Digit Recognizer
Made by Cole Corbett & Chance Page
Trained on 50,000 MNIST images for 10 epochs with standard preprocessing. This model represents a well-balanced CNN with proper regularization (dropout) and sufficient training data. It should generalize well to new handwritten digits.
Trained with data augmentation techniques including random rotations (±10°), width/height shifts (±10%), and zoom variations (±10%). This creates more diverse training examples from the same dataset, helping the model handle variations in handwriting styles, angles, and positioning.
Intentionally trained on only 1,000 images for 50 epochs to demonstrate overfitting. This model has memorized the limited training data rather than learning general patterns. It shows high training accuracy but poor generalization—watch how it may struggle with your handwriting compared to the other models.