On Using Sum-Product Networks For Multi-Label Classification

Autor(es)

Llerena, Julissa Villanueva ; Maua, Denis Deratani ; IEEE

• Tipo de documento: Artigo Científico
• Fonte: 2017 6TH BRAZILIAN CONFERENCE ON INTELLIGENT SYSTEMS (BRACIS); v. N/A, p. 6-pg., 2017-01-01.

Resumo

Multi-Label classification consists of mapping each object to a set of relevant labels. A successful approach to constructing Multi-Label classifiers is to obtain a probabilistic model of the relation between object attributes and labels. This model can then be used to classify objects by computing the most probable configuration of label relevance variable conditioned on the attributes. Sum-Product Networks (SPN) are deep probabilistic models that have shown promising results in many tasks. As with many other probabilistic models, performing Most Probable Explanation (MPE) inference is NP-hard in SPNs. In this work we investigate the use of SPNs for Multi-Label classification. We compare different approaches for learning SPNs and computing MPE. We show that SPN-based Multi-Label Classifiers are competitive against state-of-the-art classifiers in a collection of real-world experiments. (AU)

• Processo FAPESP: 16/01055-1 - Aprendizagem de Modelos Probabilísticos Tratáveis e seu Uso na Classificação Multirrótulo
• Beneficiário: Denis Deratani Mauá
• Modalidade de apoio: Auxílio à Pesquisa - Regular