What is the vanishing gradient problem?
Whether you can trace it to the chain rule and name the fixes that actually addressed it.
Machine learningStandard
What a strong answer does
The vanishing gradient problem is a backpropagation problem, not a mysterious property of depth by itself. During backprop, the gradient for an early layer is computed by repeatedly applying the chain rule through every later layer. That means multiplying many local derivatives and weight terms together. If the typical factor in that product has magnitude below one, the product decays geometrically as the network gets deeper. By the time the signal reaches the first layers, it can be so small that those layers receive almost no useful learning signal.