Neural Computation
September 1994, Vol. 6, No. 5, Pages 944-956
(doi: 10.1162/neco.1994.6.5.944)
Realization of the “Weak Rod” by a Double Layer Parallel Network
Article PDF (480.63 KB)
Abstract
The weak rod can be realized by a very natural double layer network with feedback. While the second layer is linear, the first layer is significantly nonlinear. Elements associated with the first layer are called “synaptic fuses.” An MRF argument is used to eliminate the line variables. The network makes what is cut transparent when a crease or a step edge is present.