AI Layers and Activation Functions
A neural network gets its power not from a single neuron but from stacking many neurons into layers, each layer's output reshaped by an activation function before it moves on.
From One Neuron to a Network
A single artificial neuron (or perceptron) takes some inputs, multiplies each by a weight, adds them up along with a bias, and produces one number. On its own, that is not very useful - it can only draw a straight dividing line between two outcomes. The real power of a neural network comes from arranging many of these neurons into layers and connecting layer to layer, so that the output of one layer becomes the input to the next.
Input, Hidden, and Output Layers
Every network has an input layer, which simply holds the raw data (pixel brightness values, word counts, sensor readings) without doing any calculation. Between the input and the final answer sit one or more hidden layers, where the actual pattern-detection happens - each hidden neuron combines signals from the previous layer in a slightly different way. Finally, the output layer produces the network's answer, shaped to match the task: a single number for a price prediction, a yes/no probability for a binary decision, or a list of probabilities for a multi-class decision.
- Input layer - holds the raw features, performs no computation
- Hidden layer(s) - combine and transform signals from the previous layer to detect patterns
- Output layer - produces the final prediction, shaped to fit the task
Why a Network Needs Activation Functions
If every neuron only computed a weighted sum, stacking ten layers would mathematically collapse into the same thing as one layer - weighted sums of weighted sums are still just a weighted sum. An activation function is a small nonlinear function applied to each neuron's output before passing it forward. That nonlinearity is what lets a deep stack of layers represent curves, thresholds, and complex decision boundaries instead of only straight lines.
Example
import numpy as np
def relu(x):
return np.maximum(0, x)
def sigmoid(x):
return 1 / (1 + np.exp(-x))
raw_scores = np.array([-2.0, -0.5, 0.0, 1.5, 3.0])
print("ReLU output:", relu(raw_scores))
print("Sigmoid output:", sigmoid(raw_scores))Layers also vary in width - the number of neurons inside a single layer. A wider layer can hold more distinct patterns at once, while a deeper stack of narrower layers can build up more complex ideas step by step by combining simpler ones from the layer before it. Choosing how wide and how deep to make a network is part of designing its architecture.
