Isn’t backpropagation just an efficient method for getting a gradient? It sounds like what he’s looking for an alternative to is the idea of adjusting parameter

Isn’t backpropagation just an efficient method for getting a gradient? It sounds like what he’s looking for an alternative to is the idea of adjusting parameters to match labeled training data.