CNN model accuracy deteriorate when retrained for the second time?
Help · 19 Aug 2022, 07:19 · 1

When i train my neural network model for the first time using VGG16 I got accuracy of 0.8835 and loss 0.3268,But when I retrained again without Changing any data The accuracy decreased to 0.5787 and the loss increase to 0.6034.I wonder what maybe the problem????

Discussion 1 answer

i am pretty sure the train process of most deep learning models is random.

i suggest try to keep most random state parameter as constant as possible and you might wanna take a look at kernel_initializers of VGG16. if the kernel_initializer is not set it defaults to random which means that the first weight of the first layers are set to random values 😎😎😎😎