Profile cover photo
Profile photo
Eliezer de Souza da Silva
Eliezer's interests
View all
Eliezer's posts

Post has attachment
Our paper "Large-Scale Distributed Locality-Sensitive Hashing for
General Metric Data" (Eliezer Silva, Thiago Teixeira, George Teodoro, and Eduardo Valle) was accepted at SISAP 2014 (Conference on Similarity Search and Applications). 
The idea of our hashing is to partition the data in Dirichlet domains using a collection of reference points (random and/or learned by a k-medoid algorithm); so it is a version of to the compact partition approach. Thus the hashing is the number of the partition (the number of the closest reference point). The basic idea is that nearest-neighbor points have high probability of being assigned to the same partition (or sharing the same closest point in the reference set). In this work we extend our algorithm to a distributed-memory system using dataflow programming paradigm, decomposing Voronoi LSH in five computing stages, which exploits task, pipeline, replicated
and intra-stage parallelism.

More details in the preprint:

ps1. This work does not develop the theory of the method; the theoretical results are only in my M.Sc. dissertation (which I will be presenting soon). 
ps2. I will not attend the conference, but Prof. Eduardo Valle (my advisor) will be there =)

Post has attachment
<a class='ot-hashtag' href=''>#hangoutsonair</a>Sydnei Melo and Pedro Grabois

Post has attachment

Post has shared content
Wait while more posts are being loaded