PHEW : Constructing Sparse Networks that Learn Fast
            and Generalize Well Without Training Data

                  Shreyas Malakarjun Patil 1 Constantine Dovrolis 1

             Abstract                 erally, such methods use an edge scoring mechanism for
   Methods that sparsify a network at initialization      eliminating the less important connections. Popular scoring
   are important in practice because they greatly im-      mechanisms include weight magnitudes (Han et al., 2015 ...