DNN#Assisted*Parameter*Space* Exploration*and*Visualization*for* Large*Scale*Simulations HA HAN#WE WEI*SH SHEN DE DEPARTM TMENT*OF NT*OF*C *COM OMPUTE TER*S *SCIENC NCE*A *AND*E ND*ENG NGINE NEERING NG TH THE*OH OHIO* O*STATE TE*UNIVERSITY TY 1
Traditional**Visualization*Pipeline* Disk I/O I/O Simulator Raw*data Memory Supercomputer Post9analysis • Data*are*often*very*large* • Mainly*batch*mode*processing;**Interactive*exploration*is*not*possible • Limited*parameters*to*explore
Parameter'Space'Exploration • Running&large&scale&simulations&is&very&time&and&storage& consuming& • A&physical&simulation&typically&have&a&huge¶meter&space • Ensemble&simulations&are&needed&for&analyzing&the&model&quality& and&also&identify&the&uncertainty&of&the&simulation& • It&is¬&possible&to&exhaust&all&possible&input¶meters&– time& and&space&prohibitive& 3
DNN#Assisted#Parameter#Space#Exploration • Can$we$create$visualization$of$the$simulation$outputs$without$saving$the$data? Or • Can$we$predict$the$simulation$results$without$running$the$simulation? • Why?$ : Identify$important$simulation$parameters$ : Identify$simulation$parameter$sensitivity$ : Quantify$the$uncertainty$of$the$simulation$models • Methods: : InSituNet (IEEE$SciVis’19$Best$Paper) : NNVA (IEEE$VAST’19$Best$Paper$Honorable$Mention) 4
InSituNet:*Deep*Image*Synthesis*for* Parameter*Space*Exploration*of*Ensemble* Simulations 5
Introduction • Ensemble(data(analysis(workflow • Issues 5 I/O(bottleneck(and(storage(overhead simulation parameters I/O I/O raw data ensemble post-hoc disk simulations analysis 6
Introduction • In#situ#visualization - Generating#visualizations#at#simulation#time - Storing#images#for#post-hoc#analysis visual mapping parameters view parameters simulation visualization parameters I/O I/O images in situ vis image data ensemble post-hoc disk simulations analysis 7
8 Introduction • Challenges ) Limiting.the.flexibility.of.post)hoc.exploration.and.analysis ) Raw.data.are.no.longer.available ) Incapable.of.exploring.the.simulation.parameters ) Expensive.simulations.need.to.be.conducted.for.new.parameter.settings • Our.Approach ) Studying.how.the.parameters.influence.the.visualization.results ) Predicting.visualization.results.for.new.parameter.settings
Approach(Overview visual mapping parameters view ! "#$ parameters simulation visualization ( ! %#" ℱ ! "#$ , ! %#" , ! %#&' = ( images parameters in situ vis ! %#&' ensemble simulations InSituNet A#deep#neural#network#that#models#the#function# ℱ 9
Design'of'InSituNet • Three'subnetworks'and'two'losses 3 Regressor (mapping'parameters'to'prediction'images) 3 Feature+comparator+ (computing' feature+reconstruction+loss ) 3 Discriminator (computing' adversarial+loss ) input feature Feature parameters reconstruction Comparator loss ground truth Regressor Discriminator adversarial loss prediction 10
Design'of'InSituNet • Three&subnetworks&and&two&losses 2 Regressor (mapping(parameters(to(prediction(images) 2 Feature'comparator' (computing& feature'reconstruction'loss ) 2 Discriminator (computing& adversarial'loss ) input feature Feature parameters reconstruction Comparator loss ground truth Regressor Discriminator adversarial loss prediction 11
12 • A"convolutional"neural"network"(CNN) Regressor' + , 4 Weights:" * ,"updated"during"training 4 Output:"prediction"image" ( 4 Input:"parameters" ! "#$ ," ! %#" ,"and" ! %#&' parameters input Regressor ) prediction view visual mapping simulation parameters parameters parameters (n) (m) (l) (n, 512) (m, 512) (l, 512) (512) (512) 2D convolution fully connected residual block input/output (512) (512, 512) (512, 512) (512, 512) concat Architecture"of"the"regressor (1536) (1536, 4 × 4 × 16 × k ) (4 × 4 × 16 × k ) others normalization bacth ReLU reshape (4, 4, 16 × k ) (in=16 × k , out=16 × k ) (8, 8, 16 × k ) (in=16 × k , out=8 × k ) (16, 16, 8 × k ) (in=8 × k , out=8 × k ) upsampling (32, 32, 8 × k ) (in=8 × k , out=4 × k ) upsampling (64, 64, 4 × k ) (in, out, 1, 1) (in=4 × k , out=2 × k ) (128, 128, 2 × k ) (in, out, 3, 3) (in=2 × k , out= k ) (256, 256, k ) (out, out, 3, 3) ( k , 3, 3, 3) (256, 256, 3) sum tanh a image
view visual mapping simulation parameters parameters parameters (m) (l) (n) Regressor' ! " (m, 512) (l, 512) (n, 512) (512) (512) (512) 2D convolution fully connected residual block input/output (512, 512) (512, 512) (512, 512) concat (1536) (1536, 4 × 4 × 16 × k ) (4 × 4 × 16 × k ) others normalization bacth ReLU reshape (4, 4, 16 × k ) (in=16 × k , out=16 × k ) (8, 8, 16 × k ) (in=16 × k , out=8 × k ) (16, 16, 8 × k ) (in=8 × k , out=8 × k ) (32, 32, 8 × k ) upsampling (in=8 × k , out=4 × k ) upsampling (64, 64, 4 × k ) (in, out, 1, 1) (in=4 × k , out=2 × k ) (in, out, 3, 3) (128, 128, 2 × k ) (in=2 × k , out= k ) (256, 256, k ) (out, out, 3, 3) ( k , 3, 3, 3) (256, 256, 3) sum tanh a image • A$convolutional$neural$ network$(CNN) • Weights$ # :$weights$ • Output:$prediction$ • Input:$simulation,$visual$ collected$from$all$layers image parameters mapping,$view$ 13
view visual mapping simulation parameters parameters parameters (m) (l) (n) Regressor' ! " (m, 512) (l, 512) (n, 512) (512) (512) (512) 2D convolution fully connected residual block input/output (512, 512) (512, 512) (512, 512) concat (1536) (1536, 4 × 4 × 16 × k ) (4 × 4 × 16 × k ) others normalization bacth ReLU reshape (4, 4, 16 × k ) (in=16 × k , out=16 × k ) (8, 8, 16 × k ) (in=16 × k , out=8 × k ) (16, 16, 8 × k ) (in=8 × k , out=8 × k ) (32, 32, 8 × k ) upsampling (in=8 × k , out=4 × k ) upsampling (64, 64, 4 × k ) (in, out, 1, 1) (in=4 × k , out=2 × k ) (in, out, 3, 3) (128, 128, 2 × k ) (in=2 × k , out= k ) (256, 256, k ) (out, out, 3, 3) ( k , 3, 3, 3) (256, 256, 3) sum tanh a image • Activation'function • Fully'connected'layer • Rectified'Linear'Units'(ReLU) • Weights:'matrix' ) ∈ ℝ &×( • Output:'1D'vector' ' ∈ ℝ ( • Input:'1D'vector' # ∈ ℝ & '′ = max(0, ') ' = )# 14
view visual mapping simulation parameters parameters parameters (m) (l) (n) Regressor' ! " (m, 512) (l, 512) (n, 512) (512) (512) (512) 2D convolution fully connected residual block input/output Conference%on%Computer%Vision%and%Pattern%Recognition,%pp.%770–778,%2016. 1 K.%He,%X.%Zhang,%S.%Ren,%and%J.%Sun.%Deep%residual%learning%for%image%recognition.%In%Proceedings%of%2016%IEEE% (512, 512) (512, 512) (512, 512) concat (1536) (1536, 4 × 4 × 16 × k ) (4 × 4 × 16 × k ) others normalization bacth ReLU reshape (4, 4, 16 × k ) (in=16 × k , out=16 × k ) (8, 8, 16 × k ) (in=16 × k , out=8 × k ) (16, 16, 8 × k ) (in=8 × k , out=8 × k ) (32, 32, 8 × k ) upsampling (in=8 × k , out=4 × k ) upsampling (64, 64, 4 × k ) (in, out, 1, 1) (in=4 × k , out=2 × k ) (in, out, 3, 3) (128, 128, 2 × k ) (in=2 × k , out= k ) (256, 256, k ) (out, out, 3, 3) ( k , 3, 3, 3) (256, 256, 3) sum tanh a image • Residual%block 1 • 2D%Convolutional%Layer • Adding%input%to%the%output% • Weights:%kernel% , ∈ • Output:%tensor% * ∈ ℝ &×(×) + • Input:%tensor% # ∈ ℝ &×(×) of%convolutional%layers ℝ -_(×-_&×)×) + * = , ∗ # 15
16 Loss$Function • Difference*between*the*prediction*and*the*ground*truth • Used*to*update*the*weights* ! Backpropagation input Regressor parameters prediction Loss ground truth
17 Loss$Function$+ Straightforward$Approach • Pixel&wise&loss&functions / Example:&mean&squared&error&loss&(MSE&loss& ℒ "#$ ) / Issue:&blurry&prediction&images input Regressor parameters prediction MSE&Loss Blurry&image&generated&by&a regressor&trained&with&the&MSE&loss ground truth
18 Loss$Function$+ Our$Approach • Combining(two(loss(functions(defined(by(two(subnetworks 5 Feature(comparator(5>(feature(reconstruction(loss( ℒ "#$% 5 Discriminator(5>(adversarial(loss( ℒ $&' feature input Feature Regressor reconstruction parameters Comparator loss prediction Discriminator adversarial loss ground truth
Design'of'InSituNet • Three'subnetworks'and'two'losses 3 Regressor (mapping'parameters'to'prediction'images) 3 Feature'comparator' (computing+ feature'reconstruction'loss ) 3 Discriminator (computing' adversarial1loss ) input feature Feature parameters reconstruction Comparator loss ground truth Regressor Discriminator adversarial loss prediction 19
Recommend
More recommend
