Syllabus Map

• Study map: Syllabus Study Map

Core Idea

• Derivative of composed functions multiplies local derivatives.

Key Formula

Practical Notes

Cache intermediate activations in deep models.

• Backprop uses repeated chain-rule application through layers.

Local gradients can vanish or explode.

• Product of many derivatives can become unstable.

Why This Matters for ML

• Backpropagation is repeated chain-rule application through layered computations.
• Gradient quality in deep models depends on products of local derivatives.
• Vanishing/exploding gradients are chain-rule phenomena across many layers/time steps.
• Correct chain-rule intuition is essential for designing activations and stable architectures.