HandwrittenLetters

CS 6375.002 Final Project

Project Title: Recognizing Handwritten Letters using Neural Network Algorithms

Team Members:

• Michelle Kelman
• Jihan Wang

Dataset Type:

• Kaggle - https://www.kaggle.com/datasets/crawford/emnist

Algorithm Approach:

• Own Implementation
• Neural Network models with different activation functions
• Simple PCA and t-SNE methods with 2D and 3D plot functions

How To Use:

1. Clone our project
2. Download the dataset from the Kaggle link above
3. Copy the extracted data files into the project data folder
4. Process the data: python data.py
5. Run the different models:

Activation Function	Logistic (Sigmoid)	ReLU	Hyperbolic Tangent
1 Hidden Layer Neural Network	python logistic-1.py	python relu-1.py	python tanh-1.py
scikit Neural Network MLPClassifier	python logistic-s.py	python relu-s.py	python tanh-s.py
Keras Convolutional Neural Network	python logistic-k.py	python relu-k.py	python tanh-k.py

Additional hidden layer models:

Activation Function	Logistic (Sigmoid)	ReLU	Hyperbolic Tangent
2 Hidden Layer Neural Network	python extra/logistic-2.py	python extra/relu-2.py	python extra/tanh-2.py
3 Hidden Layer Neural Network	python extra/logistic-3.py	python extra/relu-3.py	python extra/tanh-3.py

6. Get the data in the hidden layer and output layer: run forward() function
7. Run the PCA or t-SNE models