Makes a hella lot of sense to use this instead of Monte Carlo gradients. Difficulty is that we're not using a automatic differentiation backend or utility, so would need to specify model gradients manually? Links into broader points about making models amenable to stochastic optimization (mini-batches) as well as using non-numerical gradients.
Makes a hella lot of sense to use this instead of Monte Carlo gradients. Difficulty is that we're not using a automatic differentiation backend or utility, so would need to specify model gradients manually? Links into broader points about making models amenable to stochastic optimization (mini-batches) as well as using non-numerical gradients.