Skip to main content
Tokenline

Line 1 · Station 4 of 6

Neural Networks, Conceptually

Build intuition for how artificial neurons weigh inputs, stack into layers, and learn by adjusting their weights.

  • 8 min
  • +50 XP
  • 2 knowledge checks

By the end of this station, you can:

  • Describe an artificial neuron as a weighted sum compared to a threshold
  • Explain how layers let a network find more complex patterns
  • Describe learning as adjusting weights to reduce errors

The idea behind deep learning

Neural networks are the engine inside deep learning — the kind of machine learning behind face recognition, speech assistants and chatbots.

The name comes from a loose inspiration: brain cells called neurons pass signals to each other. But an artificial neural network is really just math — lots and lots of multiplying and adding numbers.

One artificial neuron

An artificial neuron takes a few inputs, multiplies each by a weight that says how important it is, and adds them up. If the total passes a threshold, the neuron “fires” (outputs a strong signal).

Say you're deciding whether to go to Friday's game. Inputs: Is it nice out? Are your friends going? Do you have a test Monday? Your friends going might carry a big weight, while the test might carry a negative one.

Interactive lab

Build a Neuron

Flip the inputs and drag the weight sliders. The neuron multiplies each input by its weight, adds them up, and says “Go” if the total reaches the threshold.

Homework done?

Friend going?

Is it raining?

Weighted sum

(1 × 3) + (1 × 2) + (0 × (−4)) = 5

5 ≥ 4

Fires → Go to the game!

Training challenge

Adjust the weights and threshold so the neuron says “Go” only when homework is done and it's not raining — whether or not your friend goes.

7 of 8 situations correct — keep adjusting.

What to notice: A neuron is just arithmetic: weights decide how much each input matters. When a network “learns,” it nudges thousands or millions of weights like these automatically, based on its mistakes.

Key terms

Neural network
A machine learning model made of many connected artificial neurons arranged in layers.
Artificial neuron
A small unit that multiplies its inputs by weights, adds them up, and produces an output based on that total.
Weight
A number that controls how much a particular input influences a neuron's output. Learning means adjusting weights.
Layer
A group of neurons at the same stage of a network. Outputs from one layer become inputs to the next.

Knowledge check 1

+10 XP on first try
In a neuron deciding whether to go to the game, the input “I have a test Monday” has a large negative weight. What does that mean?
A small neural networkThree input neurons connect to four hidden-layer neurons, which connect to two output neurons. Every connecting line is a weight. Thicker lines are stronger weights; blue lines are positive and orange lines are negative. Training adjusts these weights.InputsHidden layerOutputpositive weightnegative weightEach line = a weight that gets adjusted during training. Thicker = stronger influence.
DiagramInputs flow left to right through layers of neurons. Each connection has its own weight. Early layers catch simple patterns; later layers combine them into more complex ones.

How a network learns

  1. Guess

    Data flows through the network and it produces an output — at first, a pretty random one.

  2. Measure the error

    The output is compared with the correct label to see how wrong it was.

  3. Adjust the weights

    Every weight gets nudged slightly in the direction that would have made the error smaller.

  4. Repeat

    Over thousands or millions of examples, those small nudges add up to a network that makes useful predictions.

Myth vs. fact

  • MythA neural network is a digital copy of a human brain.

    FactIt was loosely inspired by neurons, but it works very differently. It's a math model that adjusts numbers — it has no awareness, feelings or understanding.

  • MythProgrammers set every weight by hand.

    FactLarge networks can have millions or billions of weights. They are set automatically during training, which is also why it's hard to explain exactly why a network made a certain choice.

Knowledge check 2

+10 XP on first try
A network keeps labeling photos of wolves as “husky.” During training, what actually changes to fix this?

Key takeaways

  • An artificial neuron = inputs × weights, added up, compared to a threshold.
  • Stacking neurons in layers lets networks find complex patterns.
  • Learning means automatically adjusting weights to reduce errors.
  • Neural networks are inspired by brains but are not brains.

Sources for this lesson

Finish this station

Answer every knowledge check (right or wrong) to unlock completion.

  • Knowledge check 1 — not answered yet
  • Knowledge check 2 — not answered yet

0/2 checks answered