We provide an axiomatic foundation for a class of neural-network models applied to decision-making under risk, called neural-network expected utility (NEU) models. Motivated by classic experimental findings, we weaken the independence axiom in a novel way. We show how to use simple neurons,...