ARegularizing Optimal Transportwithf-Divergences Namef(v)f (v)f 0 Dom(f (v)) Kullback-Leiblervlog (v) exp(v 1) exp(v 1) v 2R Reverse KL log (v) log (1v) 1 1vv< 0 Pearson 2 (v 1)2 v
–Neural Information Processing Systems
Theorem, 4.2 - Optimizing Neural Nets.SupposeJ (', )is 1s-stronglysmoothinl1 norm. Let n bethe Sinkhornplanbetweenˆµand ˆ atregularization , andletD: = diam (X).
Neural Information Processing Systems
Apr-25-2026, 04:57:29 GMT