Figure 7: For a board of width 4, networks trained with or without junk performed comparably.