Complex Boltzmann networks and one stage learning

The author considers the discrete Boltzmann net with complex activations and weights. In particular he shows the following: it is possible to define an energy Hamiltonian with properties similar to the one in the real case; with appropriate clamping of a set of pattern units, the usual two-stage learning can be accomplished in one stage; and this model is still strongly related to the physical Spin Glass model, although not to the simple Ising model. Simulations, extensions, and future work are discussed