Convite da Defesa do Projeto de Tese do Programa de Pós-Graduação em Ciência da Computação

A Coordenação do Programa de Pós-Graduação em Ciência da Computação tem a satisfação de convidá-lo para a Defesa de Projeto de Tese:

Compression-Based Unsupervised Sampling for Learning to Rank

Rodrigo de Magalhães Silva


Learning to Rank (L2R) methods can be used to improve ranking quality in a variety of applications. Yet a major hurdle to the adoption of L2R on private and public collections is the investment required to produce good training sets. It is possible to use Active Learning (AL) algorithms to produce smaller training sets. But these methods can be difficult to implement and may not produce good results depending on characteristics of the collections or data being sampled. They are also not very practical, requiring constant supervision and availability of human assessors. We propose an unsupervised sampling technique that relies on very general characteristics of L2R datasets. We show how this method has several advantages over AL methods for L2R, making it much easier for practitioners to use L2R on their systems and applications. Finally, based on a detailed analysis of the method's inner workings, we provide information-theoretic insights on why it is so effective.

Comissão Examinadora:


Prof. Marcos André Gonçalves - Orientador (DCC - UFMG)

Prof. Mario Sérgio Ferreira Alvim Júnior - Coorientador (DCC - UFMG)

Prof. Rodrygo Luis Teodoro Santos (DCC - UFMG)

Prof. Fabricio Murai Ferreira (DCC - UFMG)

Prof. Ricardo da Silva Torres (IC - UNICAMP)


22 de Outubro de 2019



Sala 2077 do ICEX