ICP11 - narhirep/Python-Deep-Learning GitHub Wiki

Welcome to the In Class Programming 11:

Objective: Training the Convolutional Neural Network to classify the CIFAR-10 dataset.

Implementation:

I have created a model that includes all the required layers as per question 1. The model was trained for 25 epochs. The model gave accuracy of 87 percent during training and 79 percent during testing. Since there are fewer CNN layers for extracting the features in the dataset and it is overfitting, the Usecase model performs poorly in the test process as compared to our model.

After training, saved the model to cifar10-model2 using save() method of the model.

Plotted the first 4 test images along with their label and the model predicted label. The model was able to predict all the 4 test images correctly. While in the case of Usecase, the 3rd image is incorrectly classified as Airplane when the image actually is Ship.

Plotted the graph between accuracy and Number of Epochs, Loss and Number of Epochs of the training and validation.

Video:

Conclusion: I learned about the Convolutional Neural Network and the various layers that comprise the CNN. Also, how to create, train, and save a trained model that can be loaded at a later time without having to retrain the model.