Quadrat1c / Neural-Networks

Exploring Neural Networks in C# .NET

Neural-Networks

Exploring Neural Networks in C# .NET

Activation Functions

• Logistic Sigmoid
• Hyperbolic Tangent
• Heaviside Step
• Softmax

Training Methods

• Back-propagation
• Genetic Algorithm
• Particle Swarm Optimization

4 Free Parameters

• Number of Hidden Neurons
• Momentum
• Learning Rate
• Weight Decay

Neural Network Basics

Core Concepts

• Feed-forward
• Activation Functions
• Data Encoding
• Error
• Training
• Free Parameters
• Over-fitting

Neurons

• Input Neurons (Determined by the amount of inputs)
• Hidden Neurons (The unsolved mystery of neural networks, try different amounts until wanted result)
• Output Neurons (How many choices could the final answer be? Yes or No [2 Output Neurons] Yes, No, Maybe? [3 Output Neurons])
• Bias (Normally is a dummy input usually set at '1.0')

Accuracy

It is easy to "over-fit" a neural network. You do not wan't 100& during training. However you do wan't 100% Accuracy on the Test Set.