Foto 7

Prof. Mauro Barni, University of Siena - Italy - "Information Theory and Statistics", 22,23,25,26 February, 1 March 2021

18-20 hours (5 credits)


From remote by using Google Meet. The link will be sent in due time to all students who registered to the seminar.

To register to the course, click here


Short Abstract:
Information theory has been celebrated mostly for its success in modelling information sources and communication systems, however it can be applied to many other application areas always bringing useful and somewhat unexpected insights. The relationship between information theory and statistics is one of such cases. As a matter of fact, some of the most famous results in hypothesis testing and large deviation theory can be revisited from an information theoretic perspective leading to new findings and a deeper understanding of the involved concepts. It is the goal of this course to provide a brief introduction to the use of information theoretic concepts in statistics and, conversely, use some well-known results in statistics to revisit the most celebrated theorems of information theory. In the last part of the course the concepts developed in the first lectures are applied to build a theory of adversarial hypothesis testing, aiming at determining the ultimate achievable performance when hypothesis testing is cast into an adversarial setting encompassing the presence of an adversary aiming at inducing an error in the test. The links of the theory with adversarial machine learning and AI security will discussed with examples drawn from the multimedia forensics field.

Course Contents in brief:

  1. Information theory in a nutshell
  2. The method of types and its relationship with statistics
  3. Application to large deviation theory and hypothesis testing
  4. Adversarial hypothesis testing: an information theoretic perspective
  5. Links with adversarial machine learning and multimedia forensics


22,23,25,26 February, 1 March 2021 - 9:00-13:00