Explaining Completions Produced by Embeddings of Knowledge Graphs

Autor(es)

Ruschel, Andrey ; Gusmao, Arthur Colombini ; Polleti, Gustavo Padilha ; Cozman, Fabio Gagliardi ; KernIsberner, G ; Ognjanovic, Z

• Tipo de documento: Artigo Científico
• Fonte: SYMBOLIC AND QUANTITATIVE APPROACHES TO REASONING WITH UNCERTAINTY, ECSQARU 2019; v. 11726, p. 12-pg., 2019-01-01.

Resumo

Advanced question answering typically employs large-scale knowledge bases such as DBpedia or Freebase, and are often based on mappings from entities to real-valued vectors. These mappings, called embeddings, are accurate but very hard to explain to a human subject. Although interpretability has become a central concern in machine learning, the literature so far has focused on non-relational classifiers (such as deep neural networks); embeddings, however, require a whole range of different approaches. In this paper, we describe a combination of symbolic and quantitative processes that explain, using sequences of predicates, completions generated by embeddings. (AU)

• Processo FAPESP: 16/18841-0 - Algoritmos para inferência e aprendizado de programas lógicos probabilísticos
• Beneficiário: Fabio Gagliardi Cozman
• Modalidade de apoio: Auxílio à Pesquisa - Parceria para Inovação Tecnológica - PITE