From Perceptron to Multilayer Network · seed 1 · A4, ink-friendly. The answer key prints on its own page for grown-ups.

One line is not enough

Computing · Machine Learning · ages 21-22
Name ______________________   Date ____________

  1. Why can one neuron not learn exclusive-or?

    • Yes corners move during the test, defeating the line
    • No single line groups the opposite yes corners
    • Neurons dislike corners

  2. What does one neuron compute?

    • A straight-line split with a firing mark
    • A whole deep network of many layers at once
    • A random answer each time

  3. A stack of layers with no nonlinear step behaves like one layer.

    Circle one:   True   False

  4. A neuron fires when its score passes a mark. What is that mark called?

    • An activation threshold
    • A step size for the search
    • A hidden layer of its own

  5. In a small image network, what does the first layer spot?

    • Final answers with their names attached
    • Simple edges the next layer combines
    • The training schedule

  6. Where do the network's own new features get built?

    • In the input signals, before any weights
    • In the output layer only
    • In the hidden layers between input and output

  7. Two linear layers with no step sit in a row. What can replace them?

    • A deeper stack with even more lines
    • A bigger screen
    • A single layer with matched weights

  8. Which two-layer plan solves exclusive-or by hand?

    • One neuron with a thicker line
    • One hidden unit per yes corner, then OR them
    • Two neurons side by side with no output layer
LightMySky · lightmysky.comW1-mt_TTa7KN1-IO-s1

Answer key

For grown-ups. Fold this page away before handing over the rest.

One line is not enough W1-mt_TTa7KN1-IO-s1

  1. No single line groups the opposite yes corners · Opposite yes corners need more than one straight fence.
  2. A straight-line split with a firing mark · Signals combine into a score, and the threshold draws one straight line.
  3. True · Without the nonlinear step the stack collapses into a single layer.
  4. An activation threshold · The activation threshold is the firing mark the score must pass.
  5. Simple edges the next layer combines · Early layers find edges, and deeper layers combine them into shapes.
  6. In the hidden layers between input and output · Middle layers invent the features that raw inputs lack.
  7. A single layer with matched weights · With no nonlinear step, each layer only scales and adds its inputs, and doing that twice in a row is still one round of scaling and adding. Combine the two sets of weights into one and a single layer gives the same outputs.
  8. One hidden unit per yes corner, then OR them · Each hidden unit guards one yes corner, and the output fires for either.
Worksheet · LightMySky