Elizeu (doutorando da University of British Columbia e ex-lsdano ) está nos visitando por algumas semanas. Programamos duas palestras sobre as pesquisas sendo desenvolvida por ele. A primeira palestra será hoje, a segunda quarta-feira, ambas às 16h.
Segue abaixo informações sobre o que será conversado
Abstract: The goal of this project is to understand the challenges in
porting graph algorithms to commodity, hybrid platforms; platforms
that consist of processors optimized for sequential processing and
accelerators optimized for massively-parallel processing. This study
fills the gap between current graph processing platforms that are
either expensive (e.g., supercomputers) or ine cient (e.g., commodity
clusters). Our hypothesis is that hybrid platforms (e.g.,
GPU-supported clusters) can bridge the performance-cost chasm, and
o er an attractive graph-processing solution for many graph-based
applications such as social networks and web analysis.
This work presents the first step towards designing Totem (a
graph-processing framework that leverages massively parallel hybrid
platforms. In particular, we design, implement, and evaluate core
graph algorithms (i.e., BFS, Dijkstra's algorithm, and PageRank).
Also, we discuss the future work based on the current experience
provided by these initial implementations.
(joint work with Lauro Beltrão Costa and Abdullah Gharaibeh)
Users create massive amounts of content in social media websites such as YouTube, Flickr, and del.icio.us. The increasingly large amount of content these systems accumulate poses a challenge to users that want to navigate the vast amount of content produced daily. To distribute the burden of organizing the collections of user-generated content, and to make navigation potentially more efficient, social systems often provide users with tagging features (i.e., users can annotate content they produce or find interesting with free-from keywords). These tags can, in turn, help users to sift through the set of items available in the system.
Similar to other commons-based peer-production systems, the efficiency of tagging rely on the individial contributions of participants. On the other hand, as opposed to systems desinged for resource sharing (e.g., BitTorrent, OurGrid), quantifying the value of contributions in tagging systems poses different challenges, as users produce information instead of physical resources.
This work investigates methods to assess the value of tags from the perspective of a user who is interested in navigating the system. The long term goal is to understand the aspect that influence the perceived value of tags and to inform the design of new mechanisms (e.g., incentive to produce more socially useful tags, or spam detection). In particular, in this talk I will provide an overview of the progress to date towards the aforementioned goal.
Segue abaixo informações sobre o que será conversado
Palestra 1: Massively-Parallel Graph Processing
Terça, 16h, auditório do LSD
Terça, 16h, auditório do LSD
Abstract: The goal of this project is to understand the challenges in
porting graph algorithms to commodity, hybrid platforms; platforms
that consist of processors optimized for sequential processing and
accelerators optimized for massively-parallel processing. This study
fills the gap between current graph processing platforms that are
either expensive (e.g., supercomputers) or ine cient (e.g., commodity
clusters). Our hypothesis is that hybrid platforms (e.g.,
GPU-supported clusters) can bridge the performance-cost chasm, and
o er an attractive graph-processing solution for many graph-based
applications such as social networks and web analysis.
This work presents the first step towards designing Totem (a
graph-processing framework that leverages massively parallel hybrid
platforms. In particular, we design, implement, and evaluate core
graph algorithms (i.e., BFS, Dijkstra's algorithm, and PageRank).
Also, we discuss the future work based on the current experience
provided by these initial implementations.
(joint work with Lauro Beltrão Costa and Abdullah Gharaibeh)
Palestra 2: Assessing the Value of Tags in Social Tagging Systems
Quarta, 16h, auditório do LSD
Quarta, 16h, auditório do LSD
Similar to other commons-based peer-production systems, the efficiency of tagging rely on the individial contributions of participants. On the other hand, as opposed to systems desinged for resource sharing (e.g., BitTorrent, OurGrid), quantifying the value of contributions in tagging systems poses different challenges, as users produce information instead of physical resources.
This work investigates methods to assess the value of tags from the perspective of a user who is interested in navigating the system. The long term goal is to understand the aspect that influence the perceived value of tags and to inform the design of new mechanisms (e.g., incentive to produce more socially useful tags, or spam detection). In particular, in this talk I will provide an overview of the progress to date towards the aforementioned goal.
