Neural Automata Research Articles

Criterion of Neural Generation of Automaton Functions with Time Delay

Neural automata being finite automata represented by a composition of threshold Boolean functions with time delay are considered in the paper. A simple criterion to test whether a given automaton with time delay can be represented by a neural automaton is proved.

The Category of Machines that Become Conscious

A conscious machine category is defined for which the process of becoming conscious is important. This is distinguished from the artificial intelligence category of machines and contrasted with being a conscious living organism. An example is given of a neural automaton for which becoming conscious is seen to be equated to the growth of a depictive state structure. Existing works and future possibilities are examined against this categoric distinction.

Impulse Neuron and Cellular Neural Automaton are Asymptotically Equivalent

In the article, it is established the asymptotic equivalence of dynamics of the neural networks consisting of the impulse neurons and the neural networks built of the neural cellular automata of different kinds (autogenerators and detectors). Such an equivalence takes place for appropriate parameters values and non-intersection of input impacts for each neuron. In the first chapter, we describe the model of impulse autogenerator neuron and two different models of impulse detector neuron. For these models, we prove statements about duration of the latent period for a neuron under impact. In the second chapter, we describe the model of a neural cellular automaton with autogenerator dynamics and a modification of this model with detector dynamics. For these models, we also prove statements about the latent period for a neural automaton under impact. In the third chapter, we prove some statements about asymptotic equivalence of the impulse neurons and neural cellular automata of different kinds.

Instabilities in attractor networks with fast synaptic fluctuations and partial updating of the neurons activity

We present and study a probabilistic neural automaton in which the fraction of simultaneously-updated neurons is a parameter, rhoin(0,1). For small rho, there is relaxation towards one of the attractors and a great sensibility to external stimuli and, for rho > or = rho(c), itinerancy among attractors. Tuning rho in this regime, oscillations may abruptly change from regular to chaotic and vice versa, which allows one to control the efficiency of the searching process. We argue on the similarity of the model behavior with recent observations, and on the possible role of chaos in neurobiology.

Variability of structure of abstract neural automata and the ability of thought

In the view of philosophy, a concept is the highest product of the human brain. Demonstrates abstract neural automata (ANA) – the more perfect of the brain's models have the ability of transition of concept – ability of thought. The transition of the concept of ANA results from non‐uniqueness of its limit Gibbs measure, which is the variability of structure of ANA.

Abstract neural automata on compact Riemannian manifold

By means of topological conjugate transformation, the previous theory of abstract neural automata (ANA) on d‐dimensional (d≥1) integer lattice is extended to compact Riemannian manifold. This paper points out emphatically that intelligence of ANA is related to the geometrical features. The greater the volume of relative plane, the stronger the intelligence; curved Riemannian manifold X˜ configuration space of ANA are locally flat such that the cognitive process of NAN limits the Gibbs' probability measure for a sufficiently small time i.e. the cognitive process of ANA can determine the solution in a sufficiently small time the problem. This hypothesis was supported by studying the human brain, in particular by studying Einstein's brain.

Abstract neural automata: variability of structure, thought and Riemannian volume

Believes that in the view of philosophy, a concept is the highest form of activity of human brain. This paper demonstrates Abstract Neural Automata and a more perfect brain's models that have the ability of transition of concept‐ability of thought. The transition of the concept of Abstract Neural automata results from the non‐uniqueness of its limit Gibbs measure‐variability of the structure of Abstract Neural Automata.By means of topological conjugate transformation, the previous theory of Abstract Neural Automata on a d‐dimensional (d≥1) integer lattice is extended to the compact Riemannian manifold. We have pointed out emphatically that functions of cognition and thought of Abstract Neural Automata depend crucially on its topological and the Riemannian structure, particularly, on its Riemannian volume of some relative places which are relative learning, memory, cognition and thought. Furthermore, the larger the Riemannian volume, the stronger the intelligent function. In the study of the human brain, and in particular, Einstein's brain, one has discovered such information.

Abstract neural automata

The theory of abstract neural automata has been proposed previously. Believes that in studying change of the structure of abstract neural automata, genetico‐variable structure must be considered. Abstract neural automata are considered to be cognizant machines, machines of thought. The basic effect on cognition and on the thought of heredity, such as genes is considered to be important in this discussion. Following the proof of the theorem of existence of abstract neural automata whose state space of single neuron (stochastic automaton) is Euclidean space Ed, the theorem of ergodicity of evolutionary process of genetico‐variable structure of abstract neural automata is proved.

Communications and forum

The purpose of this paper is to consider the following three kinds of cases: first, where a network of stochastic automata consists of a neuron of several types; second, where not only pair‐wise interaction of the neuron occurs, but also that between groups of the neuron; and, third, where neurons are located in d‐dimensional Euclidean space. This is based on neurons of different types having been distinguished by the neurophysiological theory of the retina of vertebrates, where the retina consists of overlapping sheets which are composed of cells receiving light, bipolar cells, ganglion cells, respectively; there are also horizontal cells and amacrine; a functional unit of the central nervous system of vertebrates is composed of the group of neurons which have the form of a network; and the case where neurons are located in d‐dimensional Euclidean space will extend theoretical significance and practical significance of abstract neural automata. The existence of abstract neural automata will be proved in all three cases.

Abstract neural automata

Considers neurons, neural networks and neural fields from the viewpoint of abstract automata. Introduces Abstract neural automata (ANA) to explain and to provide a mathematical description of neural functions and theory. Surveys some current literature including that concerned with Boltzmann machines and the author’s own view of an associative Boltzmann neural model. Provides definitions and theorems to support the author’s theses on cognition, the human brain and the role of ANA in the understanding of neural networks.

Discrete vector quantization for arbitrary distance function estimation

There are currently many vastly different areas of research involving adaptive learning. Among them are the two areas that concern neural networks and learning automata. This paper develops a method by which the general philosophies of vector quantization (VQ) and discretized automata learning can be incorporated for the computation of arbitrary distance functions. The latter is a problem which has important applications in logistics and location analysis. The input to our problem is the set of coordinates of a large number of nodes whose internode arbitrary "distances" have to be estimated. To render the problem interesting, nontrivial, and realistic, we assume that the explicit form of this distance function is both unknown and uncomputable. Unlike traditional operations research methods, which use optimized parametric functional estimators, we have utilized discretized VQ principles to first adaptively polarize the nodes into subregions. Subsequently, the parameters characterizing the subregions are learned by using a variety of methods (including, for academic purposes, a VQ strategy in the meta-domain). After an initial training phase, a system which achieves distance estimation attempts to yield an estimate of any node-pair distance without actually deriving an explicit form for the unknown function. The algorithms have been rigorously tested for the actual road-travel distances involving cities in Turkey and the results obtained are conclusive. Indeed, these present results are the best currently available from any single or hybrid strategy.