Celoslovenské podujatie na podporu rozvoja IT vo výskume, vzdelávaní a podnikaní na Slovensku.
Podujatie sa uskutočňuje v dňoch 1. 9. - 31. 10. 2018.
Pre rok 2018 sa hlavným mestom informatiky stala Žilina.
Viac na stránke www.extrapolacie2018.sk
V priestoroch FIIT STU v Bratislave sme pripravili:
1. októbra 2018 o 10.00 hod.
Aula Minor (-1.65 na 1. PP)
prof. Marián Vajteršic,
(Paris-Lodron-University Salzburg, Rakúsko a SAV, Bratislava, Slovensko)
Some Examples and Notes on Algorithms for Parallel Computers
(prezentácia je v angličtine, prednáška v slovenčine)
Parallel computing is nowadays a mainstream in computational practice. We show on some exaples, that without systems with a huge number of processing nodes it is not possible to obtain feasible solutions of challenging problems in science and technology. In this context, a note about ranking of current supercomputers systems and about the global performance development will be mentioned.
The development of parallel computing from historical perpective will be shortly summarized for the former Institute of Technical Cybernetics of the Slovak Academy of Sciences in Bratislava. Our focus will be devoted to review an intensive work in algorithmic development for the parallel computer PPS SIMD. It will be a possibility to show one example of parallelization of an strictly sequential algorithm for this specialized architecture.
In second part of the talk we wil present a current research related to parallelization of decomposition of nonnegative matrices. Nonnegative matrices arise naturally in various applications. These include text mining, document classification, clustering, spectral data analysis, face recognition and also problems in non-informatics areas like e.g. computational biology. An efficient technique to approximate a nonnegative matrix as a product of two nonnegative matrices of smaller size represents the Nonnegative Matrix Factorization (NMF), which has a nice property that also the factors are guaranteed to be nonnegative.
We show the application of NMF for image reconstruction. It will be demonstrated how the quality of the reconstruction depends on the factorization parameter and which computational and memory costs are related to it. Since these are high for large-size images, this problem is also a good candidate for parallelism. Unfortunately, there is little related work until now on parallel distributed NMF. We report on our parallelization for NMF, which is based on Newton-like iteration. It was adapted for parallel execution on computer clusters, utilizing the message passing communication paradigm. The measurements show that for sufficiently large workloads, the parallelized Newton iteration algorithm achieves an almost linear speedup, which makes it a promising candidate for large-scale NMF computations.