Information Technology Reference
In-Depth Information
It was shown that in neural structures with weights that follow the QHO model,
the weights update is described by Langevin's stochastic differential equation. It
was proved that conventional gradient algorithms are a subcase of Langevin's
equation. It can be also stated that neural networks with crisp numerical weights
and conventional gradient learning algorithms are a subset of neural structure
with weights based on the QHO model. In that way the complementarity between
classical and quantum physics was also validated in the field of neural computation.
Search WWH ::




Custom Search