Creating and sharing knowledge for telecommunications

Project: Development and Learning of Kernels for Text and Images

Acronym: DeLKeTI
Main Objective: 1. Development of techniques to approximate the universal
distance for sequences and multidimensional data; this will lead to an approximate universal kernel. Using the ability of
compression algorithms to approximate the universal
distance, kernels will be tailored to each particular problem
by exploiting different characteristics of the data.
2. Application of universal kernels in learning algorithms
such as support vector machines or kernel logistic
regression.
3. Theoretical characterization of the concentration
properties (bounds) of the approximations to the universal
distance, namely by exploiting known connection to
Shannon's information.
4. Development of kernel learning techniques, in which the
distance measure is learned from data (in supervised,
semi-supervised, or non-supervised modes), rather than
specified a priori.
5. Application of the developed kernels to problems in text
and image analysis; e.g., categorization, desambiguation,
clustering.
Reference: PTDC/EEA-TEL/72572/2006
Funding: FCT/PTDC
Start Date: 01-11-2007
End Date: 01-06-2011
Team: Mário Alexandre Teles de Figueiredo, José David Pereira Coutinho Gomes Antão
Groups: Pattern and Image Analysis – Lx
Partners: Priberam Informática, SA
Local Coordinator: Mário Alexandre Teles de Figueiredo
Links: Internal Page
Associated Publications