Multi-level Synergetic Computation in Brain

Patterns of activities of neurons serve as attractors, since they are those neuronal configurations which correspond to minimal ’free energy’ of the whole system. Namely, they realize maximal possible agreement among constitutive neurons and are most-strongly correlated with some environmental pattern. Neuronal patterns-qua-attractors have both a material and a virtual aspect. As neuronal patterns, on the one hand, patterns-qua-attractors are explicit carriers of informational contents. As attractors, on the other hand, patterns-qua-attractors are implicit mental representations which acquire a meaning in contextual relations to other possible patterns. Recognition of an external pattern is explained as a (re)construction of the pattern which is the most relevant and similar to a given environmental pattern. The identity of the processes of pattern construction, re-construction and Hebbian short-term storage is realized in a net. Perceptual processes are here modeled using Kohonen’s topology-preserving feature mapping onto cortex where further associative processing is continued. To model stratification of associative processing because of influence from higher brain areas, Haken’s multi-level synergetic network is found to be appropriate. The hierarchy of brain processes is of ”software”-type, i.e. virtual, as well as it is of ”hardware”-type, i.e. physiological. It is shown that synergetic and attractor dynamics can characterize not only neural networks, but also underlying quantum networks. Neural nets are alone not sufficient for consciousness, but interaction with the quantum level might provide effects necessary for consciousness, like, for instance, ultimate binding of perceptual features into an unified experience. It is mathematically demonstrated that associative neural networks realize information processing analogous to the quantum dynamics. Parallels in the formalism of neural models and quantum theory are listed. Basic elements of the quantum versus neural system (modeled by formal neurons and connections) are very different, but their collective processes obey similar laws. Specifically, it is shown that neuron’s weighted spatio-temporal integration of signals corresponds to the Feynman’s version of the Schrödinger equation. In the first case weights are synaptic strengths determined by the Hebb or delta correlation rule; in the second case weights are Green functions or density matrices. In both cases encodings of pattern-correlations represent memory. (Re)construction of a neuronal pattern-qua-attractor is analogous to the ”wave-function collapse”. Transformations of memory (or sub-conscious) representations to a conscious representation is modeled in the same way. Found mathematical analogies allow translation of the neural-net ”algorithm”, which in author’s simulations works very well, into a quantum one. This indicates how such quantum networks, which might be exploited by the sub-cellular levels of brain, could process information efficiently and also make it conscious.