How Does Deep Neural Network Work?

How does it work specifically for those applications that you mentioned, also, the hard to approximate functions? Well, so the way we see this working, so you have a slow to calculate function in three, five, 10, 20 dimensions. You sample it at a relatively small number of points,. Is that offline, let's say, not in your main risk management application? And then you fit the functional form that we propose, the generalized stochastic sampling form. Instead of calling your slow function, you call your fast approximation. How does it compare to deep neural network? We can say that, of course, our methods work much faster.