Material handling: a conservative domain for neural connectivity and propagation

摘要: Two important components of connectionist models are the connectivitV between units and propagation rule for mapping outputs to inputs units. The biological domains where these usuallV applied nonconservative, in that a single output signal produced bV one unit can become input zero, one, or manV subsequent matrices rules common reflect this nonconservativism both learning performance. CASCADE is sVstem performing material handling discrete parts manufacturing environment. We have described elsewhere architecture implementation CASCADE [PARU86a] its formal correspondence [PARU86c], [PARU87a] with PDP model [RUME86]. signals passes correspond phVsical objects, thus must obeV certain conservation laws not observed conventional neural architectures. This paper brieflV reviews problem domain structure CASCADE, describes CASCADE's scheme maintaining information propagating signals, reports some experiments sVstem.