A feedforward artificial neural network computes an output by passing an input vector through successive layers of neurons. Each neuron sums its weighted inputs, adds a bias and applies a non-linear activation function.
z = Σ (wᵢ · xᵢ) + b
y = f(z) f = ReLU, Sigmoid or Tanh
- Hidden layers — how many layers of neurons sit between the input and output layer; more layers let the network represent more complex functions.
- Neurons per hidden layer — the width of each hidden layer; controls how many features the network can learn at once.
- Signal speed — how fast the animated pulses travel from the input layer to the output layer, visualizing one forward pass.
- Activation function — the non-linearity
f(z) applied at every neuron; toggling it changes how strongly each neuron "fires", shown as glow brightness.
This weighted-sum-then-activate pattern, repeated across many more layers and neurons, is the core computation behind modern AI systems — from small classifiers to the large networks behind computer vision and language models.