Post and pre-compensatory Hebbian learning for categorisation

A system with some degree of biological plausibility is developed to categorise items from a widely used machine learning benchmark. The system uses fatiguing leaky integrate and fire neurons, a relatively coarse point model that roughly duplicates biological spiking properties; this allows spontaneous firing based on hypo-fatigue so that neurons not directly stimulated by the environment may be included in the circuit. A novel compensatory Hebbian learning algorithm is used that considers the total synaptic weight coming into a neuron. The network is unsupervised and entirely self-organising. This is relatively effective as a machine learning algorithm, categorising with just neurons, and the performance is comparable with a Kohonen map. However the learning algorithm is not stable, and behaviour decays as length of training increases. Variables including learning rate, inhibition and topology are explored leading to stable systems driven by the environment. The model is thus a reasonable next step toward a full neural memory model.     

On the role of anatomy in learning by the cerebellar cortex

The properties due to the location of neurons, synapses, and possibly even synaptic channels, in neuron networks are still unknown. Our preliminary results suggest that not only the interconnections but also the relative positions of the different elements in the network are of importance in the learning process in the cerebellar cortex. We have used neural field equations to investigate the mechanisms of learning in the hierarchical neural network. The numerical resolution of these equations reveals two important properties: (i) The hierarchical structure of this network has the expected effect on learning because the flow of information at the neuronal level is controlled by the heterosynaptic effect through the synaptic density-connectivity function, i.e. the action potential field variable is controlled by the synaptic efficacy field variable at different points of the neuron. (ii) The geometry of the system involves different velocities of propagation along different fibers, i.e. different delays between cells, and thus has a stabilizing effect on the dynamics, allowing the Purkinje output to reach a given value. The field model proposed should be useful in the study of the spatial properties of hierarchical biological systems.     

Speed/accuracy trade-off between the habitual and the goal-directed processes

Instrumental responses are hypothesized to be of two kinds: habitual and goal-directed, mediated by the sensorimotor and the associative cortico-basal ganglia circuits, respectively. The existence of the two heterogeneous associative learning mechanisms can be hypothesized to arise from the comparative advantages that they have at different stages of learning. In this paper, we assume that the goal-directed system is behaviourally flexible, but slow in choice selection. The habitual system, in contrast, is fast in responding, but inflexible in adapting its behavioural strategy to new conditions. Based on these assumptions and using the computational theory of reinforcement learning, we propose a normative model for arbitration between the two processes that makes an approximately optimal balance between search-time and accuracy in decision making. Behaviourally, the model can explain experimental evidence on behavioural sensitivity to outcome at the early stages of learning, but insensitivity at the later stages. It also explains that when two choices with equal incentive values are available concurrently, the behaviour remains outcome-sensitive, even after extensive training. Moreover, the model can explain choice reaction time variations during the course of learning, as well as the experimental observation that as the number of choices increases, the reaction time also increases. Neurobiologically, by assuming that phasic and tonic activities of midbrain dopamine neurons carry the reward prediction error and the average reward signals used by the model, respectively, the model predicts that whereas phasic dopamine indirectly affects behaviour through reinforcing stimulus-response associations, tonic dopamine can directly affect behaviour through manipulating the competition between the habitual and the goal-directed systems and thus, affect reaction time.     

Density dependent neurodynamics

The dynamics of a neural network depends on density parameters at (at least) two different levels: the subcellular density of ion channels in single neurons, and the density of cells and synapses at a network level. For the Frankenhaeuser-Huxley (FH) neural model, the density of sodium (Na) and potassium (K) channels determines the behaviour of a single neuron when exposed to an external stimulus. The features of the onset of single neuron oscillations vary qualitatively among different regions in the channel density plane. At a network level, the density of neurons is reflected in the global connectivity. We study the relation between the two density levels in a network of oscillatory FH neurons, by qualitatively distinguishing between three regions, where the mean network activity is (1) spiking, (2) oscillating with enveloped frequencies, and (3) bursting, respectively. We demonstrate that the global activity can be shifted between regions by changing either the density of ion channels at the subcellular level, or the connectivity at the network level, suggesting that different underlying mechanisms can explain similar global phenomena. Finally, we model a possible effect of anaesthesia by blocking specific inhibitory ion channels.     

Dynamic interactions between orexin and dynorphin may delay onset of functional orexin effects: a modeling study

Orexin (also known as hypocretin) neurons play a key role in regulating sleep-wake behavior, but the links between orexin neuron electrophysiology and function have not been explored. Orexin neurons are wake-active, and spiking activity in orexin neurons may anticipate transitions to wakefulness by several seconds. However, it is suggested that while the orexin system is necessary to maintain sustained wake bouts, orexin has little effect on brief wake bouts. In vitro experiments investigating the actions of orexin and dynorphin, a colocalized neuropeptide, on orexin neurons indicate that the dynamics of desensitization to dynorphin may represent a mechanism for modulating local network activity and resolving the apparent discrepancy between the onset of firing in orexin neurons and the onset of functional orexin effects. To investigate the role of dynorphin on orexin neuron activity, a Hodgkin-Huxley-type model orexin neuron was developed in which baseline electrophysiology, orexin/dynorphin action, and dynorphin desensitization were closely tied to experimental data. In this model framework, model orexin neuron responses to orexin/dynorphin action were calibrated by simulating the physiologic effects of static orexin and dynorphin bath application on orexin neurons. Then behavior in a small network of model orexin neurons was simulated with pure orexin, pure dynorphin, or combined orexin and dynorphin coupling based on the mechanisms established in the static case. It was found that the dynamics of desensitization to dynorphin can mediate a clear shift from a network in which firing is suppressed by dynorphin-mediated inhibition to a network of neurons with high firing rates sustained by orexin-mediated excitation. The findings suggest that dynamic interactions between orexin and dynorphin at transitions from sleep to wake may delay onset of functional orexin effects.     

Spatial firing patterns of auditory neuron network modelling by computer simulation

This communication examines, in digital computer simulated network, input signals and response patterns established at excitatory neurons' level i.e. the membrane potential of neuron soma. It is restricted to spatial patterns of the auditory neuron networks and time factor for nervous conduction and transmission is neglected compared with long maintained membrane potentials of neuron somas. The model analyzes the change in the spatial patterns of the membrane potential in the two dimensional networks of the auditory system. In order to evaluate the contribution of the various parameters, it is started that the simplest model has only one parameter, lateral inhibition. The other parameters are then added, one at a time, to successive models. The lateral inhibition is a necessary condition in the auditory nervous system if any sharpening of the response areas in the single neurons is to occur. A necessary condition for the validity of the model is that it should be applicable to the other senses such as vision and chemical patterns, taste. The threshold feature of auditory neurons aids in producing a sharpening in the neuron of the auditory relay nuclei. It does this clipping the spatial response patterns in one dimensional arrays of excitatory neurons. Recurrent inhibition seems a necessary condition in the sensory nervous system that any kinds of input signals are to be preserved over a wide range of stimulus intensity. In other words, this network has a wide dynamic range against any kinds of input signals. A simple self-recurrent negative feedback does not contribute to the sharpening, but more complex socalled averaged type does. A neuron network is capable of responding stably to stimuli with a wide range of intensity and with any kind of spatial patterns if there is a simple negative feedback mechanism. When there is no negative feedback, input signals soon disappear or saturate in the neuron network. Therefore, recurrent inhibition is the most important mechanism. Spontaneous activity appears to aid in the sharpening by providing a kind of contrast, that is by reducting the amount of activity in neurons adjacent to the excitatory area. Moreover, the effect of spontaneous activity in the model seems to make repples around the excitatory area and suggests that an introduction of activity at any stage of the networks, from whatever source for example reticulum formation and thalamus, might appreciably alter the response patterns at subsequent neuron network. This suggests that the mechanism of the consciousness that might be controlled by the thalamus and or reticular formation. These two dimensional neuron networks may be expanded to three dimensional neuron networks. The former might simulate the auditory nervous system while the latter might simulate the visual system.     

Towards an artificial brain

Three components of a brain model operating on neuromolecular computing principles are described. The first component comprises neurons whose input-output behavior is controlled by significant internal dynamics. Models of discrete enzymatic neurons, reaction-diffusion neurons operating on the basis of the cyclic nucleotide cascade, and neurons controlled by cytoskeletal dynamics are described. The second component of the model is an evolutionary learning algorithm which is used to mold the behavior of enzyme-driven neurons or small networks of these neurons for specific function, usually pattern recognition or target seeking tasks. The evolutionary learning algorithm may be interpreted either as representing the mechanism of variation and natural selection acting on a phylogenetic time scale, or as a conceivable ontogenetic adaptation mechanism. The third component of the model is a memory manipulation scheme, called the reference neuron scheme. In principle it is capable of orchestrating a repertoire of enzyme-driven neurons for coherent function. The existing implementations, however, utilize simple neurons without internal dynamics. Spatial navigation and simple game playing (using tic-tac-toe) provide the task environments that have been used to study the properties of the reference neuron model. A memory-based evolutionary learning algorithm has been developed that can assign credit to the individual neurons in a network. It has been run on standard benchmark tasks, and appears to be quite effective both for conventional neural nets and for networks of discrete enzymatic neurons. The models have the character of artificial worlds in that they map the hierarchy of processes in the brain (at the molecular, neuronal, and network levels), provide a task environment, and use this relatively self-contained setup to develop and evaluate learning and adaptation algorithms.     

A Boolean complete neural model of adaptive behavior

A multi-layered neural assembly is developed which has the capability of learning arbitrary Boolean functions. Though the model neuron is more powerful than those previously considered, assemblies of neurons are needed to detect non-linearly separable patterns. Algorithms for learning at the neuron and assembly level are described. The model permits multiple output systems to share a common memory. Learned evaluation allows sequences of actions to be organized. Computer simulations demonstrate the capabilities of the model.     

Fitting of dynamic recurrent neural network models to sensory stimulus-response data

We present a theoretical study aiming at model fitting for sensory neurons. Conventional neural network training approaches are not applicable to this problem due to lack of continuous data. Although the stimulus can be considered as a smooth time-dependent variable, the associated response will be a set of neural spike timings (roughly the instants of successive action potential peaks) that have no amplitude information. A recurrent neural network model can be fitted to such a stimulus-response data pair by using the maximum likelihood estimation method where the likelihood function is derived from Poisson statistics of neural spiking. The universal approximation feature of the recurrent dynamical neuron network models allows us to describe excitatory-inhibitory characteristics of an actual sensory neural network with any desired number of neurons. The stimulus data are generated by a phased cosine Fourier series having a fixed amplitude and frequency but a randomly shot phase. Various values of amplitude, stimulus component size, and sample size are applied in order to examine the effect of the stimulus to the identification process. Results are presented in tabular and graphical forms at the end of this text. In addition, to demonstrate the success of this research, a study involving the same model, nominal parameters and stimulus structure, and another study that works on different models are compared to that of this research.     

How Good are Formal Neurons for Modelling Real Ones?

A formal neuron has been studied mathematically. The spiking behaviour of a single neuron has been considered and the influence of the other neurons has been replaced by an average activity level. Four different kinds of spiking behaviour are predicted by the model: B (bursts), C (continuous), P (periodic) and S (silent) neurons and several real neurons can be classified within these four categories. Some properties of the spiking neuron are calculated: 1) the time between spikes, 2) the spike train length and 3) the silent time. Because these magnitudes can be measured in the laboratory, an experimental validation of the model is proposed.     

Computer modeling of mechanisms of the information processing in the olfactory bulb. I. Model of the structure-functional organizations of neuronal elements in the olfactory bulb and receptor epithelium

A computer model of the olfactory bulb was constructed. The paper describes: 1) the general architecture of a model neuron network that reflects the neurophysiological experimental and theoretical data on the structural and functional organization of the peripheral part of the olfactory system, the olfactory bulb with inputs from olfactory receptor neurons; 2) the organization of each of three levels of the model: receptors, olfactory glomeruli, and basic neurons; and 3) a scenario of the computer model work. In some aspects, in particular, in the principle of information presentation, the treatment of the role of basic neurons (mitral and tufted cells), and their interrelations in modules, the model favorably differs from the available olfactory bulb models. The model is basic and provides further refinement of the architecture, an increase in the number of modules, and the modeling of the learning process.     

Learning compositional sequences with multiple time scales through a hierarchical network of spiking neurons

Sequential behaviour is often compositional and organised across multiple time scales: a set of individual elements developing on short time scales (motifs) are combined to form longer functional sequences (syntax). Such organisation leads to a natural hierarchy that can be used advantageously for learning, since the motifs and the syntax can be acquired independently. Despite mounting experimental evidence for hierarchical structures in neuroscience, models for temporal learning based on neuronal networks have mostly focused on serial methods. Here, we introduce a network model of spiking neurons with a hierarchical organisation aimed at sequence learning on multiple time scales. Using biophysically motivated neuron dynamics and local plasticity rules, the model can learn motifs and syntax independently. Furthermore, the model can relearn sequences efficiently and store multiple sequences. Compared to serial learning, the hierarchical model displays faster learning, more flexible relearning, increased capacity, and higher robustness to perturbations. The hierarchical model redistributes the variability: it achieves high motif fidelity at the cost of higher variability in the between-motif timings.     

Sleep changes during degeneration of neurons in the substantia nigra induced by inhibitor of proteasomes lactacystin in rats

A decrease in activity of ubiquitin proteasome system results in accumulation of toxic forms of protein and cell degeneration, including dopamine (DA)-ergic neurons in the substantia nigra; these neurons are remarkable for their low proteolytic activity of proteosomes that makes them more vulnerable, especially when subjected to the neurotoxin action or Parkinson's disease (PD). The goal of the present study is to develop a model on the basis of inhibition of proteasome activity of nigral cell degeneration which is not accompanied by disturbances in motor behavior but leads to changes in sleep-wake cycle characteristic of the non-motor behaviour. We determined the optimal dose of natural inhibitor of proteasome lactacystin (0.4 mkg) and developed a preclinical model of PD in Wistar rats. We established that on the 14th day following lactacystin double (with one-week interval) bilateral injection into the substantia nigra the developing effects involved 28 % degeneration of DA-ergic neurons in the compact part of the substantia nigra, absence of disorders in motor behaviour, and increase in the total time of rapid eye movement sleep by 37 % at the second half of inactive day phase. These data and an increase in the level of key enzyme of DA synthesis tyrosine hydroxylase (TH) in survived neurons in the substantia nigra as well as the presence of the inverse correlation dependency (r = -0.8, p < 0.01) between the number of survived neurons and the level of TH inside them suggest a hypothesis that the increase in the duration of rapid eye movement sleep could be a non-motor marker of the preclinical stage of PD reflecting a reservation of compensatory potentials in the nigrostriatal system.     

Leader neurons in leaky integrate and fire neural network simulations

In this paper, we highlight the topological properties of leader neurons whose existence is an experimental fact. Several experimental studies show the existence of leader neurons in population bursts of activity in 2D living neural networks (Eytan and Marom, J Neurosci 26(33):8465–8476, 2006; Eckmann et al., New J Phys 10(015011), 2008). A leader neuron is defined as a neuron which fires at the beginning of a burst (respectively network spike) more often than we expect by chance considering its mean firing rate. This means that leader neurons have some burst triggering power beyond a chance-level statistical effect. In this study, we characterize these leader neuron properties. This naturally leads us to simulate neural 2D networks. To build our simulations, we choose the leaky integrate and fire (lIF) neuron model (Gerstner and Kistler 2002; Cessac, J Math Biol 56(3):311–345, 2008), which allows fast simulations (Izhikevich, IEEE Trans Neural Netw 15(5):1063–1070, 2004; Gerstner and Naud, Science 326:379–380, 2009). The dynamics of our lIF model has got stable leader neurons in the burst population that we simulate. These leader neurons are excitatory neurons and have a low membrane potential firing threshold. Except for these two first properties, the conditions required for a neuron to be a leader neuron are difficult to identify and seem to depend on several parameters involved in the simulations themselves. However, a detailed linear analysis shows a trend of the properties required for a neuron to be a leader neuron. Our main finding is: A leader neuron sends signals to many excitatory neurons as well as to few inhibitory neurons and a leader neuron receives only signals from few other excitatory neurons. Our linear analysis exhibits five essential properties of leader neurons each with different relative importance. This means that considering a given neural network with a fixed mean number of connections per neuron, our analysis gives us a way of predicting which neuron is a good leader neuron and which is not. Our prediction formula correctly assesses leadership for at least ninety percent of neurons.     

An adaptive visual neuronal model implementing competitive, temporally asymmetric Hebbian learning

A novel depth-from-motion vision model based on leaky integrate-and-fire (I&F) neurons incorporates the implications of recent neurophysiological findings into an algorithm for object discovery and depth analysis. Pulse-coupled I&F neurons capture the edges in an optical flow field and the associated time of travel of those edges is encoded as the neuron parameters, mainly the time constant of the membrane potential and synaptic weight. Correlations between spikes and their timing thus code depth in the visual field. Neurons have multiple output synapses connecting to neighbouring neurons with an initial Gaussian weight distribution. A temporally asymmetric learning rule is used to adapt the synaptic weights online, during which competitive behaviour emerges between the different input synapses of a neuron. It is shown that the competition mechanism can further improve the model performance. After training, the weights of synapses sourced from a neuron do not display a Gaussian distribution, having adapted to encode features of the scenes to which they have been exposed.     

A multivariate population density model of the dLGN/PGN relay

Using a population density approach we study the dynamics of two interacting collections of integrate-and-fire-or-burst (IFB) neurons representing thalamocortical (TC) cells from the dorsal lateral geniculate nucleus (dLGN) and thalamic reticular (RE) cells from the perigeniculate nucleus (PGN). Each population of neurons is described by a multivariate probability density function that satisfies a conservation equation with appropriately defined probability fluxes and boundary conditions. The state variables of each neuron are the membrane potential and the inactivation gating variable of the low-threshold Ca²⁺ current I_T. The synaptic coupling of the populations and external excitatory drive are modeled by instantaneous jumps in the membrane potential of postsynaptic neurons. The population density model is validated by comparing its response to time-varying retinal input to Monte Carlo simulations of the corresponding IFB network composed of 100 to 1000 cells per population. In the absence of retinal input, the population density model exhibits rhythmic bursting similar to the 7 to 14 Hz oscillations associated with slow wave sleep that require feedback inhibition from RE to TC cells. When the TC and RE cell potassium leakage conductances are adjusted to represent cholingergic neuromodulation and arousal of the network, rhythmic bursting of the probability density model may either persists or be eliminated depending on the number of excitatory (TC to RE) or inhibitory (RE to TC) connections made by each presynaptic cell. When the probability density model is stimulated with constant retinal input (10–100 spikes/sec), a wide range of responses are observed depending on cellular parameters and network connectivity. These include asynchronous burst and tonic spikes, sleep spindle-like rhythmic bursting, and oscillations in population firing rate that are distinguishable from sleep spindles due to their amplitude, frequency, or the presence of tonic spikes. In this context of dLGN/PGN network modeling, we find the population density approach using 2,500 mesh points and resolving membrane voltage to 0.7 mV is over 30 times more efficient than 1000-cell Monte Carlo simulations. Action Editor: David Golomb     

Context codes and the effect of noisy learning on a simplified hippocampal CA3 model

This paper investigates how noise affects a minimal computational model of the hippocampus and, in particular, region CA3. The architecture and physiology employed are consistent with the known anatomy and physiology of this region. Here, we use computer simulations to demonstrate and quantify the ability of this model to create context codes in sequential learning problems. These context codes are mediated by local context neurons which are analogous to hippocampal place-coding cells. These local context neurons endow the network with many of its problem-solving abilities. Our results show that the network encodes context on its own and then uses context to solve sequence prediction under ambiguous conditions. Noise during learning affects performance, and it also affects the development of context codes. The relationship between noise and performance in a sequence prediction is simple and corresponds to a disruption of local context neuron firing. As noise exceeds the signal, sequence completion and local context neuron firing are both lost. For the parameters investigated, extra learning trials and slower learning rates do not overcome either of the effects of noise. The results are consistent with the important role played, in this hippocampal model, by local context neurons in sequence prediction and for disambiguation across time.     

PhotoMEA: an opto-electronic biosensor for monitoring in vitro neuronal network activity

PhotoMEA is a biosensor useful for the analysis of an in vitro neuronal network, fully based on optical methods. Its function is based on the stimulation of neurons with caged glutamate and the recording of neuronal activity by Voltage-Sensitive fluorescent Dyes (VSD). The main advantage is that it will be possible to stimulate even at sub-single neuron level and to record with high resolution the activity of the entire network in the culture. A large-scale view of neuronal intercommunications offers a unique opportunity for testing the ability of drugs to affect neuronal properties as well as alterations in the behaviour of the entire network. The concept and a prototype for validation is described here in detail.     

An extended model for a spiking neuron class

This paper proposes an extension to the model of a spiking neuron for information processing in artificial neural networks, developing a new approach for the dynamic threshold of the integrate-and-fire neuron. This new approach invokes characteristics of biological neurons such as the behavior of chemical synapses and the receptor field. We demonstrate how such a digital model of spiking neurons can solve complex nonlinear classification with a single neuron, performing experiments for the classical XOR problem. Compared with rate-coded networks and the classical integrate-and-fire model, the trained network demonstrated faster information processing, requiring fewer neurons and shorter learning periods. The extended model validates all the logic functions of biological neurons when such functions are necessary for the proper flow of binary codes through a neural network.     

Shaping Neuronal Network Activity by Presynaptic Mechanisms

Neuronal microcircuits generate oscillatory activity, which has been linked to basic functions such as sleep, learning and sensorimotor gating. Although synaptic release processes are well known for their ability to shape the interaction between neurons in microcircuits, most computational models do not simulate the synaptic transmission process directly and hence cannot explain how changes in synaptic parameters alter neuronal network activity. In this paper, we present a novel neuronal network model that incorporates presynaptic release mechanisms, such as vesicle pool dynamics and calcium-dependent release probability, to model the spontaneous activity of neuronal networks. The model, which is based on modified leaky integrate-and-fire neurons, generates spontaneous network activity patterns, which are similar to experimental data and robust under changes in the model''s primary gain parameters such as excitatory postsynaptic potential and connectivity ratio. Furthermore, it reliably recreates experimental findings and provides mechanistic explanations for data obtained from microelectrode array recordings, such as network burst termination and the effects of pharmacological and genetic manipulations. The model demonstrates how elevated asynchronous release, but not spontaneous release, synchronizes neuronal network activity and reveals that asynchronous release enhances utilization of the recycling vesicle pool to induce the network effect. The model further predicts a positive correlation between vesicle priming at the single-neuron level and burst frequency at the network level; this prediction is supported by experimental findings. Thus, the model is utilized to reveal how synaptic release processes at the neuronal level govern activity patterns and synchronization at the network level.