Grzegorz M. Wojcik

Letters: Liquid state machine and its separation ability as function of electrical parameters of cell

Large artificial Hodgkin-Huxley neural networks are examined. The structures discussed in this article simulate a part of the cortex of the mammalian visual system. We use a modular architecture of the cortex divided into sub-regions. Results of parallel simulations based on the liquid computing theory are presented in some detail. Separation ability of groups of neural microcircuits is observed. We check if such property depends on electrical parameters of particular cells. Properties Liquid State Machines are derived from the types of neurons used for simulations. They are compared and discussed to some extent.

Electrical parameters influence on the dynamics of the Hodgkin-Huxley liquid state machine

The model of mammalian cortical hypercolumn was simulated using liquid state machines built of simple Hodgkin-Huxley neurons. The influence of cell electrical parameters on the system dynamics was investigated. The systematic analysis of the hypercolumn separation ability in the function of time constants, cell membrane capacitance, resistance, Hodgkin-Huxley equilibrium potential and sodium and potassium channel conductances was performed. Optimal ranges of time constants for the most effective computational abilities of the model were estimated.

Which model to use for the Liquid State Machine

The properties of separation ability and computational efficiency of Liquid State Machines depend on the neural model employed and on the connection density in the liquid column. A simple model of part of mammalians visual system consisting of one hypercolumn was examined. Such a system was stimulated by two different input patterns, and the Euclidean distance, as well as the partial and global entropy of the liquid column responses were calculated. Interesting insights could be drawn regarding the properties of different neural models used in the liquid hypercolumn, and on the effect of connection density on the information representation capability of the system.

Liquid state machine built of Hodgkin–Huxley neurons and pattern recognition

Abstract Neural networks built of Hodgkin–Huxley neurons were examined. The structure and behavior of these nets was intended to be similar to liquid state machines. They could effectively process different input signals (i.e. geometrical patterns shown to artificial eye). The analysis of output responses was performed in two ways: by means of artificial neural network and by calculating informational entropy.

Self-organising criticality in the simulated models of the rat cortical microcircuits

Two and three dimensional models of rat barrel and somatosensory cortex were simulated. Hoddgkin-Huxley and Leaky-Integrate-and-Fire neurons were used to the construction of the networks in GENESIS and PCSIM environments. The dynamics of both models was analysed. Self-organising criticality phenomena were found. Profound investigations of this behaviour showed its dependence not only on the number of connections, but also on the simulated network architecture originating, e.g., from varying probability of exocytosis or synapse creation in the selected areas of the network. For two dimensional model the results were compared to that obtained for smaller models and analysis of this comparison is presented to some extent. The three dimensional model of the rat primary somatosensory cortex is based on the ensemble of Liquid State Machines. The results obtained from that model are in good agreement with the dynamics recorded in neurophysiological experiments on real brain.

Computational approach to understanding Autism Spectrum Disorders

Every year the prevalence of Autism Spectrum of Disorders (ASD) is rising. Is there a unifying mechanism of various ASD cases at the genetic, molecular, cellular or systems level? The hypothesis advanced in this paper is focused on neural dysfunctions that lead to problems with attention in autistic people. Simulations of attractor neural networks performing cognitive functions help to assess system long-term neurodynamics. The Fuzzy Symbolic Dynamics (FSD) technique is used for the visualization of attractors in the semantic layer of the neural model of reading. Large-scale simulations of brain structures characterized by a high order of complexity requires enormous computational power, especially if biologically motivated neuron models are used to investigate the influence of cellular structure dysfunctions on the network dynamics. Such simulations have to be implemented on computer clusters in a grid-based architectures

Analysis of the neural hypercolumn in parallel PCSIM simulations

Abstract Large and sudden changes in pitch or loudness occur statistically less frequently than gradual fluctuations, which means that natural sounds typically exhibit 1/f spectra. Experiments conducted on human subjects showed that listeners indeed prefer 1/f distributed melodies to melodies with faster or slower dynamics. It was recently demonstrated by using animal models, that neurons in primary auditory cortex of anesthetized ferrets exhibit a pronounced preference to stimuli that exhibit 1/f statistics. In the visual modality, it was shown that neurons in primary visual cortex of macaque monkeys exhibit tuning to sinusoidal gratings featuring 1/f dynamics. One might therefore suspect that neurons in mammalian cortex exhibit Self-Organizing Criticality. Indeed, we have found SOC-like phenomena in neurophysiological data collected in rat primary somatosensory cortex. In this paper we concentrated on investigation of the dynamics of cortical hypercolumn consisting of about 128 thousand simulated neurons. The set of 128 Liquid State Machines, each consisting 1024 neurons, was simulated on a simple cluster built of two double quad-core machines (16 cores). PCSIM was designed as a tool for simulating artificial biological-like neural networks composed of different models of neurons and different types of synapses. The simulator was written in C++ with a primary interface dedicated for the Python programming language. As its authors ensure it is intended to simulate networks containing up to millions of neurons and on the order of billions of synapses. This is achieved by distributing the network over different nodes of a computing cluster by using Message Passing Interface. The results obtained for Leaky Integrate-and-Fire model of neurons used for the construction of the hypercolumn and varying density of inter-column connections will be discussed. Benchmarking results for using the PCSIM on the cluster and predictions for grid computing will be presented to some extent. Research presented herein makes a good starting point for the simulations of very large parts of mammalian brain cortex and in some way leading to better understanding of the functionality of human brain.

Facial expression recognition based on Liquid State Machines built of alternative neuron models

This paper presents an approach to facial expression recognition based on the theory of liquid computing. Up to date, no emotion recognition systems based on spiking neural networks exist, and our work is the first attempt in this direction. We investigated the pattern recognition ability of Liquid State Machines based on various neural models, such as integrate-and-fire, resonate-and-fire, FitzHugh-Nagumo, Morris-Lecal, Hindmarsh-Rose and Izhikevichs models. No single Liquid State Machine provided particularly good results, but a global classifier we defined merging the response of the different models achieved a very satisfactory performance in expression recognition.

Self-organised criticality in a model of the rat somatosensory cortex

Large Hodgkin-Huxley (HH) neural networks were examined and the structures discussed in this article simulated a part of the rat somatosensory cortex. We used a modular architecture of the network divided into layers and sub-regions. Because of a high degree of complexity effective parallelisation of algorithms was required. The results of parallel simulations were presented. An occurrence of the self-organised criticality (SOC) was demonstrated. Most notably, in large biological neural networks consisting of artificial HH neurons, the SOC was shown to manifest itself in the frequency of its appearance as a function of the size of spike potential avalanches generated within such nets. These two parameters followed the power law characteristic of other systems exhibiting the SOC behaviour.

Shifting spatial attention-Numerical model of Posner experiment

A key challenge for neural modelling is to create models that allow to explain neural processes responsible for human behaviour. We propose computational spatial attention model that allows us to conduct computer experiments inspired by the Posner spatial attention task. The model consists of leaky integrate-and-fire neurons. The model dynamics was investigated and analysed. The obtained results showed that the proposed model is able to focus, keep and shift attention as well as to model lesions and dysfunctions.