February 3, 2000
1:30pm

Richard Nock
Université des Antilles-Guyane (France)

Learning logical formulas having limited size: theoretical aspects and algorithms (with related results)

The work I will present in ML is related to the improvement of classifier's accuracy, succinctness and interpretability. Two types of algorithms I have co-developed will be detailed: the first one preprocesses the data, the second one unifies and improves the induction of some types of concept representations. My COLT work has been focused on formal analyses of some of the ML goals presented before for various kinds of concept representations. I will present a result on compression algorithms also known as Ockham's razors, and an application of some results in Network Supervision. This talk will end by a short presentation of some recent results on Image Segmentation in optimal time/space complexities, using specifically developed statistical concentration bounds.

Biography

Richard Nock was born in France in 1970. In 1993, he received an engineering degree in Agronomy from the graduate school of engineering: ENSAM (Ecole Nationale Superieure Agronomique de Montpellier/Institut National de la Recherche Agronomique, France), and a M.Sc. in Computer Science from the LIRMM, Univ. Montpellier II (Lab. of Computer Science, Robotics and Microelectronics, France), ranking major. He received the PhD in Computer Science in 1998 from the LIRMM, and then served as Assistant Professor in the Univ. Montpellier II and the Univ. des Antilles-Guyane.

During his military service in 1996, Richard obtained contractual consulting activities in Statistics and Computer Science for a Research Center on Virtual Reality and Human Factors of the DGA (General Army Delegation).

His research topics include Machine Learning, Statistics, Algorithmic Complexity, Image Processing, with applications to domains such as Data Mining and Network Supervision. In this talk, I will synthetize some of the research works I have pursued in Machine Learning (ML), Computational Learning Theory (COLT) and Image Processing.