New: "A mathematical theory of cooperative communication". arxiv.org/abs/1910.02822
Wondering about foundations for human-human and human-machine/robot cooperation? We have you covered! We show that models across these literatures are approximations of Optimal (belief) transport
Wondering about foundations for human-human and human-machine/robot cooperation? We have you covered! We show that models across these literatures are approximations of Optimal (belief) transport
Probabilistic cooperative Theory of Mind reasoning in the discrete case is Sinkhorn scaling. It gives the maximum likelihood belief transport plan!
Softmax greedy choice induces instability between partner's plans. Probability matching is maximally robust to violations of common ground! It is robust: Sinkhorn is a continuous (even smooth in some cases) function!
We derive an online method of correcting discovered violations of common ground through gradient methods. Theory of mind reasoning is computable through gradient methods! (And others!)
Excited to be able to prove these and other results through the connection between probabilistic cooperative models, Sinkhorn, and OT
Work with fantastic folks from my lab: Pei Wang, Junqi Wang, and Pushpi Paranamana
