Banca de DEFESA: PEDRO HENRIQUE PARREIRA

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
DISCENTE : PEDRO HENRIQUE PARREIRA
DATA : 20/01/2020
HORA: 14:00
LOCAL: Auditório, 8º andar, Bloco B, Campus SA da Fundação Universidade Federal do ABC, localizada na Avenida dos Estados, 5001, Santa Terezinha, Santo André, SP
TÍTULO:

Active learning in data streams with intermediary latency

PÁGINAS: 64
GRANDE ÁREA: Ciências Exatas e da Terra
ÁREA: Ciência da Computação
SUBÁREA: Metodologia e Técnicas da Computação
RESUMO:

A data stream has several challenging features for classic machine learning algorithms, such as the effects of concept drift and evolution, characterized by changing data distribution and the emergence of new classes along the stream, respectively. Additionally, along the stream, data arrives continuously and uninterruptedly, and as a result, a massive amount of data is produced. In many data mining applications, obtaining labels is a costly task, and in a data stream, such a task becomes even more challenging due to the massive amount of data generated. Therefore, assuming that all data will have their respective labels available becomes unlikely for many real data stream applications. Hence, when considering such a scenario, a classification model has its predictive performance negatively affected throughout the data stream due to the phenomena of concept drift and evolution. In this case, it is necessary to adapt the classification model from time to time to maintain its predictive performance, requiring additional labeled data. However, given a realistic scenario, it is not always feasible for labels to be readily available, as is commonly assumed in most approaches found in the literature. Methods in the literature consider that when some example is made available, its label is provided shortly after its prediction, i.e. the label is available with null latency, and both are used in model adaptation. However, in many real applications, the label is provided with a non-infinite delay, i.e., with intermediate latency. In the midst of such challenges, this work aims to investigate and develop a set of active learning strategies that consider the intermediate latency data stream scenario.

MEMBROS DA BANCA:
Presidente - Interno ao Programa - 1673092 - RONALDO CRISTIANO PRATI
Membro Titular - Examinador(a) Interno ao Programa - 3008222 - PAULO HENRIQUE PISANI
Membro Titular - Examinador(a) Externo à Instituição - ELAINE RIBEIRO DE FARIA - UFU
Membro Suplente - Examinador(a) Interno ao Programa - 2376122 - THIAGO FERREIRA COVOES
Membro Suplente - Examinador(a) Externo à Instituição - DIEGO FURTADO SILVA - UFSCAR