Seminar: Large cluster programming and Chapel language with example of evolutionary algorithm implementation

Description: The application of evolutionary algorithms is frequently limited by available computational resources, especially in the case of so-called hybrid or memetic evolutionary algorithms in combination with large (or even Big) training data sets. HPC implementations not only of such algorithms are complicated. Unified programming environment of Chapel language opens way to implement algorithms working on large clusters much easier way than in the case of frequently used combinations of OpenMP+MPI or even OpenMP+MPI+Cuda technologies. This talk presents an alternative approach to programming heterogeneous systems in the Chapel language, demonstrates its basic approaches and presents their application on evolutionary algorithm parallel implementation.

Difficulty: Intermediate

Language: English

Location: Faculty of Computer and Information Science, Večna pot 113
SI-1000 Ljubljana

Opis: Uporaba evolucijskih algoritmov je pogosto omejena z razpoložljivimi računalniškimi viri, zlasti v primeru tako imenovanih hibridnih ali memetičnih evolucijskih algoritmov v kombinaciji z velikimi ali celo ogromnimi množicami učnih podatkov. HPC izvedbe takih algoritmov so zapletene. Enotno programsko okolje jezika Chapel odpira pot za implementacijo algoritmov, ki delujejo na velikih grozdih, veliko lažje kot v primeru pogosto uporabljenih kombinacij tehnologij OpenMP + MPI ali celo OpenMP + MPI + Cuda. To predavanje predstavlja alternativni pristop k programiranju heterogenih sistemov v jeziku Chapel, prikazuje njegove osnovne pristope in predstavlja njihovo uporabo pri vzporedni implementaciji evolucijskih algoritmov.
Predavanje bo potekalo v angleškem jeziku.

Zahtevnost: napredna

Lokacija: Fakulteta za računalništvo in informatiko, Večna pot 113, Ljubljana

Organizator/ Organizer

Predavatelji