📈 Training Progress (Loss: ― | Accuracy: ―)

This interactive demonstration shows how a feedforward neural network learns to classify 2D data points through backpropagation. The network adjusts millions of parameters (weights and biases) to find the optimal decision boundary that separates different classes.

• Decision Boundary: The line or curve that the network learns to separate different classes
• Activation Functions: Non-linear functions (ReLU, Tanh, Sigmoid) that allow networks to learn complex patterns
• Learning Rate: Controls how much weights are adjusted during each training step
• Loss Function: Cross-entropy loss measures how wrong the network's predictions are
• Backpropagation: Algorithm that computes gradients and updates weights to minimize loss
• Overfitting: When the network memorizes training data instead of learning general patterns
• Regularization: Technique to prevent overfitting by penalizing large weights

• Start with simple datasets (Circle, Linear) before trying complex ones (Spiral, XOR)
• More hidden layers and neurons can learn complex patterns but may overfit
• Higher learning rates train faster but may overshoot the optimal solution
• Add regularization if the network is overfitting (high training accuracy, poor generalization)
• Try different activation functions - ReLU works well for most cases
• Click on the canvas to add your own training points and see how the network adapts!