As with any use case for evolutionary computation, representation matters a lot. What space of possible cupcakes will you be using? I think the threshold for people to decide to buy a cupcake is too high for this to work, too — you're mainly going to discover that people don't buy pixel-noise cupcakes at all, and then the experiment's over because you don't get any more data. Unless people do buy them, in which case people are different enough that I don't think you're going to find a reliable gradient.
![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
no subject