This is a demo of a basic convolutional neural network on the CIFAR-10 dataset. The model based on VGGNet consists of 6 convolution layers with leaky ReLU activation units, 3 max-pooling layers with dropout, and 2 fully-connected dense layers, with final softmax for classification into 10 classes. Offline test accuracy of this simple model was around 85%, trained without any image augmentation. The architecture and weights of the model were serialized from a trained Keras model into a JSON file, which is then used to run the neural network in your browser, on-the-fly with the loaded sample images (prediction isn't called until sample images are loaded). Five sample images are randomly loaded per button-click below. This demo utilizes Web Workers to run the neural networks.