I'm not familiar with topos theory, so can someone suggest a setting in which the richer mathematical structure afforded by topoi pays off, since they are presumably more expensive than traditional Euclidean models? I'm thinking of how protein models use geometric learning, as an example. How costly are topos models anyway? Could topos models be used to improve distillation; retain more capability for a given model size?
esafak•56m ago