Lubica Benuskova | University of Otago (original) (raw)

Uploads

Papers by Lubica Benuskova

NEURAL NETWORK WORLD, 1998

A significant feature of spiking neural networks with varying connection delays, such as those in... more A significant feature of spiking neural networks with varying connection delays, such as those in the brain, is the existence of strongly connected groups of neurons known as polychronous neural groups (PNGs). Polychronous groups are found in large numbers in these networks and are proposed by Izhikevich (2006a) to provide a neural basis for representation and memory. When exposed to a familiar stimulus, spiking neural networks produce consistencies in the spiking output data that are the hallmarks of PNG activation. Previous methods for studying the PNG activation response to stimuli have been limited by the template-based methods used to identify PNG activation. In this letter, we outline a new method that overcomes these difficulties by establishing for the first time a probabilistic interpretation of PNG activation. We then demonstrate the use of this method by investigating the claim that PNGs might provide the foundation of a representational system.

Computational models of metaplasticity have usually focused on the modeling of single synapses (S... more Computational models of metaplasticity have usually focused on the modeling of single synapses (Shouval et al., 2002). In this paper we study the effect of metaplasticity on network behavior. Our guiding assumption is that the primary purpose of metaplasticity is to regulate synaptic plasticity, by increasing it when input is low and decreasing it when input is high. For our experiments we adopt a model of metaplasticity that demonstrably has this effect for a single synapse; our primary interest is in how metaplasticity thus defined affects network-level phenomena. We focus on a network-level phenomenon called polychronicity, that has a potential role in representation and memory. A network with polychronicity has the ability to produce non-synchronous but precisely timed sequences of neural firing events that can arise from strongly connected groups of neurons called polychronous neural groups (Izhikevich et al., 2004). Polychronous groups (PNGs) develop readily when spiking networks are exposed to repeated spatio-temporal stimuli under the influence of spike-timing-dependent plasticity (STDP), but are sensitive to changes in synaptic weight distribution. We use a technique we have recently developed called Response Fingerprinting to show that PNGs formed in the presence of metaplasticity are significantly larger than those with no metaplasticity. A potential mechanism for this enhancement is proposed that links an inherent property of integrator type neurons called spike latency to an increase in the tolerance of PNG neurons to jitter in their inputs.

The posterior-anterior shift in aging (PASA) is a commonly observed phenomenon in functional neur... more The posterior-anterior shift in aging (PASA) is a commonly observed phenomenon in functional neuroimaging studies of aging, characterized by age-related reductions in occipital activity alongside increases in frontal activity. In this work we have investigated the hypothesis as to whether the PASA is also manifested in functional brain network measures such as degree, clustering coefficient, path length and local efficiency. We have performed statistical analysis upon functional networks derived from a fMRI dataset containing data from healthy young, healthy aged, and aged individuals with very mild to mild Alzheimer's disease (AD). Analysis of both task based and resting state functional network properties has indicated that the PASA can also be characterized in terms of modulation of functional network properties, and that the onset of AD appears to accentuate this modulation. We also explore the effect of spatial normalization upon the results of our analysis.

Abstract We have combined the nearest neighbour additive spike-timing-dependent plasticity (STDP)... more Abstract We have combined the nearest neighbour additive spike-timing-dependent plasticity (STDP) rule with the Bienenstock, Cooper and Munro (BCM) sliding modification threshold in a computational model of heterosynaptic plasticity in the hippocampal dentate gyrus. As a result we can reproduce (1) homosynaptic long-term potentiation of the tetanized input, and (2) heterosynaptic long-term depression of the untetanized input, as observed in real experiments.

Abstract Most of the interest in computational modelling is devoted to the phenomenon of long-ter... more Abstract Most of the interest in computational modelling is devoted to the phenomenon of long-term potentiation (LTP) of synaptic efficacy. Long-term depression (LTD) of synaptic connections is often overlooked. We have found that it is in fact very difficult to induce LTD experimentally and the conditions for LTD induction are much less clear than those for induction of LTP.

Abstract Previous electrophysiological experiments have documented the response of neurons in the... more Abstract Previous electrophysiological experiments have documented the response of neurons in the adult rat somatic sensory (" barrel") cortex to whisker movement after normal experience and after periods of experience with all but two whiskers trimmed close to the face (whisker" pairing").

Abstract In this paper, we elaborate upon the claim that clustering in the recurrent layer of rec... more Abstract In this paper, we elaborate upon the claim that clustering in the recurrent layer of recurrent neural networks (RNNs) reflects meaningful information processing states even prior to training. By concentrating on activation clusters in RNNs, while not throwing away the continuous state space network dynamics, we extract predictive models that we call neural prediction machines (NPMs).

Abstract Heterosynaptic long-term depression (LTD) is conventionally defined as occurring at syna... more Abstract Heterosynaptic long-term depression (LTD) is conventionally defined as occurring at synapses that are inactive during a time when neighboring synapses are activated by high-frequency stimulation. A new model that combines computational properties of both the Bienenstock, Cooper and Munro model and spike timing-dependent plasticity, however, suggests that such LTD actually may require presynaptic activity in the depressed pathway.

Abstract We model experience-dependent plasticity in the cortical representation of whiskers (the... more Abstract We model experience-dependent plasticity in the cortical representation of whiskers (the barrel cortex) in normal adult rats, and in adult rats that were prenatally exposed to alcohol. Prenatal exposure to alcohol (PAE) caused marked deficits in experience-dependent plasticity in a cortical barrel-column. Cortical plasticity was induced by trimming all whiskers on one side of the face except two.

We simulate the induction and maintenance of late long-term potentiation (L-LTP) in the hippocamp... more We simulate the induction and maintenance of late long-term potentiation (L-LTP) in the hippocampal dentate gyrus by means of a new synaptic plasticity rule that is the result of combination of the spike-timing-dependent plasticity (STDP) and the moving LTD/LTP threshold θM from the Bienenstock, Cooper and Munro (BCM) theory of synaptic plasticity.

Feedforward and recurrent neural networks (RNNs) are broadly used to simulate cognitive phenomena... more Feedforward and recurrent neural networks (RNNs) are broadly used to simulate cognitive phenomena. In this work we used two types of RNNs, one trained with a common error backpropagation technique, and the other one trained with a biologically more plausible BCM rule, to simulate actual psychophysical experiment performed with human subjects on the task of word segmentation in a continuous speech stream originating from a novel language. The only cues available for word segmentation were the transitional probabilities between syllables. We show that artificial RNNs greatly outperform humans in this task, which raises the question about suitability of these neural models in explaining processing of symbolic time series, for instance speech and language, in humans.

Long-term potentiation (LTP) is a long-lasting enhancement in signal transmission between two neu... more Long-term potentiation (LTP) is a long-lasting enhancement in signal transmission between two neurons, and represents a widely accepted experimental model for long-term memory processes. Although it is now clear that the maintenance of LTP requires new gene transcription, little is known on the genetic mechanisms underlying these changes. We assume that an LTP-related gene regulatory network has two equilibrium states in terms of gene expression levels which correspond to a pre-and post-LTP states. This network is shifted from the first to the latter by means of a perturbation, which experimentally corresponds to the high-frequency stimulus necessary to induce LTP in vivo. Based on this assumption and by means of modeling the transcriptional regulation with weight matrices, we study the properties of the main LTP-related network recently proposed in [1]. First, we classify the LTP-related genes according to their relevance to the bistable dynamic output of the network. In addition, we demonstrate how the LTP gene regulatory network architecture holds a higher tendency towards bistable behaviours than we should expect of a random network.

Abstract Recurrent neural networks (RNNs) have much larger potential than classical feed-forward ... more Abstract Recurrent neural networks (RNNs) have much larger potential than classical feed-forward neural networks. Their output responses depend also on the time position of a given input and they can be successfully used in spatio-temporal task processing. RNNs are often used in the cognitive science community to process symbol sequences that represent various natural language structures. Usually they are trained by common gradient-based algorithms such as real time recurrent learning or backpropagation through time.

Abstract. Kalman filter (KF)-based techniques used for recurrent neural networks (RNNs) training ... more Abstract. Kalman filter (KF)-based techniques used for recurrent neural networks (RNNs) training on real-valued time series have already shown their potential. On the other hand gradient descent approaches such as back-propagation through time (BPTT) or real-time recurrent learning (RTRL) algorithms are still widely used by researchers working with symbolic sequences. The aim of this work is to show how KF-based techniques used for training RNNs can deal with symbolic time series.

The paper describes the integration of brain-inspired systems to perform audiovisual pattern reco... more The paper describes the integration of brain-inspired systems to perform audiovisual pattern recognition tasks. Individual sensory pathways as well as the integrative modules are implemented using a fast version of spiking neurons grouped in evolving spiking neural network (ESNN) architectures capable of lifelong adaptation. We design a new crossmodal integration system, where individual modalities can influence others before individual decisions are made, fact that resembles some characteristics of the biological brains.

Summary We used a biologically realistic pulsed neural network model of the piriform cortex (PC) ... more Summary We used a biologically realistic pulsed neural network model of the piriform cortex (PC) presented in the book of GENESIS to investigate the underlying dynamics leading to a realistically looking 40 Hz-awake EEG.

V súčasnosti je jedným z nových pojmov kognitívnych vied (angl. cognitive science) pojem stelesne... more V súčasnosti je jedným z nových pojmov kognitívnych vied (angl. cognitive science) pojem stelesnenia poznania či poznatkov (angl. embodiment of knowledge). Poznanie sa už ďalej nedá poznávať ako niečo abstraktné a odtrhnuté od neurónového substrátu. Nevyhneme sa odhaľovaniu neurónových štruktúr či architektúr a skúmaniu dynamiky týchto zoskupení. Avšak konkrétna neurónová architektúra sa dá pochopiť iba v kontexte fylogenetickej a ontogenetickej evolúcie.

We show that representation of DNA sequences by means of Iterated Function Systems (IFS) can be u... more We show that representation of DNA sequences by means of Iterated Function Systems (IFS) can be used as a fast and useful 2D visualization tool for bioinformatics analysis and comparison of genomic data. The methods could be applied to any long symbolic sequences as well, with potential for fast searching as well.

NEURAL NETWORK WORLD, 1998

A significant feature of spiking neural networks with varying connection delays, such as those in... more A significant feature of spiking neural networks with varying connection delays, such as those in the brain, is the existence of strongly connected groups of neurons known as polychronous neural groups (PNGs). Polychronous groups are found in large numbers in these networks and are proposed by Izhikevich (2006a) to provide a neural basis for representation and memory. When exposed to a familiar stimulus, spiking neural networks produce consistencies in the spiking output data that are the hallmarks of PNG activation. Previous methods for studying the PNG activation response to stimuli have been limited by the template-based methods used to identify PNG activation. In this letter, we outline a new method that overcomes these difficulties by establishing for the first time a probabilistic interpretation of PNG activation. We then demonstrate the use of this method by investigating the claim that PNGs might provide the foundation of a representational system.

Computational models of metaplasticity have usually focused on the modeling of single synapses (S... more Computational models of metaplasticity have usually focused on the modeling of single synapses (Shouval et al., 2002). In this paper we study the effect of metaplasticity on network behavior. Our guiding assumption is that the primary purpose of metaplasticity is to regulate synaptic plasticity, by increasing it when input is low and decreasing it when input is high. For our experiments we adopt a model of metaplasticity that demonstrably has this effect for a single synapse; our primary interest is in how metaplasticity thus defined affects network-level phenomena. We focus on a network-level phenomenon called polychronicity, that has a potential role in representation and memory. A network with polychronicity has the ability to produce non-synchronous but precisely timed sequences of neural firing events that can arise from strongly connected groups of neurons called polychronous neural groups (Izhikevich et al., 2004). Polychronous groups (PNGs) develop readily when spiking networks are exposed to repeated spatio-temporal stimuli under the influence of spike-timing-dependent plasticity (STDP), but are sensitive to changes in synaptic weight distribution. We use a technique we have recently developed called Response Fingerprinting to show that PNGs formed in the presence of metaplasticity are significantly larger than those with no metaplasticity. A potential mechanism for this enhancement is proposed that links an inherent property of integrator type neurons called spike latency to an increase in the tolerance of PNG neurons to jitter in their inputs.

The posterior-anterior shift in aging (PASA) is a commonly observed phenomenon in functional neur... more The posterior-anterior shift in aging (PASA) is a commonly observed phenomenon in functional neuroimaging studies of aging, characterized by age-related reductions in occipital activity alongside increases in frontal activity. In this work we have investigated the hypothesis as to whether the PASA is also manifested in functional brain network measures such as degree, clustering coefficient, path length and local efficiency. We have performed statistical analysis upon functional networks derived from a fMRI dataset containing data from healthy young, healthy aged, and aged individuals with very mild to mild Alzheimer's disease (AD). Analysis of both task based and resting state functional network properties has indicated that the PASA can also be characterized in terms of modulation of functional network properties, and that the onset of AD appears to accentuate this modulation. We also explore the effect of spatial normalization upon the results of our analysis.

Abstract We have combined the nearest neighbour additive spike-timing-dependent plasticity (STDP)... more Abstract We have combined the nearest neighbour additive spike-timing-dependent plasticity (STDP) rule with the Bienenstock, Cooper and Munro (BCM) sliding modification threshold in a computational model of heterosynaptic plasticity in the hippocampal dentate gyrus. As a result we can reproduce (1) homosynaptic long-term potentiation of the tetanized input, and (2) heterosynaptic long-term depression of the untetanized input, as observed in real experiments.

Abstract Most of the interest in computational modelling is devoted to the phenomenon of long-ter... more Abstract Most of the interest in computational modelling is devoted to the phenomenon of long-term potentiation (LTP) of synaptic efficacy. Long-term depression (LTD) of synaptic connections is often overlooked. We have found that it is in fact very difficult to induce LTD experimentally and the conditions for LTD induction are much less clear than those for induction of LTP.

Abstract Previous electrophysiological experiments have documented the response of neurons in the... more Abstract Previous electrophysiological experiments have documented the response of neurons in the adult rat somatic sensory (" barrel") cortex to whisker movement after normal experience and after periods of experience with all but two whiskers trimmed close to the face (whisker" pairing").

Abstract In this paper, we elaborate upon the claim that clustering in the recurrent layer of rec... more Abstract In this paper, we elaborate upon the claim that clustering in the recurrent layer of recurrent neural networks (RNNs) reflects meaningful information processing states even prior to training. By concentrating on activation clusters in RNNs, while not throwing away the continuous state space network dynamics, we extract predictive models that we call neural prediction machines (NPMs).

Abstract Heterosynaptic long-term depression (LTD) is conventionally defined as occurring at syna... more Abstract Heterosynaptic long-term depression (LTD) is conventionally defined as occurring at synapses that are inactive during a time when neighboring synapses are activated by high-frequency stimulation. A new model that combines computational properties of both the Bienenstock, Cooper and Munro model and spike timing-dependent plasticity, however, suggests that such LTD actually may require presynaptic activity in the depressed pathway.

Abstract We model experience-dependent plasticity in the cortical representation of whiskers (the... more Abstract We model experience-dependent plasticity in the cortical representation of whiskers (the barrel cortex) in normal adult rats, and in adult rats that were prenatally exposed to alcohol. Prenatal exposure to alcohol (PAE) caused marked deficits in experience-dependent plasticity in a cortical barrel-column. Cortical plasticity was induced by trimming all whiskers on one side of the face except two.

We simulate the induction and maintenance of late long-term potentiation (L-LTP) in the hippocamp... more We simulate the induction and maintenance of late long-term potentiation (L-LTP) in the hippocampal dentate gyrus by means of a new synaptic plasticity rule that is the result of combination of the spike-timing-dependent plasticity (STDP) and the moving LTD/LTP threshold θM from the Bienenstock, Cooper and Munro (BCM) theory of synaptic plasticity.

Feedforward and recurrent neural networks (RNNs) are broadly used to simulate cognitive phenomena... more Feedforward and recurrent neural networks (RNNs) are broadly used to simulate cognitive phenomena. In this work we used two types of RNNs, one trained with a common error backpropagation technique, and the other one trained with a biologically more plausible BCM rule, to simulate actual psychophysical experiment performed with human subjects on the task of word segmentation in a continuous speech stream originating from a novel language. The only cues available for word segmentation were the transitional probabilities between syllables. We show that artificial RNNs greatly outperform humans in this task, which raises the question about suitability of these neural models in explaining processing of symbolic time series, for instance speech and language, in humans.

Long-term potentiation (LTP) is a long-lasting enhancement in signal transmission between two neu... more Long-term potentiation (LTP) is a long-lasting enhancement in signal transmission between two neurons, and represents a widely accepted experimental model for long-term memory processes. Although it is now clear that the maintenance of LTP requires new gene transcription, little is known on the genetic mechanisms underlying these changes. We assume that an LTP-related gene regulatory network has two equilibrium states in terms of gene expression levels which correspond to a pre-and post-LTP states. This network is shifted from the first to the latter by means of a perturbation, which experimentally corresponds to the high-frequency stimulus necessary to induce LTP in vivo. Based on this assumption and by means of modeling the transcriptional regulation with weight matrices, we study the properties of the main LTP-related network recently proposed in [1]. First, we classify the LTP-related genes according to their relevance to the bistable dynamic output of the network. In addition, we demonstrate how the LTP gene regulatory network architecture holds a higher tendency towards bistable behaviours than we should expect of a random network.

Abstract Recurrent neural networks (RNNs) have much larger potential than classical feed-forward ... more Abstract Recurrent neural networks (RNNs) have much larger potential than classical feed-forward neural networks. Their output responses depend also on the time position of a given input and they can be successfully used in spatio-temporal task processing. RNNs are often used in the cognitive science community to process symbol sequences that represent various natural language structures. Usually they are trained by common gradient-based algorithms such as real time recurrent learning or backpropagation through time.

Abstract. Kalman filter (KF)-based techniques used for recurrent neural networks (RNNs) training ... more Abstract. Kalman filter (KF)-based techniques used for recurrent neural networks (RNNs) training on real-valued time series have already shown their potential. On the other hand gradient descent approaches such as back-propagation through time (BPTT) or real-time recurrent learning (RTRL) algorithms are still widely used by researchers working with symbolic sequences. The aim of this work is to show how KF-based techniques used for training RNNs can deal with symbolic time series.

The paper describes the integration of brain-inspired systems to perform audiovisual pattern reco... more The paper describes the integration of brain-inspired systems to perform audiovisual pattern recognition tasks. Individual sensory pathways as well as the integrative modules are implemented using a fast version of spiking neurons grouped in evolving spiking neural network (ESNN) architectures capable of lifelong adaptation. We design a new crossmodal integration system, where individual modalities can influence others before individual decisions are made, fact that resembles some characteristics of the biological brains.

Summary We used a biologically realistic pulsed neural network model of the piriform cortex (PC) ... more Summary We used a biologically realistic pulsed neural network model of the piriform cortex (PC) presented in the book of GENESIS to investigate the underlying dynamics leading to a realistically looking 40 Hz-awake EEG.

V súčasnosti je jedným z nových pojmov kognitívnych vied (angl. cognitive science) pojem stelesne... more V súčasnosti je jedným z nových pojmov kognitívnych vied (angl. cognitive science) pojem stelesnenia poznania či poznatkov (angl. embodiment of knowledge). Poznanie sa už ďalej nedá poznávať ako niečo abstraktné a odtrhnuté od neurónového substrátu. Nevyhneme sa odhaľovaniu neurónových štruktúr či architektúr a skúmaniu dynamiky týchto zoskupení. Avšak konkrétna neurónová architektúra sa dá pochopiť iba v kontexte fylogenetickej a ontogenetickej evolúcie.

We show that representation of DNA sequences by means of Iterated Function Systems (IFS) can be u... more We show that representation of DNA sequences by means of Iterated Function Systems (IFS) can be used as a fast and useful 2D visualization tool for bioinformatics analysis and comparison of genomic data. The methods could be applied to any long symbolic sequences as well, with potential for fast searching as well.