Room “Sala Seminari” - Abacus Building (U14)
Foundation and recent advances
of Geometric Semantic Genetic Programming
Speaker Prof. Leonardo Vanneschi
NOVA IMS, Universidade Nova de Lisboa, Lisbon, Portugal
Associate Dean for Research
Director of the MagIC Research Center
Coordinator of the Data Analytics Lab
Abstract
Geometric Semantic Genetic Programming (GSGP) introduced a major paradigm shift in Genetic Programming by defining semantic-aware operators, capable of inducing a unimodal error surface for supervised learning problems. This property enabled remarkable optimization performance and, in some cases, strong generalization ability across several benchmark problems and real-world applications. However, traditional GSGP also suffered from a major limitation: the continuous growth of program size, producing models that are so large that they are impossible to be interpreted by humans. This seminar presents a unified research trajectory addressing the challenge, from the theoretical foundations and efficient implementation of GSGP to the recent development of SLIM_GSGP, a novel framework capable of evolving compact and potentially interpretable models, while preserving the essential geometric properties of GSGP. SLIM_GSGP is a novel paradigm, with still several open issues and much room for research, promising to open new perspectives for explainable and theoretically grounded evolutionary machine learning.
Short Bio
Leonardo Vanneschi is a Full Professor at NOVA IMS. His main research interests involve Machine Learning, Data Science, Complex Systems, and in particular Evolutionary Computation. His work can be broadly partitioned into theoretical studies on the foundations of Evolutionary Computation, and applicative work. The former covers the study of the principles of functioning of Evolutionary Algorithms, with the final objective of developing strategies to outperform the traditional techniques. The latter covers several different fields among which computational biology, image processing, personalized medicine, engineering, logistics, economics and marketing. His work has been consistently recognized and appreciated by the international community from 2000 to nowadays. In 2015, he was honoured with the Award for Outstanding Contributions to Evolutionary Computation in Europe, in the context of EvoStar, the leading European Event on Bio-Inspired Computation. In 2020, he was included in the list of the 2% best researchers in the world, both for the year 2019 and for the entire career, according to a study conducted by the Stanford University.
contact person for this Seminar: daniela.besozzi@unimib.it