Skip to content

Quiz 5: Neural Networks

CS50's Introduction to Artificial Intelligence with Python

Question 1

Network weights: w0 = -5, w1 = 2, w2 = -1, w3 = 3. Inputs: x1 = 3, x2 = 2, x3 = 4. What does the network compute using a step activation function? A ReLU activation function?

  • 0 for step, 0 for ReLU
  • 0 for step, 1 for ReLU
  • 1 for step, 0 for ReLU
  • 1 for step, 11 for ReLU ✓
  • 1 for step, 16 for ReLU
  • 11 for step, 11 for ReLU
  • 16 for step, 16 for ReLU

Why: Compute the weighted sum: −5 + (2×3) + (−1×2) + (3×4) = −5 + 6 − 2 + 12 = 11

  • Step: Output 1 if value ≥ 0 → 11 ≥ 0 → 1
  • ReLU: Output max(0, value) → max(0, 11) → 11

Question 2

How many total weights (including biases) will a fully connected neural network have with: input layer (3 units), hidden layer (5 units), output layer (4 units)?

Answer: 44 ✓

Why: Count connections + biases at each layer: - Input → Hidden: 3 × 5 = 15 weights + 5 biases = 20 - Hidden → Output: 5 × 4 = 20 weights + 4 biases = 24 - Total: 20 + 24 = 44

Question 3

A recurrent neural network listens to audio speech and classifies it by whose voice it is. What architecture fits best?

  • One-to-one (single input, single output)
  • Many-to-one (multiple inputs, single output) ✓
  • One-to-many (single input, multiple outputs)
  • Many-to-many (multiple inputs, multiple outputs)

Why: The input is a sequence of audio frames (many inputs), and the output is a single speaker classification (one output). Many-to-one captures sequential data and produces a single label.

Question 4

Apply a 2×2 max-pool to a 4×4 grayscale image. What is the result?

  • [[16, 14], [32, 30]]
  • [[22, 24], [32, 30]]
  • [[16, 12], [32, 28]] ✓
  • [[14, 12], [30, 28]]
  • [[16, 14], [22, 24]]
  • [[16, 12], [32, 30]]

Why: Max-pooling divides the image into non-overlapping 2×2 regions and takes the maximum value from each. For each quadrant, find the largest pixel value — those become the output values.