(PP19): Heterogeneous Computing for Deep Learning
Event Type
Project Poster
AI/Machine Learning/Deep Learning
Big Data Analytics
HPC workflows
Heterogeneous Systems
Performance Analysis and Optimization
TimeWednesday, June 19th3:15pm - 4pm CEST
LocationBooth N-230
DescriptionMission-critical applications involving stringent constraints (latency, throughput, power, size, etc.) present computing challenges to achieve tradeoffs among these constraints, often resulting in inefficient or sub-optimal performance. Heterogeneous computing (CPU+GPU+FPGA) offers an opportunity to address this challenge and balance the tradeoffs for achieving application objectives. Our research team at the University of Florida is studying the application of heterogeneous computing to high-energy physics (HEP) with convolutional neural networks (CNNs). This poster presents early findings in adapting state-of-the-art CNN models for HEP data analysis on a heterogeneous platform, achieving an average of 2X speedup for inferencing with naive optimization. Our team won the first-ever DELL EMC AI Challenge.
Poster PDF
Poster Authors
UF Site Director, NSF Center for Space, High-Performance, and Resilient Computing
Distinguished Engineer
Research Associate, NSF Center for Space, High-Performance, and Resilient Computing