Overview

• This project investigates two classification problems: Digit Recognition and Face Recognition

Digit Recognition

Data

• Work with the MNIST dataset.
• The digits already seprated into a training and a test set.

Network

Cost Function

• The sum of the negative log-probabilities of all the training cases

Face Recognition

Developed a single-hidden-layer neural network model with a custom loss function using PyTorch to classify faces extracted from 100,000 images
Data

• A subset of the FaceScrub dataset.
• The dataset consists of URLs of images with faces, as well as the bounding boxes of the faces.

Network

• a single-hidden-layer fully-connected network

Techniques

learning rate, batch normalization, dropout, and various optimizers，regularization