Wouldn't such local invariance tie in with flatness or shallowness of the minima ? This would tie in with the observation that flat/shallow minimas are easier

Wouldn't such local invariance tie in with flatness or shallowness of the minima ? This would tie in with the observation that flat/shallow minimas are easier to find with stochastic gradient descent and such weights generalise better.