@techreport{TR-IC-PFG-19-20,
   number = {IC-PFG-19-20},
   author  =  {Allana  de Macedo Idalgo and Alan Carvalho Corrales and
                   Lucas Alves Racoci and Ram\'irez Rivera,},
   title = {{Combinatory Study on Graph Network Predictions}},
   month = {June},
   year = {2019},
   institution = {Institute of Computing, University of Campinas},
   note = {In English, 25 pages.
    \par\selectlanguage{english}\textbf{Abstract}
       To test Deepmind's Graph Nets framework against invariance with
       respect to permutations in graphs representation, we used three
       different  problems  in  at least four different contexts: with
       permuted  and  non-permuted  training  sets;  and  permuted and
       non-permuted  testing  sets.  In all these contexts, the common
       behavior  was  that permuting testing sets reduced the accuracy
       of  the  networks  and  permuting training sets produced better
       results.  This  supports the argument that the algorithms might
       not  be  learning from the graph structure itself, but from the
       graph's computational representation.
  }
}