PruningNeuralNetworksviaCoresetsandConvex Geometry: TowardsNoAssumptions