Holographic Neural Technology, Systems and Applications

Described is a "neural" operating system based on Holographic/Quantum Neural Technology (HNeT). The core of the HNeT technology applies Hilbert space operations in both the updating of cortical memory and generation of response recall, similar in form to the QM wave function. Within HNeT, information is represented by sets of complex scalars, leading to a natural predilection towards frequency domain representations of stimuli. Conversion of real valued information sets to frequency domain representation leads to a number of desirable qualities, such as orthogonalization of highly asymmetric or non-orthogonal pattern sets, a distributed representation of information, as well as an effective means for data reduction (i.e. Fourier quantization). Higher order frequency domain representations facilitate extraction of invariants that define discriminating features, often intractable using conventional pattern classification methods. This is performed utilizing a form of neural plasticity that scans the set of higher order harmonics for discovery of such invariants. One of the most salient operational aspects of holographic/ quantum neural technology is the reduction in computational complexity over more traditional neural networks (NN). For instance, HNeT requires only binary cell structures in quite advanced application areas. Holographic/quantum neural technology also provides a dramatic increase in speed of learning and learning accuracy over traditional NN methods. The HNeT process facilitates real-time learning, in which large data sets may be learned to high accuracies following one training epoch. The HNeT core processes have been extended considerably over the past few years to incorporate a number of auxiliary features. These features include application of higher order combinatorics for pre-process of input stimuli, the application and advanced control of neural plasticity, the use of cell assemblies that facilitate "super-cell" structures similar in form to neo-cortical assemblies, and unsupervised learning structures that facilitate hyperincursive and spatio-temporal learning paradigms, among others. Current work is directed towards structures that facilitate temporal accumulation of spatial patterns at the preprocess level, prior to entry into cortical cell structures. These accumulative structures possess certain analogous features to the thalamus, permitting synthetic neural systems to learn spatio-temporal patterns such as speech. The HNeT system is biologically motivated, possessing an application programming interface (API) that allows the user to allocate specific cell types based on the granule, pyramidal, stellate, and Purkinje cells of the cerebellum and neo-cortex. This operating system permits the user to flexibly configure cell assemblies, and build cortical structures comprised of anywhere from 2 to several thousand cells.