PRACTICAL ASPECTS ALSO COVERED DURING LECTURES.
•IN ORDER TO PARTICIPATE TO THE FINAL ASSESSMENT AND TO GAIN THE CREDITS CORRESPONDING TO THE MODULE, THE STUDENT MUST HAVE ATTENDED AT LEAST 70% OF THE HOURS OF ASSISTED TEACHING ACTIVITIES.
Verification of learning
STUDENTS WILL BE ASSESSED VIA A DISCUSSION ON PROJECTS/COURSEWORKS THAT THEY WILL DEVELOP. THE DISCUSSION OF THE PROJECTS/COURSEWORKS WILL BE AIMED AT VERIFYING THE UNDERSTANDING OF THE METHODOLOGICAL ASPECTS AND THEIR APPLICATION.

THE EXAM INCLUDES THE DEVELOPMENT OF A PRESENTATION FROM THE STUDENT.
Texts
RECOMMENDED READINGS:
• RICHARD S. SUTTON AND ANDREW G. BARTO. "REINFORCEMENT LEARNING: AN INTRODUCTION". THE MIT PRESS. 2015. DISPONIBILE ONLINE
• M. HARDT, B. RECHT, “PATTERNS, PREDICTIONS AND ACTIONS: A STORY ABOUT MACHINE LEARNING”.SELF PUBLISHED. 2021. DISPONIBILE ONLIN

SUPPLEMENTARY TEACHING MATERIAL WILL BE AVAILABLE ON THE UNIVERSITY E-LEARNING PLATFORM (HTTP://ELEARNING.UNISA.IT) ACCESSIBLE TO STUDENTS USING THEIR OWN UNIVERSITY CREDENTIALS.
More Information
THE COURSE IS HELD IN ENGLISH
  BETA VERSION Data source ESSE3