% IMPORTANT: The following is UTF-8 encoded. This means that in the presence
% of non-ASCII characters, it will not work with BibTeX 0.99 or older.
% Instead, you should use an up-to-date BibTeX implementation like “bibtex8” or
% “biber”.
@INPROCEEDINGS{Klijn:843867,
author = {Klijn, Wouter and Canova, Carlos and Baumeister, Paul F.
and Yegenoglu, Alper and Denker, Michael and Pleiter, Dirk
and Grün, Sonja},
title = {{ASSET} for {JULIA}: {E}xecuting {M}assively {P}arallel
{S}pike {C}orrelation {A}nalysis on {KNL} {C}luster},
reportid = {FZJ-2018-01398},
year = {2017},
abstract = {Introduction: We developed a statistical analysis method,
ASSET, capable of detectingrepeated sequences of synchronous
events (SSE) in massively parallel spike trains(Torre et
al., 2016). Yet we have not been able to apply ASSET in its
full extent, giventhe high computational demand when
assessing significance of the SSEs. This challenge,however,
can now be overcome with the support from the High
Performance Analyticsand Computing Platform (HPAC), and
their readily available modern infrastructure.Here we
present the first steps towards analyzing
electrophysiological recordings withASSET on one of the new
pre-commercial procurement machines, JULIA, which isbased on
Intel’s new Knights Landing (KNL) processor.Motivation:
ASSET is an analysis designed to detect and quantify
activity in a synfirechain (Abeles, 1991), a feedforward
neuronal network with high convergence and divergenceof
connectivity between the layers (groups). Particular to such
a network is thatit favors the propagation of synchronous
spiking activities, which appear in measurementsas SSEs. In
ASSET, the repetitive occurrence of an identical SSE becomes
visiblein an intersection matrix as a diagonal structure
(DS) (Schrader et al., 2008; Gersteinet al., 2012), which is
evaluated automatically for significance. Currently, the
ASSETmethod can only be applied to time segments that are
considerably shorter than the fullduration of a typical
session of massively parallel electrophysiological
recordings dueto costly numerical steps in the analysis.
However, these numerical computations arecomposed of
independent steps and thus ASSET would profit from
parallelization. Asecond challenge is the core of the
algorithm, which makes extensive use of exponentialand
logarithmic operations. These operations are computational
expensive and do notlend themselves to easy array
vectorization on modern HPC hardware.Methods: After analysis
and instrumentation of ASSET, an MPI version of the
softwarewas implemented, distributing the workload across
multiple compute instances in around-robin manner. After the
work on the nodes, the partial results are collected onthe
master node and summed for the final results. In a parallel
effort we optimized thecore of the ASSET algorithm: the
exponential and logarithmic operations are
typicallycalculated using Taylor expansions. Approximate
methods perform the same mathematicaloperations faster at
the expense of an error smaller than 1E-6. This speedup
canbe further improved on by (automatic) array vectorization
of the code implementingthese methods. These techniques were
combined with C implementations using theCython programming
interface.Results: The MPI implementation allowed us to
leverage the large number of coresavailable in current
hardware and showed an order of magnitude shorter time to
solution.We will further report on the preliminary
qualitative and quantitative analysis ofthe approximate
methods and its effects on the runtime of the algorithm,
including theresults of running the algorithm on the KNL
processors of JULIA. ASSET is currentlyavailable to the
scientific community via the Electrophysiological Analysis
Toolkit(Elephant)7, and as such is also available to all
members of the Human Brain ProjectConsortium via the
Collab.Acknowledgments: Supported by Helmholtz Portfolio
Theme Supercomputing andModeling for the Human Brain (SMHB),
EU grant 604102 (Human Brain Project,HBP), EU Grant 269912
(BrainScaleS), DFG Priority Program SPP 1665 (GR 1753/4-1and
2175/1-1).REFERENCESAbeles, M. (1991). Corticonics.
Cambridge: Cambridge University Press.Gerstein, G. L.,
Williams, E. R., Diesmann, M., Grün, S., and Trengove, C.
(2012). Detecting synfire chainsin parallel spike data. J.
Neurosci. Methods 206, 54–64. doi:
10.1016/j.jneumeth.2012.02.003 PMID:22361572Schrader, S.,
Bell, M. L., Allen, D. L., Byrnes, W. C., and Leinwand, L.
A. (2008). Skeletal muscle adaptations inresponse to
voluntary wheel running in myosin heavy chain null mice. J.
Neurophysiol. 100, 2165–2176. doi:10.1152/jn.01245.200
PMID:NOPMIDTorre, E., Canova, C., Denker, M., Gerstein, G.,
Helias, M., and Grün, S. (2016). ASSET: analysis of
sequencesof synchronous events in massively parallel spike
trains. PLoS Comput. Biol. 12:e1004939. doi:
10.1371/journal.pcbi.1004939 PMID:27420734},
month = {Feb},
date = {2017-02-08},
organization = {HBP student conference 2017, Vienna
(Austria), 8 Feb 2017 - 10 Feb 2017},
subtyp = {After Call},
cin = {JSC / INM-6},
cid = {I:(DE-Juel1)JSC-20090406 / I:(DE-Juel1)INM-6-20090406},
pnm = {511 - Computational Science and Mathematical Methods
(POF3-511) / SMHB - Supercomputing and Modelling for the
Human Brain (HGF-SMHB-2013-2017) / HBP SGA1 - Human Brain
Project Specific Grant Agreement 1 (720270)},
pid = {G:(DE-HGF)POF3-511 / G:(DE-Juel1)HGF-SMHB-2013-2017 /
G:(EU-Grant)720270},
typ = {PUB:(DE-HGF)24},
url = {https://juser.fz-juelich.de/record/843867},
}