RAPIDS: GPU-powered data science

Conference presentation

• Author: Marko Kvakić https://orcid.org/0000-0003-0445-677X
• Author's institution: University of Zagreb; (SRCE, University Computing Centre)
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract (english)

Data science and machine learning is one of the world’s largest compute segment, in which small improvements in accuracy can translate into huge returns. To build the best models, data scientists work to train, evaluate, iterate, and retrain for highly accurate results and performant models. Want to learn how to engineer features and fit trees at lightning speed? This workshop focuses on performing data processing and machine learning using RAPIDS: a GPU-powered framework that enables massively parallel computations in typical data science workflows. Based on the familiar scikit-learn & pandas API, RAPIDS allows for an easy transition to GPU-powered systems and provides significant speed-ups in data processing and model development. With RAPIDS, processes that took days take minutes, making it easier and faster to build and deploy value-generating models. The workshop includes a hands-on demonstration of RAPIDS capabilities applied to a publicly available dataset (NYC Taxi Fares) and be performed on the University Computing Centre’s (SRCE) new cloud service “Vrančić”, with a GPU server housing four A100 GPUs.

• Keywords (english): AI; data science; GPU acceleration; machine learning; Rapids
• Language: english
• Peer review: No peer review
• URN:NBN: urn:nbn:hr:102:013868
• Presentation date: 2023-10-19
• Conference: Title: AI2Future; Location: Zagreb, Croatia; Organizer: CroAI; Start date: 2023-10-19; End date: 2023-10-20; Assembly type: Workshop
• Type of resource: mixedmaterial
• Access conditions: Open access
• Created on: 2023-10-24 11:20:32