Used training data obtained from Toronto Faces Dataset to train a multi-layer Neural Network for various facial expression detection via ReLu and backpropagation with weight updates and momentum
Implemented ReLU, softmax, and weight update using gradient descent and momentum with velocity during forward and back propagation.
Determined optimal hyper-parameters of learning rate, number of epochs,and batch-size, and momentum.
Implemented both KNN and a penalized logistic regression model for comparison.
Included all graphs for both training and validation sets.
Used training data obtained from Toronto Faces Dataset to train a multi-layer Neural Network for various facial expression detection via ReLu and backpropagation with weight updates and momentum