Classification　and　regression　prediction　of　access　objects　in　distributed　access　control　model　is　a　very　important　basic　task　in　distributed　access　control　model.　Machine　learning　plays　an　important　role　in　the　field　of　intelligent　access　control　in　the　future,　especially　in　the　application　of　machine　learning　methods　to　solve　classification　and　regression　problems.　The　paper　proposes　a　learning　method　of　distributed　collaborative　training,　which　can　reduce　the　communication　consumption　of　node　policy　update　and　increase　the　access　execution　margin　of　a　single　node.　Improve　model　performance.　©　2020,　Springer　Nature　Singapore　Pte　Ltd.