Seminários da Pós-Graduação em Informática
Data: 19 de outubro de 2018
Local: Sala Multiuso CIC
Horário: 14h
Palestrante: Jeremias Moreira Gomes (doutorando)
Título: Análise de Sensibilidade em Aplicações Médicas Utilizando Surrogate Models
Resumo:Sensitivity Analysis (SA) is important in image analysis workflows because it evaluates how analysis results are affected by variations in method input parameters, which helps in the development of robust and efficient image analysis methods. However, SA is expensive in the image analysis domain because of the large number of workflows runs required and the high cost of executing the workflows on large datasets. An approach that reduces the number of workflow runs in SA studies and accelerates runs on parallel machines is necessary to enable SA in large datasets.
Horário: 14h30
Palestrante: Lucas Oliveira Souza (mestrando)
Título: Accelerating Learning in Cooperative Multiagent Domains through Experience Sharing
Resumo:This work introduces a new reinforcement learning (RL) model for cooperative multiagent domains. The goal is to speed up learning by sharing experience between agents, taking advantage of the parallel exploration conducted by cooperative agents in a multiagent setting. It builds upon previous work conducted in the field of transfer learning in RL and multiagent settings.
Horário: 15h
Palestrante: Leomar Camargo de Souza (mestrando)
Título: Investigating the Safe Evolution of Delta-Oriented Product Lines
Resumo:A Software product line engineering is a well known approach for building a set of configurable systems for a specific domain; and different techniques have been used to manage product line variability, including delta-oriented programming (DOP). DOP has gained increasing attention of the academy over the last years—with contributions that go beyond variability management at the source code level. Nevertheless, little is known about the practical usage of DOP to extract and evolve a SPL. In this paper we address this issue, reporting and characterizing our experience on extracting and “partially safe” evolving two SPLs (REMINDER-PL and IRIS-PL) using DeltaJ, an existing implementation of DOP for Java. Our experience covers different evolutionary scenarios (including the introduction of optional, alternative, and or-features), source-code refactorings of delta modules, and our design decisions to deal with feature interactions that often lead to the scattering and tangling of features throughout delta modules. We present evidences that existing template catalogs for safe evolution of software product lines are general enough to represent evolutionary scenarios of a delta-oriented SPLs. We also report the results of a qualitative assessment that compared the efforts needed to extract end evolve the REMINDER-PL using both DOP and conditional compilation. Although conditional compilation requires substantially less effort to extract and evolve a SPL, DOP leads to a more modular design and simplifies the definitions of the SPL assets (including feature and configuration models). Our results bring a general understanding that these benefits supersede the costs related to designing a modular decomposition using DOP.
Profa Célia Ghedini Ralha (Este endereço de email está sendo protegido de spambots. Você precisa do JavaScript ativado para vê-lo.)
Coordenadora dos Seminários de Pós-Graduação em Informática 2018-2