Witold Kinsner

Contemporary Cybernetics and Its Facets of Cognitive Informatics and Computational Intelligence

This paper explores the architecture, theoretical foundations, and paradigms of contemporary cybernetics from perspectives of cognitive informatics (CI) and computational intelligence. The modern domain and the hierarchical behavioral model of cybernetics are elaborated at the imperative, autonomic, and cognitive layers. The CI facet of cybernetics is presented, which explains how the brain may be mimicked in cybernetics via CI and neural informatics. The computational intelligence facet is described with a generic intelligence model of cybernetics. The compatibility between natural and cybernetic intelligence is analyzed. A coherent framework of contemporary cybernetics is presented toward the development of transdisciplinary theories and applications in cybernetics, CI, and computational intelligence.

A unified approach to fractal dimensions

Many scientific papers treat the diversity of fractal dimensions as mere variations on either the same theme or a single definition. There is a need for a unified approach to fractal dimensions for there are fundamental differences between their definitions. This paper presents a new description of three essential classes of fractal dimensions based on: (i) morphology, (ii) entropy, and (iii) transforms, all unified through the generalized entropy-based Renyi fractal dimension spectrum. It discusses practical algorithms for computing 15 different fractal dimensions representing the classes. Although the individual dimensions have already been described in the literature, the unified approach presented in this paper is unique in terms of (i) its progressive development of the fractal dimension concept, (ii) similarity in the definitions and expressions, (iii) analysis of the relation between the dimensions, and (iv) their taxonomy. As a result, a number of new observations have been made, and new applications discovered. Of particular interest are behavioural processes such as dishabituation, irreversible and birth-death growth phenomena e.g., diffusion-limited aggregates, DLAs, dielectric discharges, and cellular automata, as well as dynamical nonstationary transient processes such as speech and transients in radio transmitters, multifractal optimization of image compression using learned vector quantization with Kohonen s self-organizing feature maps (SOFMs), and multifractal-based signal denoising.

A Doctrine of Cognitive Informatics (CI)

Cognitive informatics (CI) is the transdisciplinary enquiry of cognitive and information sciences that investigates into the internal information processing mechanisms and processes of the brain and natural intelligence, and their engineering applications via an interdisciplinary approach. CI develops a coherent set of fundamental theories and denotational mathematics, which form the foundation for most information and knowledge based science and engineering disciplines such as computer science, cognitive science, neuropsychology, systems science, cybernetics, software engineering, knowledge engineering, and computational intelligence. This paper reviews the central doctrine of CI and its applications. The theoretical framework of CI is described on the architecture of CI and its denotational mathematic means. A set of theories and formal models of CI is presented in order to explore the natural and computational intelligence. A wide range of applications of CI are described in the areas of cognitive computers, cognitive properties of knowledge, simulations of human cognitive behaviors, cognitive complexity of software, autonomous agent systems, and computational intelligence.

A radio transmitter fingerprinting system ODO-1

This paper presents a new method for the capture, analysis, and classification of radio transmitter transients. This method involves the use of a capturing subsystem consisting of an Icom IC-R7000 communications receiver and a Sound Blaster 16 sound card running on a PC. The radio transients are sampled at 44,100 samples per second and have 16 bits accuracy. Once the transmitter transient has been captured, a genetic algorithm selects the critical features from the wavelet coefficients for classification. The selected wavelet coefficients are considered to be fingerprints, and are presented to a back propagation neural network for transmitter classification. The capturing and analysis system, ODO-1, is able to classify both transients of the same model type as well as individual transmitters with 100% accuracy on a small data base of transmitter fingerprints.

Separation performance of ICA on simulated EEG and ECG signals contaminated by noise

Evaluates the performance of the extended-infomax independent component analysis (ICA) algorithm in a simulated biomedical blind source separation problem. Independent signals representing an alphawave and a heartbeat are generated and then mixed linearly in the presence of white or pink noise to simulate a one-minute recording of an electroencephalogram and electrocardiogram. The selected ICA algorithm separates the white and pink noises equally well. The maximum estimation signal-to-noise ratio of the source estimates is equivalent to the added noise level, so the separation is optimum to second-order. The higher-order demixing performance, as measured by the Amari index, indicates that when the noise contamination exceeds the mixing contamination the ICA separation is reduced. These results represent a lower bound to the performance of extended-infomax ICA in noisy, time-correlated electrophysiological conditions.

Competing ICA techniques in biomedical signal analysis

We present the background for the statistical decomposition of a signal called independent component analysis (ICA) and survey its application to blind source separation (BSS). We review principal component analysis (PCA), and gradient and cumulant ICA techniques for the complete noiseless BSS problem (more sensors than sources). Results for noisy systems are also discussed. The application of these techniques in the analysis of biomedical signals like EEG, ECG and fMRI, and their early success, is reviewed. We also propose the separation of the current EEG and ECG electrical recordings into independent brain (iEEG) and heart signals (iECG) in order to provide better signals for compression, browsability, and noninvasive medical diagnosis.

Is Entropy Suitable to Characterize Data and Signals for Cognitive Informatics

This paper provides a review of Shannon and other entropy measures in evaluating the quality of materials used in perception, cognition and learning processes. Energy-based metrics are not suitable for cognition, as energy itself does not carry information. Instead, morphological (structural and contextual) as well as entropy-based metrics should be considered in cognitive informatics. The data and signal transformation processes are defined and discussed in the perceptual framework, followed by various classes of information and entropies suitable for characterization of data, signals and distortion. Other entropies are also described, including the Renyi generalized entropy spectrum, Kolmogorov complexity measure, Kolmogorov-Sinai entropy and Prigogine entropy for evolutionary dynamical systems. Although such entropy-based measures are suitable for many signals, they are not sufficient for scale-invariant (fractal and multifractal) signals without complementary measures.

Signal classification through multifractal analysis and complex domain neural networks

This paper describes a system capable of classifying stochastic self-affine nonstationary signals produced by nonlinear systems. The classification and the analysis of these signals are important because these are generated by many real-world processes. The first stage of the signal classification process entails the transformation of the signal into the multifractal dimension domain, through the computation of the variance fractal dimension trajectory (VFDT). Features can then be extracted from the VFDT using a Kohonen self-organizing feature map. The second stage involves the use of a complex domain neural network and a probabilistic neural network to determine the class of a signal based on these extracted features. The results of this paper show that these techniques can be successful in creating a classification system which can obtain correct classification rates of about 87% when performing classification of such signals without knowing the number of classes.

System Complexity and Its Measures: How Complex Is Complex

The last few decades of physics, chemistry, biology, computer science, engineering, and social sciences have been marked by major developments of views on cognitive systems, dynamical systems, complex systems, complexity, self-organization, and emergent phenomena that originate from the interactions among the constituent components (agents) and with the environment, without any central authority. How can measures of complexity capture the intuitive sense of pattern, order, structure, regularity, evolution of features, memory, and correlation? This chapter describes several key ideas, including dynamical systems, complex systems, complexity, and quantification of complexity. As there is no single definition of a complex system, its complexity and complexity measures too have many definitions. As a major contribution, this chapter provides a new comprehensive taxonomy of such measures. This chapter also addresses some practical aspects of acquiring the observables properly.

Speech segmentation using multifractal measures and amplification of signal features

This paper describes a fast multiscale time-domain technique for the analysis of natural speech waveforms in the presence of noise. The technique is based on the variance fractal dimension trajectory algorithm that is used not only to detect the external boundaries of an utterance, but also its internal pauses representing the unvoiced speech. The algorithm can also identify internal features of phonemes. The features can be amplified so that the speech utterances can be segmented into sentences, words and phonemes. This approach is superior to other energy-based boundary-detection techniques. These observations are based on extensive experimental results on speech utterances digitized at 44.1 kilosamples per second, with 16 bits in each sample.