A gradient is a vector of partial derivatives describing how a scalar function changes across multiple variables. In a deep network, gradients support learning; a chain-rule expression relates loss derivatives across network layers. section 8.2
∂ℒ/∂θ
i = ∂ℒ/∂y ∂y/∂f
l ∂f
l/∂f
l-1 ⋯ ∂f
i/∂θ
i
The equation expresses loss derivatives as a chain through network layers. section 8.2


























