how to choose specific GPUs on multi-GPUs server to train my CNN？

Our lab has a server with 8 GPUs(0,1,2,...,7),I want to choose No.1 and No.3 GPUs to train my CNN,how to do? The code is shown below.thanks

clear

trainingData = imageDatastore('C:\Users\Administrator\Desktop\matlab code\Training',... 'IncludeSubfolders',true,'LabelSource','foldernames');

testData = imageDatastore('C:\Users\Administrator\Desktop\matlab code\Test',... 'IncludeSubfolders',true,'LabelSource','foldernames');

layers = [ ... imageInputLayer([227 227 3]) convolution2dLayer(5,8) reluLayer() maxPooling2dLayer(2,'Stride',2)

            convolution2dLayer(5,48)
            reluLayer()
            maxPooling2dLayer(2,'Stride',2)

            convolution2dLayer(5,48)
            reluLayer()
            dropoutLayer('Name','drop1')

            fullyConnectedLayer(2)  
            softmaxLayer()
            classificationLayer()];
            options = trainingOptions('sgdm',...
            'InitialLearnRate',0.0001,...
            'MiniBatchSize',16,...
            'MaxEpoch',1,...
            'ExecutionEnvironment','multi-gpu')

        net = trainNetwork(trainingData,layers,options);
        YTest = classify(net,testData);
        TTest = testData.Labels;
        accuracy = sum(YTest == TTest)/numel(TTest)