There are well-documented examples of sudden switches in human cognitive states, including visual illusions. Cognitive switches, however, are much more common than often assumed. In fact, recent research shows that abrupt switches happen several times per second in the human mind. It is of interest to explore the neural mechanisms which may lead to the observed cognitive switches and their significance in intelligent behaviors. In this work, first we briefly summarize the state-of-art of detecting cognitive switches and describe the corresponding neural processes. We argue that rapid switches between relatively stable states are inevitable attributes of brain dynamics and they are the very source of intelligent behaviors.The key question is whether the switching patterns of brain dynamics and human cognition are obscure evolutionary/historical artifacts which should be ignored in engineering designs, or they are crucial manifestations of intelligence to be incorporated in AI systems as well. Our approach is based on the observation that intelligence is neither local nor global, rather it is a delicate balance between integration and fragmentation tendencies, and sequential switches are the expressions of such competing tendencies. This motivates the development of practically useful modeling tools for intelligence using switching oscillatory dynamics. A practical model is based on percolation theory and phase transitions over random graphs, which provide a powerful tool for rigorously describe neural processes as intermittent phase transitions over the cerebral cortex. The introduced results lead to recommendations to use sequential switching patterns as the underlying modus operandi of sustainable AI systems, which can become true partners of human beings in future endeavors.

EEG-based identification of brain states has advanced significantly, enabling cognitive monitoring during daily tasks. Brain dynamics can be studied as a Markovian, Semi-Markovian, or Non-Markovian stochastic process, prescribed by Transition Probability Matrices derived from EEG measurements. This study models brain dynamics as discrete Markov chains using second-by-second dominant frequencies derived from the power spectrum of high-density array EEG signals. By analyzing transition and limiting probabilities across modalities, we reveal distinct neural signatures that differentiate engaged from meditative states. Though preliminary, these findings highlight the method’s potential to enhance BCI systems and deepen our understanding of brain dynamics, brain health and the benefits of meditation in future studies.

Embodiment is a key aspect of human intelligence, related to our ability of to identify the context of the individual experiences at a given time, corresponding to the natural constraints represented by our body. Learning from higher cognitive functions and social coordination between humans can support building intelligent robot systems and facilitates harmonious human-machine interactions. This position paper provides an overview of neural structures and neural dynamics contributing to human cognitive functions, including multisensory integration, Gestalt formation, perception, and building sensory associations. Embodied cognitive principles are illustrated through the intentional action-perception cycle. The results are applied to the design novel algorithms for brain-inspired cognitive robotics. Example scenarios include imitation learning, and the emergence of dialogue patterns in social robotics settings.

For many years, topological data analysis (TDA) and deep learning (DL) have been considered separate data analysis and representation learning approaches, which have nothing in common. The root cause of this challenge comes from the difficulties in building, extracting, and integrating TDA constructs, such as barcodes or persistent diagrams, within deep neural network architectures. Therefore, the powers of these two approaches are still on their islands and have not yet combined to form more powerful tools for dealing with multiple complex data analysis tasks. Fortunately, we have witnessed several remarkable attempts to integrate DL-based architectures with topological learning paradigms in recent years. These topology-driven DL techniques have notably improved data-driven analysis and mining problems, especially within graph datasets. Recently, graph neural networks (GNNs) have emerged as a popular deep neural architecture, demonstrating significant performance in various graph-based analysis and learning problems. Explicitly, within the manifold paradigm, the graph is naturally considered as a topological object (e.g., the topological properties of the given graph can be represented by the edge weights). Therefore, integrating TDA and GNN is considered an excellent combination. Many well-known studies have recently presented the effectiveness of TDA-assisted GNN-based architectures in dealing with complex graph-based data representation analysis and learning problems. Motivated by the successes of recent research, we present systematic literature about this nascent and promising research direction in this article, which includes general taxonomy, preliminaries, and recently proposed state-of-the-art topology-driven GNN models and perspectives.

Significant efforts has been made in recent years to identify neural correlates of consciousness. The present report discusses several concepts used in models and interpretations of brain measurements, including self-organized criticality, neural avalanches, phase transitions, and scale-free structures and dynamics. These concepts have various advantages and limitations when applied to brain dynamics and cognitive functions, in particular consciousness. Analysis of the properties of the cortical neural tissue indicates that non-local interactions manifested in the projections by long axons lead to the emergence of spatiotemporal activity patterns, which are unique to the cortex and have not been observed in any other physical or biological substrates. A relevant graph-theoretical model, which includes nonlocal effects, has been used to interpret the experimental findings. We propose the hypothesis that the non-local interactions produce unique physiological conditions with large-scale phase gradients and switches between synchrony and absence of synchrony, are the manifestations of conscious experience. We expand on the Global Workspace Theory (GWT) of consciousness to illustrate the hypothesis and outline the directions for future research.

Recent advances in human brain monitoring provide increasingly detailed insights into the spatio-temporal neurodynamic processes contributing to cognition and consciousness. The view that cognition is not a smooth temporal process, rather it is a sequence of metastable states which have a duration of around 100 ms, becomes increasingly recognized. The cinematic theory of cognition is a potential approach to interpret these experimental findings. Recent extensions explore the possible link between the cinematic sequence and conscious broadcast events postulated by the Global Workspace Theory (GWT). The present work summarizes the existing arguments on this issue and elaborates on potential implications of this theory on the development of novel technologies for humancentered and efficient human-computer interaction.

Scalp electroencephalography (EEG) provides a practical tool for the identification and characterization of various brain states, including healthy and diseased conditions. We measure brain dynamics on the scalp via a HydroCel Geodesic Sensor Net, 128 electrodes dense-array EEG. We compute the pragmatic information index (PI) by Hilbert analysis of the EEG signals of 20 healthy participants. We compare 6 task modalities, combining different audio-visual stimuli, leading to various mental states, predominantly relaxed versus engaged. We analyze PI values to classify different brain states. We show significant differences between the measured neural signatures depending on the task modalities, based on both qualitative and quantitative analysis. The results can help to develop tools for medical diagnostics of stress-related mental conditions.

The new generation of digital intelligence technology enables knowledge creation, dissemination, and application to undergoing parallel changes. Scientific systems face an increasingly uncertain, diverse, and complex environment, making adopting multidisciplinary, interdisciplinary, and transdisciplinary approaches to research issues inevitable. Existing scientific systems follow linear value streams, leading to problems, such as inefficiency, unfairness, and knowledge monopoly. Decentralized science (DeSci) is a new scientific development paradigm based on Web3, Metaverses, and decentralized autonomous organizations and operations (DAOs) technologies, that can solve organizational and management problems in scientific systems through organizing, coordinating, and executing techniques. However, new economic theories and methods are still needed to effectively solve the problem of linear value flow in scientific systems. Metaeconomics based on the parallel intelligence theory, also known as decentralized economics (DeEco), provides a new approach and idea for redesigning the economic system of scientific markets. Thus, this article proposes a research framework and core mechanisms of DeSci MetaMarkets based on parallel economic theory to provide effective and practical methodologies for scientific system governance.

The presence of respiration-locked sensory cortical activity has been documented in various mammalian species in recent years, including cognitively-relevant gamma oscillations (30-80 Hz). This work builds on previous evidence suggesting that respiration has a direct influence on oscillatory activity in human sensory, motor and association cortical areas. The entrainment of cortical activity patterns by respiratory phase suggests a direct influence of respiration on cognitive processing, which represents a possible neuronal mechanism behind the well-documented but unexplained effects of respiratory exercises on emotional and cognitive functions. We explore possible interpretation of those findings in the context of the cinematic view of cognition, in particular cortical phase transitions. In addition to traditional Fourier-based correlational analysis of cortical signals, we introduce Hilbert analysis, which allows to monitor rapid phase synchronization-desynchronization transitions. Our results support the hypothesis that respiration-locked cortical activity is linked to phase transitions in the cortex, measured by discontinuities of the instantaneous phase of the analytic signal determined by the Hilbert transform. Taken together, these findings suggest that respiration acts as master clock exerting a subtle but unfailing synchronizing influence on the temporal organization of dynamic cortical activity patterns and the cognitive processes they control.

Decentralized science (DeSci) is a hot topic emerging with the development of Web3 or Web3.0 and decentralized autonomous organizations (DAOs) and operations. DeSci fundamentally differs from the centralized science (CeSci) and Open Science (OS) movement built in the centralized way with centralized protocols. It changes the basic structure and legacy norms of current scientific systems via reshaping the cooperation mode, value system, and incentive mechanism. As such, it can provide a viable path for solving bottleneck problems in the development of science, such as oligarchy, silos, and so on, and make science more fair, free, responsible, and sensitive. However, DeSci itself still faces many challenges, including scaling, balancing the quality of participants, system suboptimal loops, lack of accountability mechanism, and so on. Taking these into consideration, this article presents a systematic introduction of DeSci, proposes a novel reference model with a six-layer architecture, addresses the potential applications, and also outlines the key research directions in this emerging field. This article is committed to providing helpful guidance and reference for future research efforts on DeSci.

Spiking neural networks (SNNs) have great potential for energy-efficient implementation of Deep Neural Networks (DNNs) on dedicated neuromorphic hardware. Recent studies demonstrated competitive performance of SNNs compared with DNNs on image classification tasks, including CIFAR-10 and ImageNet data. The present work focuses on using SNNs in combination with deep reinforcement learning in ATARI games, which involves additional complexity as compared to image classification. We review the theory of converting DNNs to SNNs and extending the conversion to Deep Q-Networks (DQNs). We propose a robust representation of the firing rate to reduce the error during the conversion process. In addition, we introduce a new metric to evaluate the conversion process by comparing the decisions made by the DQN and SNN, respectively. We also analyze how the simulation time and parameter normalization influence the performance of converted SNNs. We achieve competitive scores on 17 top-performing Atari games. To the best of our knowledge, our work is the first to achieve state-of-the-art performance on multiple Atari games with SNNs. Our work serves as a benchmark for the conversion of DQNs to SNNs and paves the way for further research on solving reinforcement learning tasks with SNNs.

A 50th birthday is an important milestone in the life of any individual and certainly in the development of collectives and organizations. January 2021 is such a milestone in the life of the IEEE Transactions ofSystems, Man,andCybernetics, which had its very first issue published in January 1971. Our 50th Anniversary Issue celebrates this remarkable achievement by introducing 20 survey articles from early pioneers of our field, as well as works by leading scientists conducting research at the cutting edge of systems science, with a focus on human aspects, cybernetics, with crucial societal impact. Our goal is to provide a vista of the remarkable developments in the past, giving a snapshot of the present state of our field, and indicating possible avenues for future progress in our rapidly changing research discipline.

To commemorate the 50th anniversary of the IEEE Transactions on Systems, Man, and Cybernetics: Systems, this article examines and reports on its past to current topical coverage of systems science and engineering toward exploring the evolving focus of the research community. Results of a systematic bibliometric analysis are presented with associated conclusions, implications, and summary of topical areas. In addition, respective views regarding the current state of the field and where it is headed are offered by recent leaders of the IEEE Systems, Man, and Cybernetics Society, including its continued relevance and role in the advancement of systems technology.

Leading mainstream image processing approaches produce excellent performance using convolutional neural networks trained by backpropagation (BP) learning rules. Unsupervised learning approaches have been popular due to their biological significance, though they typically underperform compared to BP results. In this work, we demonstrate that features extracted in an unsupervised manner using the biologically inspired Hebbian learning rule in a winner-take-all setting, perform competitively with BP on the image classification task. The convolutional filters learned by Hebbian rule are smoother than filters learned using BP. The quality of the two training approaches is compared based on metrics such as the speed of training and classification accuracy. We demonstrate that the extracted features of unsupervised learning are more robust to noise as compared to BP.

Spiking neural networks (SNNs) are a promising candidate for biologically-inspired and energy efficient computation. However, their simulation is restrictively time consuming, and creates a bottleneck in developing competitive training methods with potential deployment on neuromorphic hardware platforms, even on simple tasks. To address this issue, we provide an implementation of mini-batch processing applied to clock-based SNN simulation, leading to drastically increased data throughput. To our knowledge, this is the first general-purpose implementation of mini-batch processing in a spiking neural networks simulator, which works with arbitrary neuron and synapse models. We demonstrate nearly constant-time scaling with batch size on a simulation setup (up to GPU memory limits), and showcase the effectiveness of large batch sizes in two SNN application domains, resulting in ≈880X and ≈24X reductions in wall-clock time respectively. Different parameter reduction techniques are shown to produce different learning outcomes in a simulation of networks trained with spike-timing-dependent plasticity. Machine learning practitioners and biological modelers alike may benefit from the drastically reduced simulation time and increased iteration speed this method enables.

Non-invasive brain imaging techniques are popular tools for monitoring the cognitive state of human participants. This work builds on our previous studies using the HydroCel Geodesic Sensor Net, 256 electrodes dense-array electro-encephalography (EEG). The studies analyze dominant frequencies of temporal power spectral densities for each of the EEG electrodes. The experiments involve three modalities: Meditation, Math Mind, and (c) Open Eyes condition. Here we perform an analysis of the Shannon entropy index and Pearson's skewness coefficient in order to test their fitness to classify different brain states. The results help to develop a comprehensive methodology to understand brain dynamics.

Welcome from the Editor-in-Chief, I am pleased to welcome the readers of the IEEE Transactions on Systems, Man, and Cybernetics: Systems, as the new Editor-in-Chief of the journal. It is my distinct honor to serve as EIC, following the great traditions of the Transactions. I would like to use this opportunity to summarize my thoughts on the state of our Transactions, its past, present, and future perspectives.

Cognition and brain dynamics manifests a delicate balance between processes at various temporal and spatial scales. The conversion between microscopic neural pulses and waves of mesoscopic activity of neural masses is a crucial research subject. In this work we analyze the hierarchy of neural structures and dynamics, with an emphasis on pulse-wave-pulse conversion. Our models describe pulse-to-wave conversion, as well as the feedback of action potentials on neurons in the recovery mode. We study the behavior of neural populations in the background state of activity, as well as under perturbations due to external stimuli. Simulations results are employed for the interpretation of electrocorticogram data obtained with rabbits trained using classical conditioning paradigm.

Embodiment is a key feature of biological intelligence, and energy-awareness can be viewed as the ultimate expression of situated intelligence. Energy constraint is often ignored, or it has just secondary role in typical cutting-edge AI approaches. For example, Deep Learning Networks often require huge amount of data/time/energy/resources, which may not be readily available in various practical scenarios. We outline a simple computational model of spiking neural activity lined link astrocytes responsible for energy management in brains. We analyze oscillatory neural dynamics and its modulation by astrocytes mediating energy constrains. We indicate the potential benefits of the proposed design to be implemented on neuromorphic computational platforms.

Recent advances in brain imaging techniques require the development of advanced models of brain networks and graphs. Previous work on percolation on lattices and random graphs demonstrated emergent dynamical regimes, including zero- and non-zero fixed points, and limit cycle oscillations. Here we introduce graph processes using lattices with excitatory and inhibitory nodes, and study conditions leading to spatio-temporal oscillations. Rigorous mathematical analysis provides insights on the possible dynamics and, of particular concern to this work, conditions producing cycles with very long periods. A systematic parameter study demonstrates the presence of phase transitions between various regimes, including oscillations with emergent metastable patterns. We studied the impact of external stimuli on the dynamic patterns, which can be used for encoding and recall in robust associative memories.

We present a system comprising a hybridization of self-organized map (SOM) properties with spiking neural networks (SNNs) that retain many of the features of SOMs. Networks are trained in an unsupervised manner to learn a self-organized lattice of filters via excitatory-inhibitory interactions among populations of neurons. We develop and test various inhibition strategies, such as growing with inter-neuron distance and two distinct levels of inhibition. The quality of the unsupervised learning algorithm is evaluated using examples with known labels. Several biologically-inspired classification tools are proposed and compared, including population-level confidence rating, and n-grams using spike motif algorithm. Using the optimal choice of parameters, our approach produces improvements over state-of-art spiking neural networks.

Spiking neural networks are motivated from principles of neural systems and may possess unexplored advantages in the context of machine learning. A class of convolutional spiking neural networks is introduced, trained to detect image features with an unsupervised, competitive learning mechanism. Image features can be shared within subpopulations of neurons, or each may evolve independently to capture different features in different regions of input space. We analyze the time and memory requirements of learning with and operating such networks. The MNIST dataset is used as an experimental testbed, and comparisons are made between the performance and convergence speed of a baseline spiking neural network.

Energy-awareness is a relatively less-studied aspect of artificial intelligence and the development of intelligent devices. Here we analyze metabolic processes in brains and their role in higher cognitive activities. To describe important aspects of the energy management in large-scale populations of the cortical tissue, we introduce a novel hierarchical capillary-astrocyte-neuron (CAN) model. CAN is a highly simplified model, still it can describe important aspects of synchronization-desynchronization transitions observed in brain imaging experiments. Energy constraints act as regularization terms in equations of brain dynamics models, producing oscillatory modes across cortical regions. These oscillations are considered neural correlates of higher cognition and awareness. The introduced oscillatory CAN arrays present a possible approach towards developing energy-efficient dynamical memories and learning systems.

We develop a methodology to assess knowledge representation in deep neural networks trained to recognize classes of objects. We measure the abstraction level by studying correlations between the neuron activation levels of different layers based on image class. The approach is developed and tested using CIFAR-10 dataset and MatConvNet toolbox. The results show that different kinds of layers, convolutional or pooling, have different effect on the representation. The observations also point to a tendency for incremental increase in the abstraction measure, sometimes interrupted by more significant jumps, which may indicate a qualitative transition between abstraction levels. We describe and interpret current results and outline the direction of future work.

The human brain is an energy hungry organ. How that brain manages its energy consumption in maintaining its health and executing sensori-motor and cognitive functions is an important but overlooked research area in contemporary cognitive neuroscience. It is argued here that the principal method whereby the human brain manages its energy utilization is through maintaining a relatively elevated level of activity in what can be referred to as “resting state networks” (RSN). The elevated energy consumption in the human brain's varied RSNs is driven and maintained by a physiological mechanism we call the Frame-Formation Energy Cycle (FFEC). Running the FFEC cycle is metabolically expensive and therefore offers a mechanism to explain the increased energy consumption in human-brain RSNs as compared to regions not involved in such networks.

Previous experimental studies on rabbits using electrocorticograms (ECoGs) over the cortical surface indicate spatio-temporal dynamics in the form of amplitude modulation (AM) patterns, which intermittently collapse at theta rates and give rise to rapidly propagating phase modulated (PM) patterns. The observed dynamics have been shown to be of cognitive relevance carrying useful information on the meaning of sensory information perceived by the subject. We have extended these studies to human scalp EEG measurements, which show evidence that cognitively relevant AM and PM patterns are observable by non-intrusive experimental techniques as well. The present work develops experimental techniques for studying cognitively relevant spatio-temporal neural dynamics using a high-density EEG array. Theoretical considerations indicate that the required spatial resolution to detect and categorize amplitude and phase patterns should be in the range of 3–5 mm. A prototype 1-dimensional array (MINDO-48S) has been developed, which has 48 electrodes in a flexible linear array of 5 mm spacing. The present work focuses on the extraction of broadly distributed spatio-temporal patterns, which carry cognitively relevant information. Preliminary analysis of the signal-to-noise ratio indicates that the sensitivity of the experiment allows the predicted AM patterns to be measured.

Many processes in nature display quasi-periodic behavior, including variable stars in distant galaxies and oscillations in brains. In this work we model quasi-periodic lightcurves using neuropercolation, which describes complex spatio-temporal oscillations arising from random cellular automata near criticality. We show that neuropercolation is able to model lightcurves from various stars of the gamma-Doradus type with high accuracy even in the presence of missing data points and unevenly distributed time series. We provide physical interpretation of the obtained results.

This contribution presents a brief review of experimental and theoretical approaches to brain dynamics, including concepts describing the unity of brain, mind, and body. We dedicate this essay to the memory of Walter J. Freeman, III, with a focus on his pioneering work in the past 60 years establishing the new field of computational neuroscience, including mass action in the nervous system, field theories of cognition and intelligence, and his quest towards inventing novel engineering approaches for brain-machine interfaces.

Learning in the mammalian brain is commonly modeled through changing synaptic connections in cortical networks. Dynamical brain models indicate that learning leads to the formation of limit cycle oscillations across cortical areas and that the oscillatory regimes re-emerge when the learnt input is presented to the system. In this work, learning is modeled using a graph-theoretical model, which captures salient characteristics of the learning process. We introduce a random graph that combines a torus with lattice edges and additional random edges, which have power law length distribution. On this graph, we consider bootstrap percolation with excitatory and inhibitory vertices. Theoretical and numerical studies indicate the presence of various dynamical regimes on these graphs. Here, the transitions between fixed-point and limit cycle attractors are analyzed. We link this transition to changes in cortical networks during category learning, which have been observed in animal experiments using electro-cortiograph (ECoG) arrays over sensory cortices. We discuss how learning leads to categorization and strategy formation, and how the theoretical modeling results can be used for designing learning and adaptation in computationally aware intelligent machines.

The effort to integrate emotions into human-computer interaction (HCI) system has attracted broad attentions. Automatic emotion recognition enables the HCI to become more intelligent and user friendly. Although numerous studies have been performed in this field, emotion recognition is still an extremely challenging task, especially in real-world practice usage. In this work, probabilistic neural network (PNN), with advantage of simple, efficient, and easy to train, was employed to recognize emotions elicited by watching music videos from scalp EEG. The publicly available DEAP emotion database was used to validate our algorithms. The powers of 4 frequency bands of EEG were extracted as features. The results show that the mean classification accuracy of PNN is 81.21% for valence(≥5 and <;5) and 81.26% for arousal(≥5 and <;5) across 32 subjects, similar with the results of SVM. In addition, they demonstrate that higher frequency bands (beta and gamma) play more important role in emotion classification than lower ones (theta and alpha). For the purpose of practical emotion recognition system, we proposed a ReliefF-based channel selection algorithm to reduce the number of used channels for convenience in practical usage. The results show that while using PNN, the 98% of the maximum classification accuracy can be obtained with only 9 (for valence) and 8 (for arousal) best channels, however, 19 (for valence) and 14 (for arousal) channels are needed while using SVM.

In this paper we first point out a fatal drawback that the widely used Granger causality (GC) needs to estimate the autoregressive model, which is equivalent to taking a series of backward recursive operations which are infeasible in many irreversible chemical reaction models. Thus, new causality (NC) proposed by Hu et al. (2011) is theoretically shown to be more sensitive to reveal true causality than GC. We then apply GC and NC to motor imagery (MI) which is an important mental process in cognitive neuroscience and psychology and has received growing attention for a long time. We study causality flow during MI using scalp electroencephalograms from nine subjects in Brain-computer interface competition IV held in 2008. We are interested in three regions: Cz (central area of the cerebral cortex), C3 (left area of the cerebral cortex), and C4 (right area of the cerebral cortex) which are considered to be optimal locations for recognizing MI states in the literature. Our results show that: 1) there is strong directional connectivity from Cz to C3/C4 during left- and right-hand MIs based on GC and NC; 2) during left-hand MI, there is directional connectivity from C4 to C3 based on GC and NC; 3) during right-hand MI, there is strong directional connectivity from C3 to C4 which is much clearly revealed by NC than by GC, i.e., NC largely improves the classification rate; and 4) NC is demonstrated to be much more sensitive to reveal causal influence between different brain regions than GC.

Signals measured by electroencephalogram (EEG) arrays were decomposed using Hubert Transformations to produce the spatial amplitude and phase modulation (AM and PM) patterns. Spatial PM patterns intermittently exhibit synchronization-desynchronization transitions. During desynchronization, the spatial PM patterns intermittently conform to conic shapes. These phase cones mark the onset of emergent AM patterns, which carry cognitive content. In this work, various temporal band pass filters were applied to study the frequency dependence of phase cones in the beta-gamma range (10-40 Hz). The results are interpreted in the context of the cognitive cycle of knowledge generation.

In this paper, the basic building blocks of the intentional neurodynamics cycle are studied using probabilistic cellular automata. We apply the mean field neuropercolation model to describe the dynamics of coupled excitatory-inhibitory neural populations. The model exhibits a phase transition between a background point attractor and narrow-band (limit-cycle) behavior, depending of the choice of system parameters. Metastable dynamics near criticality is investigated. We show the existence of various regions with unstable and multiple stable equilibria. The results are relevant to modeling cognitive phase transitions observed in brains during awareness experience.

Propagating phase patterns of neural activity in the cortex have been found to serve as useful markers for the identification of neural correlates of cognition. In this work we develop an automatic method to detect phase propagation in the form of cones using adaptive mixture models. The present work is the first demonstration of this methodology in actual Electrocorticogram (ECoG) data. The paper discusses results obtained with this novel method.

This contribution presents a review of mathematical approaches to modeling brains and higher cognitive activity, including consciousness. We dedicate this paper to John Gerard Taylor on the somber occasion of remembering his lifelong contribution to science, with a focus on his pioneering work on neural networks and brain studies.

Cyber security is an important issue in today's global computer networks. Advanced clustering methods are relevant for efficient data mining over the web. KIII is a biologically plausible neural network model. In its multi-layer architecture there are excitatory and inhibitory neurons, which present lateral, feedforward, and delayed feedback connections between layers in a massive way. KIII has been successfully employed in classification and pattern recognition tasks. In this work we develop a methodology to use KIII for community detection. It is shown that clustering methods that employ KIII related to cybersecurity achieve better results, despite the amount of data available by such application.

Electrocorticogram (ECoG) analysis of human subjects demonstrated that beta-gamma oscillations carry perceptual information in spatial patterns across the cortex when the subjects were engaged in task-oriented activities. A hypothesis was tested that similar patterns could be found in the scalp EEG of human subjects during visual stimulation. Signals were continuously recorded from scalp electrodes and band-pass filtered. The Fast Fourier transform provides the phase, which is used to obtain directional phase information relating to cognitive tasks. Spatial patterns of EEG phase modulation were identified and classified with respect to stimulus. The obtained results suggest that the scalp EEG can yield information about the timing of episodically synchronized brain activity in higher cognitive function, so as to support mechanisms of brain-computer interfacing.

We apply Hilbert transforms to the analysis of phase relationship in elecrocoticogram (ECoG) signals in order to explore a set of meaningful information theoretic measures. This analysis leads to a methodology to derive meaning from experimentally observed brain dynamics under various states induced by sensory stimuli. We explore the possibility to represent periods of habituation and learning based on instantaneous frequency signals and introducing a new set of parameters, based on the concept of pragmatic information.

The brain is a thermodynamic system operating far from equilibrium. Its function is to extract microscopic sensory information from the volleys of action potentials (pulses) that are delivered by immense arrays of sensory receptors, construct the macroscopic meaning of the information, and store, retrieve, and update that meaning by incorporating it into its knowledge base. The function is executed repetitively in the action-perception-assimilation cycle. Each cycle commences by a phase transition, in which the immense population comprising each sensory cortex condenses from a gas-like state to a liquid-like state. It ends with return of the cortex to the expectant gas-like state. We have modeled the microscopic thermodynamics of the cycle using quantum field theory. Our new result is modeling cortical macroscopic thermodynamics with the generalized Carnot cycle, in which the energy required for the construction of knowledge is supplied by brain metabolism and is dissipated as heat by the cerebral circulation. What makes the application possible is the unprecedented precision with which spatial patterns of ECoG are measured, thus providing precise state variables with which to represent energy vs. entropy. We present experimental evidence that these isothermal processes are coupled by adiabatic cooling and heating. We postulate that the action-perception-assimilation cycle comprises minimally three consecutive Carnot cycles required for basic perception, assimilation, and decision, and more cycles with greater complexity of cognitive tasks at hand.

A cognitively inspired framework referred to as Modeling Fields Theory (MFT) is utilized as the basic methodology for learning categorical data, represented by large binary vectors. The presented solution, referred to as accelerated MAP, allows simultaneous learning and selection of the number of models. The key element of accelerated MAP is a steady increase of the regularization penalty combined with gradual decrease of the model vagueness. The operation of this algorithm on real world data is illustrated by applying the algorithm to a text categorization problem. The relationship between the described algorithm and the vague-to-crisp process logic and the dynamical system approach to cognition are discussed.

The KIV (K-4) model is based on biological attributes found in the limbic system of a salamander. Higher forms of organisms including humans have a limbic system which incorporates the sensory cortex, hippocampus and entorhinal cortex/amygdala of the brain. The KIV model has been used successfully for classification and prediction tasks. We propose the use of the KIV model as a metaphor of the limbic system of the human brain. The brain states of normal/pathological (seizure)/restoration are modeled to further understand the pathological states of the brain and propose a titration therapy through this model.

Numerous scientists claim that the memristor may be a real breakthrough in the fields of electronic and circuit design. For this reason, it is important to study what dynamics arise in memristive networks and speculate about how they could be used for meaningful tasks. In this paper, we focus on the phenomenon of percolation in memristive networks, studying the theoretical aspects and performing SW simulations.

Optimization in large-scale networks - such as large logistical networks and electric power grids involving many thousands of variables - is a very challenging task. In this paper, we present the theoretical basis and the related experiments involving the development and use of visualization tools and improvements in existing best practices in managing optimization software, as preparation for the use of “metamodeling” - the insertion of complex neural networks or other universal nonlinear function approximators into key parts of these complicated and expensive computations; this novel approach has been developed by the new Center for Large-Scale Integrated Optimization and Networks (CLION) at University of Memphis, TN.

Experiments conducted in brains by electroencephalographic and magnetoencephalographic techniques reveal widespread coherent oscillations. The oscillations over multiple frequency bands overlap and result in signals with broad spectra. Previous studies showed that various frequencies can be modeled by probabilistic cellular automata with coupled inhibitory and excitatory interactions. In this work we show that coupled oscillator layers can create broad-spectrum chaotic oscillations with power spectral densities over long times segments converging to Brown noise features. Models of cortical neurodynamics provide an interpretation of the observed phenomena.

Understanding neocortical dynamics at mesoscopic level is an important area of experimental neuroscience. ECoG signals reveal us intercortical communications of neural populations in the form of spatial patterns appeared both in amplitude (AM) and phase (PM) modulation of gamma and beta waves. Neocortex shows multiple overlapping autonomous AM-PM phase transition patterns during cognitive processing. We propose an efficient digital filtering method for the capturing abrupt phase transitions defined in analytic phase domain. Phase transitions occurring on the surface of cortex can cover an area ranging from a few hypercolumns to the entire hemisphere. We develop an accurate and adaptable digital filter which is robust to variations in the bandpass filter characteristics and able to separate real transitions from artifacts caused by phase slips. We study complex polynomials which are derived from pseudo spectrum estimation of analytic signals reflecting the dynamics of grid topology. We classify the roots of this complex polynomial defined at each sample according to their location either outside or inside the unit disk in complex plane. The analysis of root characteristics enables us to identify phase transitions. The results are demonstrated using actual ECoG signals.

KIV, a biologically inspired neural network with non-convergent dynamics is considered. This contribution builds on the previous studies concerning components of KIV and considers sensor integration. The method is demonstrated using a simple character recognition task. The significance of the results in the framework of biologically plausible sensor integration is discussed.

Recent studies have focused on the phenomena of abnormal electrical brain activity which may transition into a debilitating seizure state through the entrainment of large populations of neurons. Starting from the initial epileptogenisis of a small population of abnormally firing neurons, to the mobilization of mesoscopic neuron populations behaving in a synchronous manner, a prediction methodology has been formulated that compares the initial epileptogenisis to distant neuron populations. As two neuron populations begin to operate in a synchronized manner, the respective signals phase lock, manifesting into a seizure state. The normal non-linear dynamic signal captured through an EEG enters a semi-periodic state, which can be quantified into a seizure state. A method for capturing synchronous behavior of the pathological brain state is described. An individual patient based phase-locking threshold is introduced for seizure prediction and for differentiating seizure and non-seizure states.

Model field theory (MFT) is a powerful tool of pattern recognition, which has been used successfully for various tasks involving noisy data and high level of clutter. Detection of spatio-temporal activity patterns in EEG experiments is a very challenging task and it is well-suited for MFT implementation. Previous work on applying MFT for EEG analysis used Gaussian assumption on the mixture components. The present work uses non-Gaussian components for the description of propagating phase-cones, which are more realistic models of the experimentally observed physiological processes. This work introduces MFT equations for non-Gaussian transient processes, and describes the identification algorithm. The method is demonstrated using simulated phase cone data.

The KIV model is a biologically inspired hierarchical model that describes non-linear dynamics found in brains. Previous animal and human EEG measurements indicated the presence of jumps in the spatio-temporal EEG patterns, which are relevant to cognitive processing. The present work introduces the KIV model to simulate phase transitions in EEG signals. Phase transitions have non-stationary and intermittent characteristics, which make automated detection a very difficult task. We analyze the simulated EEG signals using various statistical methods. We describe various classification methods to identify simulated phase transitions, which will be used to automate the detection process in actual EEG signals.

Cellular simultaneous recurrent neural network (SRN) has been shown to be a function approximator more powerful than the multilayer perceptron (MLP). This means that the complexity of MLP would be prohibitively large for some problems while SRN could realize the desired mapping with acceptable computational constraints. The speed of training of complex recurrent networks is crucial to their successful application. This work improves the previous results by training the network with extended Kalman filter (EKF). We implemented a generic cellular SRN (CSRN) and applied it for solving two challenging problems: 2-D maze navigation and a subset of the connectedness problem. The speed of convergence has been improved by several orders of magnitude in comparison with the earlier results in the case of maze navigation, and superior generalization has been demonstrated in the case of connectedness. The implications of this improvements are discussed.

K Sets are nonlinear recurrent connectionist models proposed to emulate the brain dynamics. They can be used as dynamic memories encoding in non-equilibrium attractors. As multidimensional non-linear systems, they are extremely hard to analyze. Their dynamics is strongly believed to be related to the itinerant chaos introduced by Tsuda. In this contribution we design a system with attractor switching based on the previously obtained results. This is a step towards better understanding of the K models and building powerful chaotic neural memory systems.

Dynamic logic (DL) approach establishes a unified framework for the statistical description of mixtures using model-based neural networks. In the present work, we extend the previous results to dynamic processes where the mixture parameters, including partial and total energy of the components are time-dependent. Equations are derived and solved for the estimation of parameters which vary in time. The results provide optimal approximation to a broad class of pattern recognition and process identification problems with variable and noisy data. The introduced methodology is demonstrated on the example of identification of propagating phase gradients generated by intermittent fluctuations in non-equilibrium neural media.

Model-based algorithms that attempt to localize targets and estimate the structure within a building using data from external sensors have received much attention in recent years. The potential benefits to homeland security and urban warfare are exceedingly apparent. Accurately estimating the thickness and dielectric constant of the exterior wall could prove to be a critical first step in determining the layout within, i.e., the location of interior walls, doorways, stairwells, etc. However, data collection using a linear sensor arrangement yields an ambiguous two dimensional objective function for the wall parameters, rendering maximum likelihood methods ineffective. We show that a spatially diverse aperture obviates the wall parameter ambiguity and allows accurate estimation of thickness and permittivity using dynamic logic, an iterative model-based approach to maximum likelihood.

We demonstrate the operation of the SODAS approach (self-organized ontogenetic development of autonomous systems) for on-line processing of sensory inputs and onboard dynamic behavior tasking using SRR2K (sample return rover) platform at the planetary robotics indoor facility of JPL. SODAS employs a biologically inspired dynamic neural network architecture operating on the principle of chaotic neural dynamics manifesting intentionality in the style of brains. Intentional behavior includes the cyclic operation of prediction, testing by action, sensing, perceiving, and assimilating the learned features. The experiments illustrate robust obstacle avoidance combined with goal-oriented navigation by the SRR2K robot.

A dynamical neural model that is strongly biologically motivated is applied to learning and retrieving binary patterns. This neural network, known as Freeman's K-sets, is trained with Hebbian rule and habituation to memorize the input patterns by associating them with an attractors formed in the state space. After the patterns are memorized noisy input is given to the network to recover the original. We compare the results of this recall for a different number of memories and compare them with performance of the Hopfield model. We show capacity of the dynamical system exceeds that of the Hopfield network and the noisy recall degrades at a slower pace as the number of the patterns is growing. Experimental results indicate that the critical load paramter, which gives approximation of the network capacity, is higher in K-model than in the Hopfield network. Significant advantage of K-model is achieved at a larger training set size, when compared to Hopfield model.

Cellular simultaneous recurrent neural network has been suggested to be a function approximator more powerful than the MLP's, in particular for solving approximate dynamic programming problems. The 2D maze navigation has been considered as a proof-of-concept task. Present work improves the previous results by training the network with extended Kalman filter (EKF). The original EKF algorithm has been slightly modified. The speed of convergence has been improved by several orders of magnitude in comparison with the earlier results. The implications of this improvement are discussed.

Critical properties of dynamical models of neural populations are studied. Synchronization of the firing of widely dispersed neurons enables the emergence of spatial patterns of cortical activity. Using neuropercolation model introduced in the literature in the past few years, the dynamics of neural populations is studied near critical regimes. Behavior is evaluated with respect to long-range axonal density and sparseness of feedback between excitatory and inhibitory populations. The results show that 1/falphaspectral behavior emerges near criticality, where exponent alpha is function of the critical state of the system.

In this paper we have suggested a sparse three dimensional array model for the brain. Entries of the array are synaptic weights as functions of time. This is a typical four dimensional spatiotemporal model, where each event has three spatial and one temporal dimension. Here we have concentrated on cortical computation over this model. Our model naturally indicates a duality between memory (both working and long term) and cognition, in the sense that they are mutually transformable in either direction. It also gives some support to the notion of grandmother cells.

When we look at the dynamics produced by biological neuronal populations, we are immediately struck by the fact that aperiodic, chaotic-like dynamics appear to be the normal operating state of such systems. Recent work has shown that such aperiodic dynamics, at least in perceptual systems, may not only be the result of random perturbations experienced by the system from external stimulation, but that the brain itself generates aperiodic dynamics to deal more flexibly and reliably with noisy environmental stimulation. Complex systems concepts are helping us to understand the properties of nonlinear systems that are fundamental for the emergence of complex spatiotemporal patterns in natural and biological systems. Advances in neuroscience and computational neurodynamics are applying these concepts of self-organization to understanding the spatiotemporal patterns observed in biological brains. In this article, we introduce a neural population model that is capable of replicating the generation of these types of aperiodic dynamics observed in biological brains. We use the model to self-organize cognitive maps in an autonomous agent through the agent's interaction with its environment. We show how such high-dimensional spatiotemporal dynamics may be shaped by environmental input and learning to form chaotic attractors that come to represent “meanings” for the agent. We discuss how the internal generation of such aperiodic dynamics may aid in the formation and recognition of such noisy environmental stimuli in biological organisms in general and in our simulated agents specifically. © 2006 Wiley Periodicals, Inc. Int J Int Syst 21: 955–971, 2006.

The concepts of meaning and intentionality play a central role in understanding animal and human intelligence, and they are crucial for creating artifacts with intelligent behavior and robust autonomy in real-life scenarios. The Intentional Dynamical Systems Symposium IDS'04 was held in Memphis, Tennessee, April 26–27, 2004, and it gave a forum to researchers working on the intensively developing interdisciplinary area of dynamical approaches to embedded cognition, based on self-organized development of autonomous systems. The present issue contains selected papers from the symposium addressing dynamical aspects of intentionality and intelligent behavior. All papers have been carefully peer reviewed to meet the standards of the journal. Contributions to this issue cover neuro-physiological foundations of human cortical dynamics, computational neuro-pharmacology and hippocampal modeling, development of KIII models for cognitive map generation, utilization of KII and KIII sets as logical gates and associative memories for odor recognition, evolutionary development of control for autonomous robot, and incorporating emotional representations in interactive robots. © 2006 Wiley Periodicals, Inc. Int J Int Syst 21: 875–879, 2006.

There has been intense research in feature binding to understand the parallel processing of features in visual information processing. The synchronization of spiking neurons is important for successful feature binding. In this work, we propose a novel approach to feature binding in spiking neurons using chaotic synchronization. We exploit each image pixel intensity value as individual neuron to generate chaotic time series. We generate the coupled map lattice series for neighborhood interaction and synchronization in spatiotemporal space. The largest cluster in the time series with similar chaotic synchronization parameter is used to generate segmented image. We obtain proof-of-concept application of our model in MR image clustering and compare our results with the existing Otsu adaptive segmentation technique.

We consider the model of interacting neural populations to be the main building block of K-sets, as suggested by W.J. Freeman. The full KM set's dynamics is understood through building the system up from the reduced KII. Theoretical condition for stability of the intermediate KII model is derived and the regions of structural stability of the full KII model are identified based on numeric data.

A biologically inspired dynamical neural network model called KIV is used in this work to design autonomous agents. The KIV set models the vertebrate limbic system. Previous studies indicated that KIV is able to provide a control algorithm for navigation and decision-making for autonomous mobile agents. In this work we use Hilbert transform to capture global synchronized spatio-temporal patterns of amplitude modulation in KIV. We identify phase transition in the simulated amygdala and show that it shares several important features of EEC signals.

Novel type of dynamical neural network, Freeman's K-models has been inspired by the biology and has been studied by researches in a number of research groups. This work is dedicated to providing a unifying test bed implementation that is capable of satisfying needs of possibly the most of these people. Given popularity of MATLABreg computational environment among engineers and scientists our effort is to develop a toolbox that would give way to simulate and experiment with the K-models in a simple manner as well as provide flexible tools for creation of applications that build on this dynamical model. The implementation is presented and detail of the design solutions are provided in this work. Also the successful usage of the toolbox is illustrated with the application to financial time series prediction by the K-models method

A biologically inspired approach to sensory fusion and decision-making in a network of interacting autonomous agents is outlined. The underlying biological model (KIV) explores the hierarchy of dynamically interacting units, i.e., sensory cortices. Multi-sensory percept formation in vertebrates is used for modeling multi-agent cooperation in robot networks. Each agent autonomously performs its task, e.g., classification and pattern recognition. The autonomous units weakly interact to produce a coherent, goal-oriented behavior at the level of the overall network. High-level decision-making is manifested through the sequence of intermittent phase transitions in the network coordination unit, which is modeled based on the operation of the entorhinal cortex

Mesoscopic level neurodynamics study the collective dynamical behavior of neural populations. Such models are becoming increasingly important in understanding large-scale brain processes. Brains exhibit aperiodic oscillations with a much more rich dynamical behavior than fixed-point and limit-cycle approximation allow. Here we present a discretized model inspired by Freeman's K-set mesoscopic level population model. We show that this version is capable of replicating the important principles of aperiodic/chaotic neurodynamics while being fast enough for use in real-time autonomous agent applications. This simplification of the K model provides many advantages not only in terms of efficiency but in simplicity and its ability to be analyzed in terms of its dynamical properties. We study the discrete version using a multilayer, highly recurrent model of the neural architecture of perceptual brain areas. We use this architecture to develop example action selection mechanisms in an autonomous agent.

Biological brains are saturated with complex dynamics. Artificial neural network models abstract much of this complexity away and represent the computational process of neuronal groups in terms of simple point, and sometimes periodic attractors. But is this abstraction justified? Aperiodic dynamics are known to be essential in the formation of perceptual mechanisms and representations in biological organisms. Advances in neuroscience and computational neurodynamics are helping us to understand the properties of nonlinear systems that are fundamental in the self-organization of stable, complex patterns for perceptual, memory and other cognitive mechanisms in biological brains. Much of this new understanding of the principles of self organization in biological brains has yet to be modeled or used to improve the performance of autonomous robotic and virtual agents. In this paper we present a model of an autonomous agent learning appetitive/aversive behaviors using a neuronal group model capable of such aperiodic dynamics. We demonstrate how such dynamics are useful in the self-organization of perception and behavior, and discuss the use of aperiodic dynamics in the self-organization of cognitive mechanisms in autonomous agents.

The aim of This work is to demonstrate that the dynamic KIV architecture can be used to integrate various sensory signals to achieve an efficient goal oriented navigation, when the robot has no a priori information about the simulated Martian environment. Navigation through space commonly involves goal-seeking and obstacle-avoidance. We show how a robot equipped with landmark detectors and eight infrared sensors can accomplish this task using a biologically inspired artificial brain. KIV demonstrates robust multisensory fusion with fast learning of goal-oriented behavior.

K2 sets are basic building blocks of dynamical neural network memories called K3. The K3s are strongly biologically motivated models of neural organization and functioning at the mesoscopic level in the cortex of vertebrate brains. The present study focuses on the fixed point and the limit cycle attractors in K2s. It considers the eigenvalues of the linearized K2 system and outlines the conditions under which it exhibits limit cycle oscillations. The derived conditions are instrumental in tuning the parameters of the K3 models having sustained chaotic oscillations.

This work aims at studying dynamical models of neural networks, which exhibit transitions between quasistable states of various complexities. We use the biologically motivated KIII model, which is a high-dimensional dynamical system with extremely fragmented boundaries between limit cycles, tori, fixed points, and chaotic attractors. We study the role of additive noise in the development of itinerant trajectories. Noise broadens the region of the dominance of chaotic attractors. This result is especially useful in the application of KIII and makes it possible to select parameter regions where KIII can operate as a robust dynamic system and associative memory device.

KIII is a strongly biologically inspired neural network model. It has a multi-layer architecture with excitatory and inhibitory neurons, which have massive lateral, feedforward, and delayed feedback connections between layers. KIII has been shown previously to be an efficient tool of classification and pattern recognition. In this work, we develop a methodology to use KIII for multi-step time series prediction. The method is applied for the IJCNN CATS benchmark data.

We use a biologically inspired dynamic neural network model to accomplish goal-oriented navigation by a mobile robot in a real environment with obstacles. This model is the KIV model of the brain. Real time navigation is a challenging task, especially when there is no a priori information about the environment. Our robot EMMA is designed to be autonomous using various sensory inputs, which are integrated to achieve an efficient navigation task. This paper focuses on the design, implementation, and evaluation of the performance of EMMA and gives a proof-of-principle in a real environment.

We use the biologically inspired dynamic neural network architecture KIV to achieve robust goal-oriented navigation in a physical environment with obstacles. KIV operates on the principle of chaotic neurodynamics, in the style of brains. It performs the task of multi-sensory fusion, recognition, and decision-making in real time. We use the Sony AIBO robot to demonstrate the operation of our algorithm. AIBO's video camera and infra sensors have been complemented with an external camera for monitoring of the robot's position. The performance of the autonomous system is evaluated using goal-oriented navigation.

We present a model of spatial navigation based on the non-convergent dynamics of brain activity. The system includes a hippocampal module that processes global spatial information and a cortical module that deals with local sensory information. We test the model using several spatial navigation paradigms: goal finding, shortcutting and detouring. Computer simulations show that the performance of the agent qualitatively matches that of animals and related models. This new approach provides a novel interpretation of how the brain accomplishes spatial navigation.

In this work, the KIV model is used for the description of the interaction between the sensory and cortical systems, the hippocampus, the amygdala, and the septum. Neural activity patterns in KIV determine the emergence of global spatial encoding to implement the orientation function of a simulated animal. Our results embody the mechanisms, which we believe support the generation of cognitive maps in the hippocampus, based on the sensory input-based destabilization of cortical spatio-temporal patterns. We illustrate learning results using the example of simulated navigation in a 2D environment.

In this paper, the KIII dynamic neural network is introduced and it is applied to the prediction of complex temporal sequences. In our approach, KIII gives a step-by-step prediction of the direction of the currency exchange rate change. Previously, various multiplayer perceptron (MLP) networks and recurrent neural networks have been successfully implemented for this application. Results obtained by KIII compare favorably with other methods.

Inspired by a neuronal architecture, we show how to produce dynamical behaviors in a special kind of probabilistic cellular neural network system. We demonstrate that the spatial and temporal behavior of neural activity undergoes sudden changes if the connection structure and noise component are varied. We characterize quantitatively phase transitions using the activation and cluster size. We indicate the potential role our present results may play in developing the theory of computation using non-convergent neurodynamic principles, called neurpercolation.

We present a model of spatial navigation based on the KIII dynamical model of perception developed by Walter Freeman in the 70's. We use a KIII model of the hippocampus that learns global orientation based on pre-defined landmarks and a KIII model of the sensory cortex that provides local sensory information about obstacles. We test the model using a task that requires the exploration of a previously unknown environment and the navigation towards a goal location. Computer simulations show that the simulated agent learns the position of the goal. The model provides a novel description of how navigation and way finding in the style of the brain.

This paper presents a novel approach inspired by the immune system that allows the application of conventional classification algorithms to perform anomaly detection. This approach appears to be very useful where only positive samples are available to train an anomaly detection system. The proposed approach uses the positive samples to generate negative samples that are used as training data for a classification algorithm. In particular, the algorithm produces fuzzy characterization of the normal (or abnormal) space. This allows it to assign a degree of normalcy, represented by membership value, to elements of the space.

We show that fuzzy neural nets can rate the emotional feedback message implicit in human facial expressions at a level comparable to human performance. We also identify primitive and derived features in facial expressions and animations. We also report on a partial solution to the difficult inverse problem of generating naturalistic facial expressions from summary abstract descriptions of the emotional feedback to be conveyed.

Explores bioinformatics-based modeling of immunological instabilities. We develop an algorithm for analyzing stability-instability properties of complex systems and use the developed technique to induce transitions in physical, biological and engineering systems. As a case study, we analyze the phenomena of tumor destabilization or spontaneous biological regression of a malignant focus-lymphocyte interactive system. Using stochastic noise analysis, we model high-dimensional collective oscillations of nonlinear elements and study nonautonomous systems with oscillation-induced phase transitions between lowand high-dimensional states. The associated nonlinear immunodynamical phenomenon of nonequilibrial destabilization of a malignant tumor is analyzed in terms of the Prigogine-Glansdorff (1971) stability theorem of dynamical systems theory.

We elaborate on information coding in chaotic neural networks. Noise plays a peculiar role in chaotic systems. We describe the constructive role of noise in stabilizing chaotic trajectories in Freeman s KIII model. KIII is a biologically plausible model of dynamic memories which has been established to interpret EEG measurements in the olfactory system. The results are illustrated on the example of encoding of noisy data in spatio-temporal aperiodic oscillatory patterns in neural networks.

Stability issues in Freeman's KIII model of the olfactory system are addressed. A possible mechanism for switching between various parts of the KIII attractor in response to external stimuli is outlined. It is shown that, in harmony with Haken's slaving principle, any particular system state (memory slot) can be described as a mixture of trajectories corresponding to a number of different attractors. Phase diagrams of the crucial KII sub-sets are constructed to illustrate the inherent multivaluedness (fuzziness) of the attractor landscape. In digital simulations, the overlap of the attractors can be conveniently embodied in a noisy representation, due to the fragmentation of the attractor basins and attractor crowding. Examples are introduced to illustrate the use of the observed phenomena to achieve the desired switching properties of the model.

This paper explores a multimodular architecture of an intelligent information system and proposes a method for adaptation. The method is based on evaluating which of the modules need to be adapted based on the performance of the whole system on new data. These modules are then trained selectively on the new data until they improve their performance and the performance of the whole system. The modules are fuzzy neural networks, especially designed to facilitate adaptive training and knowledge discovery, and spatial temporal maps. A particular case study of spoken language recognition is presented along with some preliminary experimental results of an adaptive speech recognition system. © 1998 John Wiley & Sons, Inc.

A connectionist-based time-series analysis method is described that includes chaotic characterization, fractal analysis together with statistical data processing in an adaptive fuzzy neural network environment. The applied fuzzy neural network (FuNN) can utilize as well as generate knowledge during an iterative learning and adaptation procedure. Two major aspects of the present work are (1) incorporating knowledge into the fuzzy neural network based on the nonlinear deterministic, chaotic analysis of the signals and (2) refining and updating the knowledge base by the FuNN using adaptive learning techniques. Examples include the standard gas-furnace benchmark data analysis and also an application to a case study of multivariate signal analysis as part of a project for establishing a plantwise monitoring and process control system. © 1998 John Wiley & Sons, Inc.