ECTS - University of Pisa
| Modules | Area | Type | Hours | Teacher(s) | |
| SYMBOLIC AND EVOLUTIONARY ARTIFICIAL INTELLIGENCE | ING-INF/05 | LEZIONI | 60 |
|
Programma disponibile solo in lingua inglese, essendo il corso tenuto in inglese.
Pertanto occorre cliccare sul pulsante accanto, per avere il programma, la modalità d'esame, eccetera.
Students are expcted to be able to understand the principles of multi-objective optimization and to develop their own algorithm for multi-objective evolutionary optimization.
Moreover, they will be able to understand and apply the principles of reinforcement learning, to solve complex learning problems there the alternative learning methods struggle.
Students are also expected to be able to use Non-Archimedean algorithms to numerically solve lexicographic multi-objective problems, including game theory problems involving priorities among the payoffs.
Finally, the studends are expected to be aware of the challenges posed by the design of hardware accelerators for machine learning and neural networks, with special emphasis on the use of alternative representations for small-precision real numbers, vectorized CPUs, etc.
The students must present a project. The project will be evaluated. If the evaluation is positive, then the students can access the oral examination. The positive outcome of the oral examination concludes the exam.
Students will be able to extend existing software libraries, or implement one from scratch.
Students will be able to present, in a written report the results of their activity.
During the computer lab sessions, small projects will be carried out in order make practice over the theoretical concepts acquired during the theoretical lessons.
Students will be able to manage the responsibility of managing a team project
During the lab sessions, the accuracy and precision of the activities carried out will be evaluated.
Having attended the course of "Optimization Methods and Game Theory" is an asset, although not a mandatory one.
The same concerning the course of "Computational Intelligence and Data Mining".
This course aims at providing students with a unifying overview about modern artificial
intelligence. First of all symbolic artificial intelligence is introduced, along with the depth-first
and breadth-first exploration methods. Then the concept of agent is covered, together with the introduction of multiple agent systems as a unifying model of many distributed AI systems regularly used today. In particular, swarm and evolutionary intelligence are two paradigms based on multi-agent systems, which have proved to be very effective in solving applications that require a distributed approach.
Then the theory of Reinforcement Learning will be covered, along with lab sessions.
The final part of the course is devoted to advanced topics in artificial intelligence, such as: how to speed up deep neuro-fuzzy networks (using novel representations for real numbers and implementing the associated hardware accelerators), the design of one-class classifiers, and the implementation of neural networks with infinitesimal or infinite weights, to solve lexicographic multi-objective learning tasks, etc.
Kalyamnoy Deb, "Multi-objective optimization using evolutionary algorithms", 2005.
Richard Sutton, Andrew G. Barto, "Reinforcement Learning: An Introduction", second edition, 2018.
The students must present a project. The project will be evaluated. If the evaluation is positive, then the students can access the oral examination. The positive outcome of the oral examination concludes the exam.
(none)
