Neural Networks Evaluation

This library and set of test programs aim to evaluate the learning and usage of multi-layer perceptron (MLP) networks. The Neural library implements different differentiable and nondifferentiable activation functions and has support for different learning strategies.

Demo programs

• XOR problem
• Handwritten digit classification

Transfer/activation functions

• Linear
• Sigmoid
• Tanh
• Softplus
• Rectifier
• Heaviside (Step)

Learning algorithms

• Momentum-based gradient descent with regularization

Cost functions

• Logistic
• Sum of squared errors