Broad intuition · Study map 13 of 85
Neurons, layers, and representations
Suggested preparation: Evaluation: the scoreboard shapes the system
Picture it
A layer mixes signals, then bends the result with a nonlinear function. Stacking layers lets simple transformations compose into complicated behavior. The weights are learned settings; activations are the changing signals for this input.
What to understand
- Weights, biases, activations, and nonlinearities
- Forward pass, loss, residual connections, and normalization
Follow the references
1 · Video series
Andrej Karpathy · Neural Networks: Zero to Hero ↗Start with micrograd. Draw a small computation graph before following the Python implementation.
You understand it when…
Why would stacking only linear layers still be equivalent to one linear transformation?
This is a study map. Follow the reference for the full explanation; use the question to test your mental model. Browse the reference shelf.