Gaussian Join Tree classifiers with applications to mass spectra classification
Publication Type:
Conference PaperSource:
6th European Workshop on Probabilistic Graphical Models, DECSAI, University of Granada, Granada, p.19-26 (2012)ISBN:
978-84-15536-57-4URL:
http://leo.ugr.es/pgm2012/proceedings/proceedings.pdfAbstract:
Classi?ers based on probabilistic graphical models are very e?ective. In continuous domains, parameters for those classi?ers are usually adjusted by maximum likelihood. When
data is scarce, this can easily lead to over?tting. Nowadays, models are sought in domains
where the number of data items is small and the number of variables is large. This
is particularly true in the realm of bioinformatics. In this work we introduce Gaussian
Join Trees (GJT) classi?ers to try to partially overcome this issue by performing exact
bayesian model averaging over the parameters. We use two di?erent mass spectra classi?cation datasets for cancer prediction to compare GJT classi?ers with those learnt by
maximum likelihood.
Improving function filtering for computationally demanding DCOPs
Publication Type:
Conference PaperSource:
Workshop on Distributed Constraint Reasoning at IJCAI 2011, Barcelona, p.99-111 (2011)Abstract:
In this paper we focus on solving DCOPs in computationally demanding scenarios. GDL optimally solves DCOPs, but requires exponentially large cost functions, being impractical in such settings. Function filtering is a technique that reduces the size of cost functions. We improve the effectiveness of function filtering to reduce the amount of resources required to optimally solve DCOPs. As a result, we enlarge the range of problems solvable by algorithms employing function filtering.
Divide and Coordinate: solving DCOPs by agreement
Publication Type:
Conference PaperSource:
Proc. of the 9th International Joint Conference on Autonomous Agents and Multi-Agent Systems (AAMAS'10), IFAAMAS, Canada, p.149-156 (2010)Keywords:
DCOP; Multi-agent Optimization; Divide and Coordinate; DaCSA; approximate algorithms; boundsAbstract:
In this paper we investigate an approach to provide approximate, anytime algorithms for DCOPs that can provide quality guaran- tees. At this aim, we propose the divide-and-coordinate (DaC) ap- proach. Such approach amounts to solving a DCOP by iterating (1) a divide stage in which agents divide the DCOP into a set of simpler local subproblems and solve them; and (2) a coordinate stage in which agents exchange local information that brings them closer to an agreement. Next, we formulate a novel algorithm, the Divide and Coordinate Subgradient Algorithm (DaCSA), a computational realization of DaC based on Lagrangian decompositions and the dual subgradient method. By relying on the DaC approach, DaCSA provides bounded approximate solutions. We empirically evaluate DaCSA showing that it is competitive with other state-of- the-art DCOP approximate algorithms and can eventually outperform them while providing useful quality guarantees.
