Towards neuro-inspired symbolic models of cognition: linking neural dynamics to behaviors through asynchronous communications

Grounding Mental Representations in a Virtual Multi-Level Functional Framework

According to the associative theory of learning, reactive behaviors described by stimulus-response pairs result in the progressive wiring of a plastic brain. In contrast, flexible behaviors are supposedly driven by neurologically grounded mental states that involve computations on informational contents. These theories appear complementary, but are generally opposed to each other. The former is favored by neuro-scientists who explore the low-level biological processes supporting cognition, and the later by cognitive psychologists who look for higher-level structures. This situation can be clarified through an analysis that independently defines abstract neurological and informational functionalities, and then relate them through a virtual interface. This framework is validated through a modeling of the first stage of Piaget's cognitive development theory, whose reported end experiments demonstrate the emergence of mental representations of object displacements. The neural correlates grounding this emergence are given in the isomorphic format of an associative memory. As a child's exploration of the world progresses, his mental models will eventually include representations of space, time and causality. Only then epistemological concepts, such as beliefs, will give rise to higher level mental representations in a possibly richer propositional format. This raises the question of which additional neurological functionalities, if any, would be required in order to include these extensions into a comprehensive grounded model. We relay previously expressed views, which in summary hypothesize that the ability to learn has evolved from associative reflexes and memories, to suggest that the functionality of associative memories could well provide the sufficient means for grounding cognitive capacities.

Modeling the Synchronization of Multimodal Perceptions as a Basis for the Emergence of Deterministic Behaviors

Living organisms have either innate or acquired mechanisms for reacting to percepts with an appropriate behavior e.g., by escaping from the source of a perception detected as threat, or conversely by approaching a target perceived as potential food. In the case of artifacts, such capabilities must be built in through either wired connections or software. The problem addressed here is to define a neural basis for such behaviors to be possibly learned by bio-inspired artifacts. Toward this end, a thought experiment involving an autonomous vehicle is first simulated as a random search. The stochastic decision tree that drives this behavior is then transformed into a plastic neuronal circuit. This leads the vehicle to adopt a deterministic behavior by learning and applying a causality rule just as a conscious human driver would do. From there, a principle of using synchronized multimodal perceptions in association with the Hebb principle of wiring together neuronal cells is induced. This overall framework is implemented as a virtual machine i.e., a concept widely used in software engineering. It is argued that such an interface situated at a meso-scale level between abstracted micro-circuits representing synaptic plasticity, on one hand, and that of the emergence of behaviors, on the other, allows for a strict delineation of successive levels of complexity. More specifically, isolating levels allows for simulating yet unknown processes of cognition independently of their underlying neurological grounding.

Deterministic characteristics of spontaneous activity detected by multi-fractal analysis in a spiking neural network with long-tailed distributions of synaptic weights

Cortical neural networks maintain autonomous electrical activity called spontaneous activity that represents the brain's dynamic internal state even in the absence of sensory stimuli. The spatio-temporal complexity of spontaneous activity is strongly related to perceptual, learning, and cognitive brain functions; multi-fractal analysis can be utilized to evaluate the complexity of spontaneous activity. Recent studies have shown that the deterministic dynamic behavior of spontaneous activity especially reflects the topological neural network characteristics and changes of neural network structures. However, it remains unclear whether multi-fractal analysis, recently widely utilized for neural activity, is effective for detecting the complexity of the deterministic dynamic process. To verify this point, we focused on the log-normal distribution of excitatory postsynaptic potentials (EPSPs) to evaluate the multi-fractality of spontaneous activity in a spiking neural network with a log-normal distribution of EPSPs. We found that the spiking activities exhibited multi-fractal characteristics. Moreover, to investigate the presence of a deterministic process in the spiking activity, we conducted a surrogate data analysis against the time-series of spiking activity. The results showed that the spontaneous spiking activity included the deterministic dynamic behavior. Overall, the combination of multi-fractal analysis and surrogate data analysis can detect deterministic complex neural activity. The multi-fractal analysis of neural activity used in this study could be widely utilized for brain modeling and evaluation methods for signals obtained by neuroimaging modalities.

Detecting prostate cancer using deep learning convolution neural network with transfer learning approach

Prostate Cancer in men has become one of the most diagnosed cancer and also one of the leading causes of death in United States of America. Radiologists cannot detect prostate cancer properly because of complexity in masses. In recent past, many prostate cancer detection techniques were developed but these could not diagnose cancer efficiently. In this research work, robust deep learning convolutional neural network (CNN) is employed, using transfer learning approach. Results are compared with various machine learning strategies (Decision Tree, SVM different kernels, Bayes). Cancer MRI database are used to train GoogleNet model and to train Machine Learning classifiers, various features such as Morphological, Entropy based, Texture, SIFT (Scale Invariant Feature Transform), and Elliptic Fourier Descriptors are extracted. For the purpose of performance evaluation, various performance measures such as specificity, sensitivity, Positive predictive value, negative predictive value, false positive rate and receive operating curve are calculated. The maximum performance was found with CNN model (GoogleNet), using Transfer learning approach. We have obtained reasonably good results with various Machine Learning Classifiers such as Decision Tree, Support Vector Machine RBF kernel and Bayes, however outstanding results were obtained by using deep learning technique.

Primal-size neural circuits in meta-periodic interaction

Experimental observations of simultaneous activity in large cortical areas have seemed to justify a large network approach in early studies of neural information codes and memory capacity. This approach has overlooked, however, the segregated nature of cortical structure and functionality. Employing graph-theoretic results, we show that, given the estimated number of neurons in the human brain, there are only a few primal sizes that can be attributed to neural circuits under probabilistically sparse connectivity. The significance of this finding is that neural circuits of relatively small primal sizes in cyclic interaction, implied by inhibitory interneuron potentiation and excitatory inter-circuit potentiation, generate relatively long non-repetitious sequences of asynchronous primal-length periods. The meta-periodic nature of such circuit interaction translates into meta-periodic firing-rate dynamics, representing cortical information. It is finally shown that interacting neural circuits of primal sizes 7 or less exhaust most of the capacity of the human brain, with relatively little room to spare for circuits of larger primal sizes. This also appears to ratify experimental findings on the human working memory capacity.

Probabilistically segregated neural circuits and subcritical linguistics

Early studies of cortical information codes and memory capacity have assumed large neural networks, which, subject to evenly probable binary (on/off) activity, were found to be endowed with large storage and retrieval capacities under the Hebbian paradigm. Here, we show that such networks are plagued with exceedingly high cross-network connectivity, yielding long code words, which are linguistically non-realistic and difficult to memorize and comprehend. Noting that the neural circuit activity code is jointly governed by somatic and synaptic activity states, termed neural circuit polarities, we show that, subject to subcritical polarity probability, random-graph-theoretic considerations imply small neural circuit segregation. Such circuits are shown to represent linguistically plausible cortical code words which, in turn, facilitate storage and retrieval of both circuit connectivity and firing-rate dynamics.

An adaptive decoder design based on the receding horizon optimization in BMI system

In a motor brain-machine interface system, since the electroencephalogram signal is changing through out the process of the arm movement, the offline trained decoder with fixed weights is often unable to convert the electroencephalogram signal accurately, resulting in poor recovery of joint motor function. In this paper, a receding horizon optimization strategy is chosen to online update the decoder weights and design an adaptive Wiener-filter-based decoder. Firstly, a classical Wiener-filter-based decoder with fixed weights is brief reviewed. Secondly, the weights in Wiener-filter-based decoder are updated by minimizing the cost function, which is composed by the sum of squared position errors in the given horizon at each sampling time. The simulation shows that the recovery effect of joint motor function and neuron activity in the BMI system with the adaptive decoder are both better than that in the BMI system with the fixed decoder.

Complex temporal patterns processing by a neural mass model of a cortical column

It is well known that neuronal networks are capable of transmitting complex spatiotemporal information in the form of precise sequences of neuronal discharges characterized by recurrent patterns. At the same time, the synchronized activity of large ensembles produces local field potentials that propagate through highly dynamic oscillatory waves, such that, at the whole brain scale, complex spatiotemporal dynamics of electroencephalographic (EEG) signals may be associated to sensorimotor decision making processes. Despite these experimental evidences, the link between highly temporally organized input patterns and EEG waves has not been studied in detail. Here, we use a neural mass model to investigate to what extent precise temporal information, carried by deterministic nonlinear attractor mappings, is filtered and transformed into fluctuations in phase, frequency and amplitude of oscillatory brain activity. The phase shift that we observe, when we drive the neural mass model with specific chaotic inputs, shows that the local field potential amplitude peak appears in less than one full cycle, thus allowing traveling waves to encode temporal information. After converting phase and amplitude changes obtained into point processes, we quantify input-output similarity following a threshold-filtering algorithm onto the amplitude wave peaks. Our analysis shows that the neural mass model has the capacity for gating the input signal and propagate selected temporal features of that signal. Finally, we discuss the effect of local excitatory/inhibitory balance on these results and how excitability in cortical columns, controlled by neuromodulatory innervation of the cerebral cortex, may contribute to set a fine tuning and gating of the information fed to the cortex.

Points and lines inside human brains

Starting from the tenets of human imagination, i.e., the concepts of lines, points and infinity, we provide a biological demonstration that the skeptical claim "human beings cannot attain knowledge of the world" holds true. We show that the Euclidean account of the point as "that of which there is no part" is just a conceptual device produced by our brain, untenable in our physical/biological realm: currently used terms like "lines, surfaces and volumes" label non-existent, arbitrary properties. We elucidate the psychological and neuroscientific features hardwired in our brain that lead us humans to think to points and lines as truly occurring in our environment. Therefore, our current scientific descriptions of objects' shapes, graphs and biological trajectories in phase spaces need to be revisited, leading to a proper portrayal of the real world's events: miniscule bounded physical surface regions stand for the basic objects in a traversal of spacetime, instead of the usual Euclidean points. Our account makes it possible to erase of a painstaking problem that causes many theories to break down and/or being incapable of describing extreme events: the unwanted occurrence of infinite values in equations. We propose a novel approach, based on point-free geometrical standpoints, that banishes infinitesimals, leads to a tenable physical/biological geometry compatible with human reasoning and provides a region-based topological account of the power laws endowed in nervous activities. We conclude that points, lines, volumes and infinity do not describe the world, rather they are fictions introduced by ancient surveyors of land surfaces.

Symbolic Modeling of Asynchronous Neural Dynamics Reveals Potential Synchronous Roots for the Emergence of Awareness

A new computational framework implementing asynchronous neural dynamics is used to address the duality between synchronous vs. asynchronous processes, and their possible relation to conscious vs. unconscious behaviors. Extending previous results on modeling the first three levels of animal awareness, this formalism is used here to produce the execution traces of parallel threads that implement these models. Running simulations demonstrate how sensory stimuli associated with a population of excitatory neurons inhibit in turn other neural assemblies i.e., a kind of neuronal asynchronous wiring/unwiring process that is reflected in the progressive trimming of execution traces. Whereas, reactive behaviors relying on configural learning produce vanishing traces, the learning of a rule and its later application produce persistent traces revealing potential synchronous roots of animal awareness. In contrast, to previous formalisms that use analytical and/or statistical methods to search for patterns existing in a brain, this new framework proposes a tool for studying the emergence of brain structures that might be associated with higher level cognitive capabilities.

A feature-based neurocomputational model of semantic memory

According with a featural organization of semantic memory, this work is aimed at investigating, through an attractor network, the role of different kinds of features in the representation of concepts, both in normal and neurodegenerative conditions. We implemented new synaptic learning rules in order to take into account the role of partially shared features and of distinctive features with different saliency. The model includes semantic and lexical layers, coding, respectively for object features and word-forms. Connections among nodes are strongly asymmetrical. To account for the feature saliency, asymmetrical synapses were created using Hebbian rules of potentiation and depotentiation, setting different pre-synaptic and post-synaptic thresholds. A variable post-synaptic threshold, which automatically changed to reflect the feature frequency in different concepts (i.e., how many concepts share a feature), was used to account for partially shared features. The trained network solved naming tasks and word recognition tasks very well, exploiting the different role of salient versus marginal features in concept identification. In the case of damage, superordinate concepts were preserved better than the subordinate ones. Interestingly, the degradation of salient features, but not of marginal ones, prevented object identification. The model suggests that Hebbian rules, with adjustable post-synaptic thresholds, can provide a reliable semantic representation of objects exploiting the statistics of input features.

Personality development in psychotherapy: a synergetic model of state-trait dynamics

Theoretical models of psychotherapy not only try to predict outcome but also intend to explain patterns of change. Studies showed that psychotherapeutic change processes are characterized by nonlinearity, complexity, and discontinuous transitions. By this, theoretical models of psychotherapy should be able to reproduce these dynamic features. Using time series derived from daily measures through internet-based real-time monitoring as empirical reference, we earlier presented a model of psychotherapy which includes five state variables and four trait variables. In mathematical terms, the traits modulate the shape of the functions which define the nonlinear interactions between the variables (states) of the model. The functions are integrated into five coupled nonlinear difference equations. In the present paper, we model how traits (dispositions or competencies of a person) can continuously be altered by new experiences and states (cognition, emotion, behavior). Adding equations that link states to traits, this model not only describes how therapeutic interventions modulate short-term change and fluctuations of psychological states, but also how these can influence traits. Speaking in terms of Synergetics (theory of self-organization in complex systems), the states correspond to the order parameters and the traits to the control parameters of the system. In terms of psychology, trait dynamics is driven by the states-i.e., by the concrete experiences of a client-and creates a process of personality development at a slower time scale than that of the state dynamics (separation of time scales between control and order parameter dynamics).