HOW TO INCREASE TESTING ACCURACY IN CNN?

Question

Saira 2020-6-29

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/556924-how-to-increase-testing-accuracy-in-cnn

回答： o.cefet cefet 2020-9-1

I have 5600 training images. I have extracted features using Principal Component Analysis (PCA). Now I am applying CNN on feature vector. Testing accuracy is 18%. How to increase testing accuracy?

Feature vector size: 640*1

Training code:

layers = [

imageInputLayer([1 640 1]);

reluLayer

fullyConnectedLayer(7);

softmaxLayer();

classificationLayer()];

options = trainingOptions('sgdm', 'Momentum',0.95, 'InitialLearnRate',0.0001, 'L2Regularization', 1e-4, 'MaxEpochs',20);

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

vaibhav mishra 2020-6-30

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/556924-how-to-increase-testing-accuracy-in-cnn#answer_459046

Hi,

there can be different ways to increase the test accuracy.

If your training accuracy is good but test accuracy is low then you need to introduce regularization in your loss function, or you need to increase your training set.

if your training accuracy increased and then decreased and then your test accuracy is low, you are over training your model so try to reduce the epochs.

if your both training and testing accuracy are less then try to either change your model architecture, or increase the training data or decrease learning rate or increase the number of epochs.

Feel free to ask for any clarification.

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Saira 2020-7-2

Thanks Vaibhav

请先登录，再进行评论。

Answer 2

Saira 2020-7-9

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/556924-how-to-increase-testing-accuracy-in-cnn#answer_463163

Hi,

I changed the training parameters but the testing accuracy is still not increasing. What should I do to increase accuracy?

Code:

options = trainingOptions('adam', 'InitialLearnRate',3e-4, 'SquaredGradientDecayFactor',0.99, 'L2Regularization',0.521, ... 'MaxEpochs',500, 'MiniBatchSize',64, 'Shuffle','every-epoch', 'Plots','training-progress');