Contribution to a conference proceedings

FZJ-2021-01284

http://join2-wiki.gsi.de/foswiki/pub/Main/Artwork/join2_logo100x88.png Scaling Up a Multispectral Resnet-50 to 128 GPUs

Sedona, R. (Corresponding author)FZJ* ; Cavallaro, G.FZJ* ; Jitsev, J.FZJ* ; Strube, A.FZJ* ; Riedel, M.FZJ* ; Book, M.

2020
IEEE

This record in other databases:  

Please use a persistent id in citations: http://hdl.handle.net/2128/27298  doi:10.1109/IGARSS39084.2020.9324237

Abstract: Similarly to other scientific domains, Deep Learning (DL)holds great promises to fulfil the challenging needs of RemoteSensing (RS) applications. However, the increase in volume,variety and complexity of acquisitions that are carried outon a daily basis by Earth Observation (EO) missions generatesnew processing and storage challenges within operationalprocessing pipelines. The aim of this work is to show thatHigh-Performance Computing (HPC) systems can speed upthe training time of Convolutional Neural Networks (CNNs).Particular attention is put on the monitoring of the classificationaccuracy that usually degrades when using large batchsizes. The experimental results of this work show that thetraining of the model scales up to a batch size of 8,000, obtainingclassification performances in terms of accuracy in linewith those using smaller batch sizes.

---

Contributing Institute(s):

1. Jülich Supercomputing Center (JSC)

Research Program(s):

1. 512 - Data-Intensive Science and Federated Computing (POF3-512) (POF3-512)
2. PhD no Grant - Doktorand ohne besondere Förderung (PHD-NO-GRANT-20170405) (PHD-NO-GRANT-20170405)

Appears in the scientific report 2021

---

Database coverage:
; ; BIOSIS Previews ; Current Contents - Life Sciences ; Ebsco Academic Search ; IF < 5 ; JCR ; NCBI Molecular Biology Database ; SCOPUS ; Science Citation Index Expanded ; Thomson Reuters Master Journal List ; Web of Science Core Collection ; Zoological Record

---