Multi-Label Classification of Handwritten Digits Using CNNs | COMP 551 Course Project

This project involved building a Convolutional Neural Network (CNN) to classify multi-digit handwritten images from the Modified MNIST dataset. The process included:

• Digit Localization: Developed an algorithm to crop and isolate digits from the center of each image.
• CNN Architecture: Implemented a 4-layer convolutional network with pooling and dropout layers, achieving 97.7% accuracy on the Kaggle leaderboard.
• Optimization: Enhanced model performance through adaptive learning rates, batch size experimentation, and hyperparameter tuning.