Module Check: Inside Neural Networks
A neuron has inputs [1, -2, 3] and weights [0.5, 0.5, 0.2]. What is the weighted sum?
Flashcards — click each card to reveal the answer
A neuron has inputs [1, -2, 3] and weights [0.5, 0.5, 0.2]. What is the weighted sum?
Which part of a trained neural network stores its learned knowledge?
A network has no activation functions — just weighted sums at every layer. What is its effective power?
During backpropagation, gradients travel in which direction?
A face recognition system trained mostly on faces photographed indoors performs poorly on outdoor photos. What failure mode is this?
What does a very large positive weight between two neurons mean?
A neural network is a layered composition of weighted sums and non-linear activations. Training finds weight values that minimize a loss function, using backpropagation to compute gradients and gradient descent to update weights. Depth allows hierarchical learning. Weights are the knowledge. Activation functions are the spark. The forward pass is the prediction. Backpropagation is how it improves. That is the whole story.
Teach It to Someone Else
- This is the ultimate test of understanding: explain it clearly to someone who has never heard of neural networks.
- Step 1: Find a partner — a family member, friend, or classmate who has not done this module.
- Step 2: Using only everyday language (no jargon without immediate explanation), explain these five ideas in order:
- a. What an artificial neuron does.
- b. What weights are and why they matter.
- c. Why networks are organized in layers.
- d. What the forward pass is.
- e. How the network improves by nudging weights.
- Step 3: Ask your partner to describe back to you, in their own words, how a network learns. Note what they got right and where they were confused.
- Step 4: Every gap in their understanding reveals a gap in your explanation. Revise and try again.
- Step 5: Write two or three sentences summarizing what you found hardest to explain clearly — that is the concept you should review one more time.