The DMIP group focussed initially on the learning
of comprehensible models on different paradigms: Functional Logic
Programs (FLP), decision trees and decision lists.
The team was formed in 1997 with the
initial goal of extending ILP to other declarative languages, and,
since then, has been acquiring a broader view of the field, exploring
different techniques and applications in the broader areas of machine learning and data mining.
Currently, the research areas are:
Machine Learning and Data Mining
Knowledge Discovery
Multi-paradigm Inductive Programming
ROC analysis, cost-sensitive
learning and model evaluation for decision support
Agreement Technologies.
Agent Intelligence Evaluation.
MML induction and Solomonoff prediction.
Probabilistic (inductive) programming.
Inductive Debugging.
Nonetheless, the primary interest is still
the learning of comprehensible or declarative models from data.
For a more comprehensive account of the team's activities and projects, you can take a look a this presentation.
Learning
Systems and software:
We have developed three learning systems:
The FLIP system (1998-2001, click here for downloading the software and for
more information): implements a framework for the Induction of
Functional Logic Programs (IFLP) from facts. This can be seen as an
extension to the now consolidated field of Inductive Logic Programming
(ILP). Inspired in the inverse resolution operator of ILP, the system
is based on the reversal of narrowing, the more usual operational
mechanism for Functional Logic Programming. The main advantages of the FLIP system over the most used ILP systems are a
natural handling of functions, without the use of mode or determinism
declarations, and its power for inducing short recursive programs. Its
applications are mainly program synthesis, program debugging and data
mining of small highly structured documents.
The SMILES system (2001-2002, click here for downloading the software and for
more information): a machine learning system that integrates many
different features from other machine learning techniques and paradigms
and, more importantly, it presents several innovations in almost all of
these features. In particular, SMILES extends classical decision tree learners in
many ways (new splitting criteria, non-greedy search, new partitions,
extraction of several and different solutions), it has an anytime
handling of resources, and has a sophisticated and quite effective
handling of costs. In this way, SMILES combines and improves the recent interest in
hypotheses combination (e.g. boosting) and cost-sensitive learning (a
priori and a posteriori class assignments, ROC analysis) outperforming
previous systems in many situations. Its applications are basically
data-mining and any other machine learning task where decision trees
could be useful.
The DBDT system (2004-2010, click here for downloading the software and for
more information): is a machine learning algorithm that integrates decision tree learning and center splitting.
Roughly speaking, the inferred classifer can be viewed as a tree of attribute prototypes (The value distribution
of an attribute is represented by a set of prototypes.). An instance is linked to one prototype or other depending
on its proximity. .