Chain rule
Chain rule: The differentiation rule for composite functions, multiplying local sensitivities along the computation path. Backpropagation applies it systematically from outputs to parameters. Repeated products also explain amplification or attenuation of deep gradients. This perspective connects the definition with practical behaviour in artificial intelligence systems.
In italiano: Regola della CatenaLegge IV · Loss, gradiente e ottimizzatoriBackpropagation applies it systematically from outputs to parameters. Repeated products also explain amplification or attenuation of deep gradients.