This is second project on DLFND class. In this project, I classify images from the CIFAR-10 dataset, and I access the floydhub.com for using GPU on this project.
- I can get 64.4% accuracy, I know that it is not a high result. Beacause, I use a simple CNN, and over 50% accuracy is not bad. I'll study few more skills about it, then I can get better accuracy than now.
- The number of epochs, which I set 100, appears excessive. The validation accuracy does not improve beyond 20 epochs, also the likelihood of overfitting is increased. I should tune hyperparameter values variety ways, then the accuracy of validation will be close to 80%.