Provably Strict Generalisation Benefit for Equivariant Models
Bryn Elesedy 1 Sheheryar Zaidi 2
Abstract as Sokolic et al. (2017); Sannai & Imaizumi (2019), cover
It is widely believed that engineering a model to only the worst case performance of algorithms. These works
be invariant/equivariant improves generalisation. use complexity measures to find tighter upper bounds on
Despite the growing popularity of this approach, ...


雷达卡




京公网安备 11010802022788号







