L04 · Neural Networks

L04 · Phase 3 · 25 min

Neural Networks

“How does the system turn these plain numerical representations into something useful?”

Neural Network Playground

One toy neuron takes two sensor readings from the condo system — temperature and flow rate, each scaled to 0–1 — and combines them into a single “pump alert” signal.

temperature (x₁) = 0.40flow rate (x₂) = 0.60

weight w₁ (temperature) = 2.20

weight w₂ (flow rate) = -1.40

bias b = -0.50

weighted sum = w₁x₁ + w₂x₂ + b = (2.20×0.40) + (-1.40×0.60) + -0.50 = -0.46
output = sigmoid(weighted sum) = 0.39

Prototype note: one neuron, two inputs — real networks stack thousands of these across many layers, each layer feeding the next. The maths shown (weighted sum, then an activation function) is exactly what scales up; nothing here is simplified beyond the network’s size.

Depth ladder

A neuron takes a handful of numbers in, multiplies each by an adjustable 'weight' that says how much it matters, adds them up along with a baseline adjustment, and produces one number out. Nudge the sliders below and watch two sensor readings turn into a single alert signal.

Knowledge check

Given a toy neuron with two weighted inputs and a bias, compute the output for sample values.