Modelling studies on the computational function of fast temporal structure in cortical circuit activity

The interplay between modelling and experimental studies can support the exploration of the function of neuronal circuits in the cortex. We exemplify such an approach with a study on the role of spike timing and gamma-oscillations in associative memory in strongly connected circuits of cortical neurones. It is demonstrated how associative memory studies on different levels of abstraction can specify the functionality to be expected in real cortical neuronal circuits. In our model overlapping random configurations of sparse cell populations correspond to memory items that are stored by simple Hebbian coincidence learning. This associative memory task will be implemented with biophysically well tested compartmental neurones developed by Pinsky and Rinzel . We ran simulation experiments to study memory recall in two network architectures: one interconnected pool of cells, and two reciprocally connected pools. When recalling a memory by stimulating a spatially overlapping set of cells, the completed pattern is coded by an event of synchronized single spikes occurring after 25-60 ms. These fast associations are performed even at a memory load corresponding to the memory capacity of optimally tuned formal associative networks (>0.1 bit/synapse). With tonic stimulation or feedback loops in the network the neurones fire periodically in the gamma-frequency range (20-80 Hz). With fast changing inputs memory recall can be switched between items within a single gamma cycle. Thus, oscillation is not a primary coding feature necessary for associative memory. However, it accompanies reverberatory feedback providing an improved iterative memory recall completed after a few gamma cycles (60-260 ms). In the bidirectional architecture reverberations do not express in a rigid phase locking between the pools. For small stimulation sets bursting occurred in these cells acting as a supportive mechanism for associative memory.     

Recollection ability of three-dimensional correlation matrix associative memory

This paper presents the approximate expression of the recollection probability of a three-dimensional correlation matrix autoassociative memory, in which each memorized pattern consists of binary (+1 or-1) elements, and discusses the recollection ability in comparison with that of the conventional two-dimensional correlation matrix associative memory. An associative memory using the correlation properties between memorized patterns desires the condition that the memorized patterns are mutually orthogonal or approximately orthogonal. The three-dimensional correlation matrix associative memory that inscribes the third order correlation of a memorized pattern heightens the recollection ability even though the above condition is not satisfied.     

A hierarchical neural network model for associative memory

A hierarchical neural network model with feedback interconnections, which has the function of associative memory and the ability to recognize patterns, is proposed. The model consists of a hierarchical multi-layered network to which efferent connections are added, so as to make positive feedback loops in pairs with afferent connections. The cell-layer at the initial stage of the network is the input layer which receives the stimulus input and at the same time works as an output layer for associative recall. The deepest layer is the output layer for pattern-recognition. Pattern-recognition is performed hierarchically by integrating information by converging afferent paths in the network. For the purpose of associative recall, the integrated information is again distributed to lower-order cells by diverging efferent paths. These two operations progress simultaneously in the network. If a fragment of a training pattern is presented to the network which has completed its self-organization, the entire pattern will gradually be recalled in the initial layer. If a stimulus consisting of a number of training patterns superposed is presented, one pattern gradually becomes predominant in the recalled output after competition between the patterns, and the others disappear. At about the same time when the recalled pattern reaches a steady state in he initial layer, in the deepest layer of the network, a response is elicited from the cell corresponding to the category of the finally-recalled pattern. Once a steady state has been reached, the response of the network is automatically extinguished by inhibitory signals from a steadiness-detecting cell. If the same stimulus is still presented after inhibition, a response for another pattern, formerly suppressed, will now appear, because the cells of the network have adaptation characteristics which makes the same response unlikely to recur. Since inhibition occurs repeatedly, the superposed input patterns are recalled one by one in turn.     

Fine-Grained,Local Maps and Coarse,Global Representations Support Human Spatial Working Memory

While sensory processes are tuned to particular features, such as an object''s specific location, color or orientation, visual working memory (vWM) is assumed to store information using representations, which generalize over a feature dimension. Additionally, current vWM models presume that different features or objects are stored independently. On the other hand, configurational effects, when observed, are supposed to mainly reflect encoding strategies. We show that the location of the target, relative to the display center and boundaries, and overall memory load influenced recall precision, indicating that, like sensory processes, capacity limited vWM resources are spatially tuned. When recalling one of three memory items the target distance from the display center was overestimated, similar to the error when only one item was memorized, but its distance from the memory items'' average position was underestimated, showing that not only individual memory items'' position, but also the global configuration of the memory array may be stored. Finally, presenting the non-target items at recall, consequently providing landmarks and configurational information, improved precision and accuracy of target recall. Similarly, when the non-target items were translated at recall, relative to their position in the initial display, a parallel displacement of the recalled target was observed. These findings suggest that fine-grained spatial information in vWM is represented in local maps whose resolution varies with distance from landmarks, such as the display center, while coarse representations are used to store the memory array configuration. Both these representations are updated at the time of recall.     

Optimal association with partly missing key vectors

A new association scheme which can still recall appropriate data when some key elements are missing (blank) is presented. The traditional associative memory models are designed to deal with complete (memorized) keys, but in the real world, key elements are often missing due to error, equipment failure, observation difficulty, etc. The traditional models, in this case, can not have an optimal association except for special cases. When an incomplete key containing blanks is given, we wish to get the same data, as nearly as possible, as would be obtained with the complete key. In this paper, the optimal associative memory model which operates with partly missing keys is proposed. The model is constructed on the basis of the theory of the pseudoinverse of matrices. Even from the incomplete keys which contain a large percentage of blanks, the model recalls the appropriate data optimally under the MSE criterion. From the results of computer simulations, we can show that the model has the expected ability.     

Innervation pattern of suboesophageal ventral unpaired median neurones in the honeybee brain

In honeybees (Apis mellifera), the biogenic amine octopamine has been shown to play a role in associative and non-associative learning and in the division of labour in the hive. Immunohistochemical studies indicate that the ventral unpaired median (VUM) neurones in the suboesophageal ganglion (SOG) are putatively octopaminergic and therefore might be involved in the octopaminergic modulation of behaviour. In contrast to our knowledge about the behavioural effects of octopamine, only one neurone (VUMmx1) has been related to a behavioural effect (the reward function during olfactory learning). In this study, we have investigated suboesophageal VUM neurones with fluorescent dye-tracing techniques and intracellular recordings combined with intracellular staining. Ten different VUM neurones have been found including six VUM neurones innervating neuropile regions of the brain and the SOG exclusively (central VUM neurones) and four VUM neurones with axons in peripheral nerves (peripheral VUM neurones). The central VUM neurones innervate the antennal lobes, the protocerebral lobes (including the lateral horn) and the mushroom body calyces. Of these, a novel mandibular VUM neurone, VUMmd1, exhibits the same branching pattern in the brain as VUMmx1 and responds to sucrose and odours in a similar way. The peripheral VUM neurones innervate the antennal and the mandibular nerves. In addition, we describe one labial unpaired median neurone with a dorsal cell body, DUMlb1. The possible homology between the honeybee VUM neurones and the unpaired median neurones in other insects is discussed. This work was supported by the DFG ME 365/24-2.     

A model of distributed type associative memory with quantized Hadamard transform

This paper proposes a new correlation matrix network model of associative memory in brain. Each memorized pattern which consists of binary (+1 or-1) elements is preprocessed by a quantized Hadamard transform to increase selectivity. The association ability of a correlation matrix network model depends on the orthogonality between key patterns by which the corresponding memorized patterns are associatively recalled. In a brain model, however, it is rare that the key patterns are mutually orthogonal since they are memorized patterns themselves. The quantized Hadamard transform, presented in this paper, renders the memorized patterns approximately orthogonal. The model is tested by computer simulation.     

Associative Memory and Segmentation in an Oscillatory Neural Model of the Olfactory Bulb

We discuss the first few stages of olfactory processing in the framework of a layered neural network. Its central component is an oscillatory associative memory, describing the external plexiform layer, that consists of inhibitory and excitatory neurons with dendrodendritic interactions. We explore the computational properties of this neural network and point out its possible functional role in the olfactory bulb. When receiving a complex input that is composed of several odors, the network segments it into its components. This is done in two stages. First, multiple odor input is preprocessed in the glomerular layer via a decorrelation mechanism that relies on temporal independence of odor sources. Second, as the recall process of a pattern consists of associative convergence to an oscillatory attractor, multiple inputs are identified by alternate dominance of memory patterns during different sniff cycles. This could explain how quick analysis of mixed odors is subserved by the rapid sniffing behavior of highly olfactory animals. When one of the odors is much stronger than the rest, the network converges onto it, thus displaying odor masking.     

Basic module for an adaptive control system based on neurone information processing

The design of such devices as robotic aids for handicapped people, powered prostheses and manipulative aids such as page turners would benefit from the use of an adaptive control system. Much recent work on adaptive networks has been based on simplified models of the information processing capabilities of neurones. Neurones are now known to be capable of association learning and memory and this study incorporates these features in a neurone model. A single neuronal input system, the NMDA-type glutamate receptor, is modelled by deriving finite difference equations from its reaction dynamics so that the concentration of several molecules in the receptor can be plotted as a function of time. The model shows association learning taking place at the glutamate receptor. A whole neurone with ten glutamate receptor regions is also modelled and shows that a neurone should be capable of recognizing patterns of inputs. As the neurone model is complicated and slow to run, a much simplified form of the model is described which embodies the basic features of neurone information processing in a simple algorithm.     

Computational dynamics of gradient bistable networks

We describe a neural-like, homogeneous network consisting of coupled bistable elements and we study its abilities of learning, pattern recognition and computation. The technique allows new possibilities of pattern recognition, including the memorization and perfect recall of several memory patterns, without interference from spurious states. When the coupling strength between elements exceeds a critical value, the network readily converges to a unique attractor. Below this critical value one could perfectly recall all memorized patterns.     

The formation of the area centralis of the retinal ganglion cell layer in the chick

In adult domestic chickens, the neurones in the retinal ganglion cell layer are very unevenly disposed such that there is a sixfold increase in neurone density from the retinal edge to the retinal centre. The formation of the high ganglion-cell-density area centralis was studied on chick retinal wholemounts from the 8th day of incubation (E8) to 4 weeks after hatching (4WAH). The density of viable neurones and the number and the distribution of pyknotic neurones in the ganglion cell layer were estimated across the whole retina. Between E8 and E10, the distribution of neurones in the ganglion cell layer was anisodensitic with 53,000 mm-2 in the centre compared to 34,000 mm-2 in the periphery of the retina. Thereafter, a progressively steeper gradient of neurone density developed, which decreased from 24,000 mm-2 in the retinal centre to 6000 mm-2 at the retinal periphery by 4WAH. Neuronal pyknosis in the ganglion cell layer was observed between E9 and E17. From E11 onwards, consistently more pyknotic neurones were found in the peripheral than in the central retina. It was estimated that over the period of cell death approximately twice as many neurones died per unit area in the retinal periphery than in the centre. Retinal area measurements and estimation of neurone densities in the ganglion cell layer after the period of neurone generation and neurone death indicated differential retinal expansion, with more expansion in the peripheral than in the central retina. These observations allow us to conclude that the formation of the area centralis of the chick retina involves (1) slightly higher cell generation in the retinal centre, (2) higher rate of cell loss in the retinal periphery and (3) differential retinal expansion.     

Plant memory: a tentative model

All memory functions have molecular bases, namely in signal reception and transduction, and in storage and recall of information. Thus, at all levels of organisation living organisms have some kind of memory. In plants one may distinguish two types. There are linear pathways from reception of signals and propagation of effectors to a type of memory that may be described by terms such as learning, habituation or priming. There is a storage and recall memory based on a complex network of elements with a high degree of integration and feedback. The most important elements envisaged are calcium waves, epigenetic modifications of DNA and histones, and regulation of timing via a biological clock. Experiments are described that document the occurrence of the two sorts of memory and which show how they can be distinguished. A schematic model of plant memory is derived as emergent from integration of the various modules. Possessing the two forms of memory supports the fitness of plants in response to environmental stimuli and stress.     

Plant-odour-specific receptor neurones on the antennae of female and male Spodoptera littoralis

Abstract. Receptor neurones with high selectivity and sensitivity to plant odours were found within short sensilla trichodea on the antenna of both female and male Spodoptera littoralis (Boisd.) (Lepidoptera: Noctuidae) by using single-sensillum recording techniques. In 112 sensilla from females and forty-one from males, twenty-four different receptor neurone types were characterized according to their specificity. Altogether, twenty-six plant and three sex pheromone compounds were tested. Receptor neurones responding with high specificity to flower odours, green leaf volatiles, oviposition deterrents and other general host plant odours were identified. In twenty-one receptor neurone types, responses were elicited by one or several plant compounds, and in three types responses were elicited by sex pheromone compounds. The majority of the receptor neurones responded to only one or two of the tested compounds. In general, only one of the two receptor neurones in a sensillum responded to any of the compounds tested. An exception was a receptor neurone responding to plant odours (green leaf volatiles) and another receptor neurone responding to a sex pheromone compound ([Z]-7-dodecenyl (acetate), which occurred in the same sensillum. The majority of the receptor neurones displayed a high sensitivity to plant odours. No morphological difference was identified the different sensillum types.     

Physical models of neural networks

The state of art in computer modelling of neural networks with associative memory is reviewed. The available experimental data are considered on learning and memory of small neural systems, on isolated synapses and on molecular level. Computer simulations demonstrate that realistic models of neural ensembles exhibit properties which can be interpreted as image recognition, categorization, learning, prototype forming, etc. A bilayer model of associative neural network is proposed. One layer corresponds to the short-term memory, the other one to the long-term memory. Patterns are stored in terms of the synaptic strength matrix. We have studied the relaxational dynamics of neurons firing and suppression within the short-term memory layer under the influence of the long-term memory layer. The interaction among the layers has found to create a number of novel stable states which are not the learning patterns. These synthetic patterns may consist of elements belonging to different non-intersecting learning patterns. Within the framework of a hypothesis of selective and definite coding of images in brain one can interpret the observed effect as the "idea? generating" process.     

Elementary transcallosal connections of the rabbit sensomotor cortex]

Antidromic responses of two callosal neurones to a local electrical stimulation of the rabbit sensorimotor cortex may be recorded simultaneously with one microelectrode in the homotopic cortical area. In such recording conditions the relative amplitude of extracellularly recorded action potentials of the two neurones is determined primarily by the distance between these neurones and the electrode's tip. In response to the stimulation of the symmetrical area transcallosal monosynaptic excitation of the callosal neurone may occur; two callosal neurones may exite monosynaptically one and the same recorded neurone. The results suggest the existence of clusters or columns, formed jointly by the bodies and terminals of callosal neurones; a functional interconnection between symmetrical clusters or columns may exist, in particular a positive feedback.     

Mapping and Deciphering Neural Codes of NMDA Receptor-Dependent Fear Memory Engrams in the Hippocampus

Mapping and decoding brain activity patterns underlying learning and memory represents both great interest and immense challenge. At present, very little is known regarding many of the very basic questions regarding the neural codes of memory: are fear memories retrieved during the freezing state or non-freezing state of the animals? How do individual memory traces give arise to a holistic, real-time associative memory engram? How are memory codes regulated by synaptic plasticity? Here, by applying high-density electrode arrays and dimensionality-reduction decoding algorithms, we investigate hippocampal CA1 activity patterns of trace fear conditioning memory code in inducible NMDA receptor knockout mice and their control littermates. Our analyses showed that the conditioned tone (CS) and unconditioned foot-shock (US) can evoke hippocampal ensemble responses in control and mutant mice. Yet, temporal formats and contents of CA1 fear memory engrams differ significantly between the genotypes. The mutant mice with disabled NMDA receptor plasticity failed to generate CS-to-US or US-to-CS associative memory traces. Moreover, the mutant CA1 region lacked memory traces for “what at when” information that predicts the timing relationship between the conditioned tone and the foot shock. The degraded associative fear memory engram is further manifested in its lack of intertwined and alternating temporal association between CS and US memory traces that are characteristic to the holistic memory recall in the wild-type animals. Therefore, our study has decoded real-time memory contents, timing relationship between CS and US, and temporal organizing patterns of fear memory engrams and demonstrated how hippocampal memory codes are regulated by NMDA receptor synaptic plasticity.     

Selective receptor neurone responses to E-β-ocimene, β-myrcene, E,E-α-farnesene and homo-farnesene in the moth Heliothis virescens, identified by gas chromatography linked to electrophysiology

An important question in olfaction is for which odorants receptor neurones have evolved. In the present study, olfactory receptor neurones on the antennae of the tobacco budworm moth Heliothis virescens were screened for sensitivity to naturally occurring plant-produced volatiles by the use of gas chromatography linked to electrophysiology. Volatiles of host as well as non-host plants collected by headspace techniques were used for stimulating the neurones, sequentially via two columns, one polar and one nonpolar installed in parallel in the gas chromatograph. Three types of neurones presented in this paper responded to one, two or three compounds for which the retention times were determined in both column types. The chemical structures of the active components were determined on the basis of mass spectrometry linked to gas chromatography, indicating E-beta-ocimene and beta-myrcene as stimulants for neurone type 1, E,E-alpha-farnesene for neurone type 2 and homo-farnesene for neurone type 3. Re-testing authentic materials verified the identifications for the type 1 neurones. The results demonstrate a high specificity for the three types of neurones by strong responses to one or two structurally similar compounds out of hundreds present in a large variety of plants. The study exemplifies plant odour detection by narrowly tuned receptor neurones in a polyphagous moth species.     

The Role of Inhibition in an Associative Memory Model of the Olfactory Bulb

The external plexiform layer is where theinteractions between the mitral (excitatory) and granule (inhibitory)cells of the olfactory bulb (OB) take place. Two outstanding features ofthese interactions are that they aredendrodendritic and that there seem to be nonebetween excitatory cells. The latter are usually credited with the role of forming Hebbian cell assemblies.Hence, it would seem that this structure lacks the necessaryingredients for an associative memory system.In this article we show that in spite of these two properties thissystem can serve as an associative memory. Our model incorporates theessential anatomical characteristics of the OB. The memories in oursystem, defined by Hebbian mitral assemblies, are activated viathe interactions with the inhibitory granule cells. The nonlinearityis introduced in our model via a sigmoid function that describesneurotransmitter release in reciprocal dendrodendritic synapses. Thecapacity (maximal number of odors that can be memorized) depends onthe sparseness of coding that is being used. For very low memoryactivities, the capacity grows as a fractional power of the number ofneurons. We validate the theoretical results by numericalsimulations. An interesting result of our model is that its capacityincreases as a function of the ratio of inhibitory to excitatorypopulations. This may provide an explanation for the dominance ofinhibitory cells in the olfactory bulb.     

On optimal associative recall by an incomplete key

A new association scheme is proposed. The fundamental principle of the conventional associative memory models is to solve the matrix equation which is made by the complete (memorized) keys and responses. Therefore, when an incomplete key pattern which is a fraction of memorized key is given to the models as a key, these models can not have an optimal association except for a special case. Analyzing the property of the incomplete key pattern, in this paper, we propose a new association model which behaves optimally for an incomplete key pattern. From the result of the computer simulation, we understand that this model has the expected ability.     

Invariant Features of Memory Capacity

Our purpose in these experiments was to study short-term (immediate) and long-term memory in the memorization of the same subject matter. By memory capacity is meant the number of units a person is able to reproduce in one repetition, or on an average in one repetition. In short-term reception and full reproduction of the material to be memorized, short-term memory capacity is commensurate with the number of units reproduced. Long-term memory capacity reflects the ability to accumulate as well as to retain information. When the material to be remembered exceeds the short-term memory capacity, the first reproduction is incomplete, and multiple presentation and repetition of the information are necessary for error-free and complete reproduction. In this case the memory capacity is equal to the number of units contained in the presented material divided by the number of repetitions.