NN Basics: Deep CNNs for CIFAR10 with image augmentation скачать в хорошем качестве

NN Basics: Deep CNNs for CIFAR10 with image augmentation

CIFAR10 using deep convolution neural networks Review MNIST performance using CNNs Understand CIFAR10 dataset Apply the same method used with MNIST Increasing the depth of CNNs Improving the accuracy through data augmentation Error Analysis Note some changes need to happen: 1. kernel_initializer should be 'he_normal' instead of 'he_uniform' 2. lr_schedule is not effective until it is used callback of model.fit() 3. lr_scheduler = LearningRateScheduler lr_schedule 4. callbacks = [lr_scheduler] def lr_schedule(epoch): lr = 1e-3 if epoch gt 180: lr *= 0.5e-3 elif epoch gt 160: lr *= 1e-3 elif epoch gt 120: lr *= 1e-2 elif epoch gt 80: lr *= 1e-1 return lr model.fit(x_train, y_train, batch_size=batch_size, epochs=epochs, validation_data=(x_test, y_test), shuffle=True, callbacks=callbacks)