A hippocampal cognitive prosthesis: multi-input, multi-output nonlinear modeling and VLSI implementation

This paper describes the development of a cognitive prosthesis designed to restore the ability to form new long-term memories typically lost after damage to the hippocampus. The animal model used is delayed nonmatch-to-sample (DNMS) behavior in the rat, and the "core" of the prosthesis is a biomimetic multi-input/multi-output (MIMO) nonlinear model that provides the capability for predicting spatio-temporal spike train output of hippocampus (CA1) based on spatio-temporal spike train inputs recorded presynaptically to CA1 (e.g., CA3). We demonstrate the capability of the MIMO model for highly accurate predictions of CA1 coded memories that can be made on a single-trial basis and in real-time. When hippocampal CA1 function is blocked and long-term memory formation is lost, successful DNMS behavior also is abolished. However, when MIMO model predictions are used to reinstate CA1 memory-related activity by driving spatio-temporal electrical stimulation of hippocampal output to mimic the patterns of activity observed in control conditions, successful DNMS behavior is restored. We also outline the design in very-large-scale integration for a hardware implementation of a 16-input, 16-output MIMO model, along with spike sorting, amplification, and other functions necessary for a total system, when coupled together with electrode arrays to record extracellularly from populations of hippocampal neurons, that can serve as a cognitive prosthesis in behaving animals.

Restorative encoding memory integrative neural device: "REMIND"

Construction and application of a neural prosthesis device that enhances existing and replaces lost memory capacity in humans is the focus of research described here in rodents. A unique approach for the analysis and application of neural population firing has been developed to decipher the pattern in which information is successfully encoded by the hippocampus where mnemonic accuracy is critical. A nonlinear dynamic multi-input multi-output (MIMO) model is utilized to extract memory relevant firing patterns in CA3 and CA1 and to predict online what the consequences of the encoded firing patterns reflect for subsequent information retrieval for successful performance of delayed-nonmatch-to-sample (DNMS) memory task in rodents. The MIMO model has been tested successfully in a number of different contexts, each of which produced improved performance by a) utilizing online predicted codes to regulate task difficulty, b) employing electrical stimulation of CA1 output areas in the same pattern as successful cell firing, c) employing electrical stimulation to recover cell firing compromised by pharmacological agents and d) transferring and improving performance in naïve animals using the same stimulation patterns that are effective in fully trained animals. The results in rodents formed the basis for extension of the MIMO model to nonhuman primates in the same type of memory task that is now being tested in the last step prior to its application in humans.

Modeling of the nervous system: from molecular dynamics and synaptic modulation to neuron spiking activity

The brain is a perfect example of an integrated multi-scale system, as the complex interactions taking place at the molecular level regulate neuronal activity that further modifies the function of millions of neurons connected by trillions of synapses, ultimately giving rise to complex function and behavior at the system level. Likewise, the spatial complexity is accompanied by a complex temporal integration of events taking place at the microsecond scale leading to slower changes occurring at the second, minute and hour scales. In the present study we illustrate our approach to model and simulate the spatio-temporal complexity of the nervous system by developing a multi-scale model integrating synaptic models into the neuronal and ultimately network levels. We apply this approach to a concrete example and demonstrate how changes at the level of kinetic parameters of a receptor model are translated into significant changes in the firing of a pyramidal neuron. These results illustrate the abilities of our modeling approach and support its direct application to the evaluation of the effects of drugs, from functional target to integrated system.

Towards a large-scale biologically realistic model of the hippocampus

Real neurobiological systems in the mammalian brain have a complicated and detailed structure, being composed of 1) large numbers of neurons with intricate, branching morphologies--complex morphology brings with it complex passive membrane properties; 2) active membrane properties--nonlinear sodium, potassium, calcium, etc. conductances; 3) non-uniform distributions throughout the dendritic and somal membrane surface of these non-linear conductances; 4) non-uniform and topographic connectivity between pre- and post-synaptic neurons; and 5) activity-dependent changes in synaptic function. One of the essential, and as yet unanswered questions in neuroscience is the role of these fundamental structural and functional features in determining "neural processing" properties of a given brain system. To help answer that question, we're creating a large-scale biologically realistic model of the intrinsic pathway of the hippocampus, which consists of the projection from layer II entorhinal cortex (EC) to dentate gyrus (DG), EC to CA3, DG to CA3, and CA3 to CA1. We describe the computational hardware and software tools the model runs on, and demonstrate its viability as a modeling platform with an EC-to-DG model.

Implementation of topographically constrained connectivity for a large-scale biologically realistic model of the hippocampus

In order to understand how memory works in the brain, the hippocampus is highly studied because of its role in the encoding of long-term memories. We have identified four characteristics that would contribute to the encoding process: the morphology of the neurons, their biophysics, synaptic plasticity, and the topography connecting the input to and the neurons within the hippocampus. To investigate how long-term memory is encoded, we are constructing a large-scale biologically realistic model of the rat hippocampus. This work focuses on how topography contributes to the output of the hippocampus. Generally, the brain is structured with topography such that the synaptic connections formed by an input neuron population are organized spatially across the receiving population. The first step in our model was to construct how entorhinal cortex inputs connect to the dentate gyrus of the hippocampus. We have derived realistic constraints from topographical data to connect the two cell populations. The details on how these constraints were applied are presented. We demonstrate that the spatial connectivity has a major impact on the output of the simulation, and the results emphasize the importance of carefully defining spatial connectivity in neural network models of the brain in order to generate relevant spatiotemporal patterns.

Analog low-power hardware implementation of a Laguerre-Volterra model of intracellular subthreshold neuronal activity

The right level of abstraction for a model mimicking a neural function is often difficult to determine. There are trade-offs between capturing biological complexities on one hand and the scalability and efficiency of the model on the other. In this work, we describe a nonlinear Laguerre-Volterra model of the synaptic temporal integration of input spikes to postsynaptic potentials. This model is then efficiently implemented using analog subthreshold circuits and can serve as a foundation for future large-scale hardware systems that can emulate multi-input multi-output (MIMO) spike transformations in populations of neurons. The normalized mean square error in estimating real data using the circuit implementation of this model is less than 15%. The model components are modular and its parameters are adjustable for modeling temporal integration by neurons in other brain regions. The total power consumption of this nonlinear Laguerre-Volterra system is less than 5nW.

Implementation of activity-dependent synaptic plasticity rules for a large-scale biologically realistic model of the hippocampus

A large-scale computational model of the hippocampus should consider plasticity at different time scales in order to capture the non-stationary information processing behavior of the hippocampus more accurately. This paper presents a computational model that describes hippocampal long-term potentiation/depression (LTP/LTD) and short-term plasticity implemented in the NEURON simulation environment. The LTP/LTD component is based on spike-timing-dependent plasticity (STDP). The short-term plasticity component modifies a previously defined deterministic model at a population synapse level to a probabilistic model that can be implemented at a single synapse level. The plasticity mechanisms are validated and incorporated into a large-scale model of the entorhinal cortex projection to the dentate gyrus. Computational expense of the added plasticity was also evaluated and shown to increase simulation time by less than a factor of two. This model can be easily included in future large-scale hippocampal simulations to investigate the effects of LTP/LTD and short-term plasticity in conjunction with other biological considerations on system function.

Functional connectivity between Layer 2/3 and Layer 5 neurons in prefrontal cortex of nonhuman primates during a delayed match-to-sample task

The prefrontal cortex (PFC) has been postulated to play critical roles in cognitive control and the formation of long-term memories. To gain insights into the neurobiological mechanism of such high-order cognitive functions, it is important to understand the input-output transformational properties of the PFC micro-circuitry. In this study, we identify the functional connectivity between the Layer 2/3 (input) neurons and the Layer 5 (output) neurons using a previously developed generalized Volterra model (GVM). Input-output spike trains are recorded from the PFCs of nonhuman primates performing a memory-dependent delayed match-to-sample task with a customized conformal ceramic multi-electrode array. The GVM describes how the input spike trains are transformed into the output spike trains by the PFC micro-circuitry and represents the transformation in the form of Volterra kernels. Results show that Layer 2/3 neurons have strong and transient facilitatory effects on the firings of Layer 5 neurons. The magnitude and temporal range of the input-output nonlinear dynamics are strikingly different from those of the hippocampal CA3-CA1. This form of functional connectivity may have important implications to understanding the computational principle of the PFC.

Spatio-temporal inter-ictal activity recorded from human epileptic hippocampal slices

Epilepsy is a medical syndrome that produces seizures affecting a variety of mental and physical functions. The actual mechanisms of the onset and termination of the seizure are still unclear. While medical therapies can suppress the symptoms of seizures, 30% of patients do not respond well. Temporal lobectomy is a common surgical treatment for medically refractory epilepsy. Part of the hippocampus is removed from the patient. In this study, we have developed an in vitro epileptic model in human hippocampal slices resected from patients suffering from intractable mesial temporal lobe epilepsy. Using a planar multielectrode array system, spatio-temporal inter-ictal activity can be consistently recorded in high-potassium (8 mM), low-magnesium (0.25 mM) aCSF with additional 100 µM 4-aminopyridine. The induced inter-ictal activity can be recorded in different regions including dentate, CA1 and Subiculum. We hope the experimental model built in this study will help us understand more about seizure generation, as well as providing insights into prevention and novel therapeutics.

Simulation of postsynaptic glutamate receptors reveals critical features of glutamatergic transmission

Activation of several subtypes of glutamate receptors contributes to changes in postsynaptic calcium concentration at hippocampal synapses, resulting in various types of changes in synaptic strength. Thus, while activation of NMDA receptors has been shown to be critical for long-term potentiation (LTP) and long term depression (LTD) of synaptic transmission, activation of metabotropic glutamate receptors (mGluRs) has been linked to either LTP or LTD. While it is generally admitted that dynamic changes in postsynaptic calcium concentration represent the critical elements to determine the direction and amplitude of the changes in synaptic strength, it has been difficult to quantitatively estimate the relative contribution of the different types of glutamate receptors to these changes under different experimental conditions. Here we present a detailed model of a postsynaptic glutamatergic synapse that incorporates ionotropic and mGluR type I receptors, and we use this model to determine the role of the different receptors to the dynamics of postsynaptic calcium with different patterns of presynaptic activation. Our modeling framework includes glutamate vesicular release and diffusion in the cleft and a glutamate transporter that modulates extracellular glutamate concentration. Our results indicate that the contribution of mGluRs to changes in postsynaptic calcium concentration is minimal under basal stimulation conditions and becomes apparent only at high frequency of stimulation. Furthermore, the location of mGluRs in the postsynaptic membrane is also a critical factor, as activation of distant receptors contributes significantly less to calcium dynamics than more centrally located ones. These results confirm the important role of glutamate transporters and of the localization of mGluRs in postsynaptic sites in their signaling properties, and further strengthen the notion that mGluR activation significantly contributes to postsynaptic calcium dynamics only following high-frequency stimulation. They also provide a new tool to analyze the interactions between metabotropic and ionotropic glutamate receptors.

Nonlinear dynamic modeling of neuron action potential threshold during synaptically driven broadband intracellular activity

Activity-dependent variation of neuronal thresholds for action potential (AP) generation is one of the key determinants of spike-train temporal-pattern transformations from presynaptic to postsynaptic spike trains. In this study, we model the nonlinear dynamics of the threshold variation during synaptically driven broadband intracellular activity. First, membrane potentials of single CA1 pyramidal cells were recorded under physiologically plausible broadband stimulation conditions. Second, a method was developed to measure AP thresholds from the continuous recordings of membrane potentials. It involves measuring the turning points of APs by analyzing the third-order derivatives of the membrane potentials. Four stimulation paradigms with different temporal patterns were applied to validate this method by comparing the measured AP turning points and the actual AP thresholds estimated with varying stimulation intensities. Results show that the AP turning points provide consistent measurement of the AP thresholds, except for a constant offset. It indicates that 1) the variation of AP turning points represents the nonlinearities of threshold dynamics; and 2) an optimization of the constant offset is required to achieve accurate spike prediction. Third, a nonlinear dynamical third-order Volterra model was built to describe the relations between the threshold dynamics and the AP activities. Results show that the model can predict threshold accurately based on the preceding APs. Finally, the dynamic threshold model was integrated into a previously developed single neuron model and resulted in a 33% improvement in spike prediction.

Determination of relevant neuron-neuron connections for neural prosthetics using time-delayed mutual information: tutorial and preliminary results

BACKGROUND

Identification of functional dependence among neurons is a necessary component in both the rational design of neural prostheses as well as in the characterization of network physiology. The objective of this article is to provide a tutorial for neurosurgeons regarding information theory, specifically time-delayed mutual information, and to compare time-delayed mutual information, an information theoretic quantity based on statistical dependence, with cross-correlation, a commonly used metric for this task in a preliminary analysis of rat hippocampal neurons.

METHODS

Spike trains were recorded from rats performing delayed nonmatch-to-sample task using an array of electrodes surgically implanted into the hippocampus of each hemisphere of the brain. In addition, spike train simulations of positively correlated neurons, negatively correlated neurons, and neurons correlated by nonlinear functions were generated. These were evaluated by time-delayed mutual information (MI) and cross-correlation.

RESULTS

Application of time-delayed MI to experimental data indicated the optimal bin size for information capture in the CA3-CA1 system was 40 ms, which may provide some insight into the spatiotemporal nature of encoding in the rat hippocampus. On simulated data, time-delayed MI showed peak values at appropriate time lags in positively correlated, negatively correlated, and complexly correlated data. Cross-correlation showed peak and troughs with positively correlated and negatively correlated data, but failed to capture some higher order correlations.

CONCLUSIONS

Comparison of time-delayed MI to cross-correlation in identification of functionally dependent neurons indicates that the methods are not equivalent. Time-delayed MI appeared to capture some interactions between CA3-CA1 neurons at physiologically plausible time delays missed by cross-correlation. It should be considered as a method for identification of functional dependence between neurons and may be useful in the development of neural prosthetics.

A cortical neural prosthesis for restoring and enhancing memory

A primary objective in developing a neural prosthesis is to replace neural circuitry in the brain that no longer functions appropriately. Such a goal requires artificial reconstruction of neuron-to-neuron connections in a way that can be recognized by the remaining normal circuitry, and that promotes appropriate interaction. In this study, the application of a specially designed neural prosthesis using a multi-input/multi-output (MIMO) nonlinear model is demonstrated by using trains of electrical stimulation pulses to substitute for MIMO model derived ensemble firing patterns. Ensembles of CA3 and CA1 hippocampal neurons, recorded from rats performing a delayed-nonmatch-to-sample (DNMS) memory task, exhibited successful encoding of trial-specific sample lever information in the form of different spatiotemporal firing patterns. MIMO patterns, identified online and in real-time, were employed within a closed-loop behavioral paradigm. Results showed that the model was able to predict successful performance on the same trial. Also, MIMO model-derived patterns, delivered as electrical stimulation to the same electrodes, improved performance under normal testing conditions and, more importantly, were capable of recovering performance when delivered to animals with ensemble hippocampal activity compromised by pharmacologic blockade of synaptic transmission. These integrated experimental-modeling studies show for the first time that, with sufficient information about the neural coding of memories, a neural prosthesis capable of real-time diagnosis and manipulation of the encoding process can restore and even enhance cognitive, mnemonic processes.

Integrated multiscale modeling of the nervous system: predicting changes in hippocampal network activity by a positive AMPA receptor modulator

One of the fundamental characteristics of the brain is its hierarchical organization. Scales in both space and time that must be considered when integrating across hierarchies of the nervous system are sufficiently great as to have impeded the development of routine multilevel modeling methodologies. Complex molecular interactions at the level of receptors and channels regulate activity at the level of neurons; interactions between multiple populations of neurons ultimately give rise to complex neural systems function and behavior. This spatial complexity takes place in the context of a composite temporal integration of multiple, different events unfolding at the millisecond, second, minute, hour, and longer time scales. In this study, we present a multiscale modeling methodology that integrates synaptic models into single neuron, and multineuron, network models. We have applied this approach to the specific problem of how changes at the level of kinetic parameters of a receptor-channel model are translated into changes in the temporal firing pattern of a single neuron, and ultimately, changes in the spatiotemporal activity of a network of neurons. These results demonstrate how this powerful methodology can be applied to understand the effects of a given local process within multiple hierarchical levels of the nervous system.

Changes of hippocampal CA3-CA1 population nonlinear dynamics across different training sessions in rats performing a memory-dependent task

Delayed-nonmatch-to-sample (DNMS) task is memory-dependent. Hippocampal CA3 and CA1 cells were shown to be encoding the required spatial and temporal information to complete this task. In order to identify possible changes in neural population nonlinear dynamics during learning of the DNMS task, we have first modeled the input-output transformation of spike trains across brain subregions from learning animals using a multiple-input, multiple-output (MIMO) nonlinear dynamic model. The feedforward and feedback kernels describing the relations between hippocampal CA3 and CA1 subregions have shown significant changes at different training sessions.

Nonlinear dynamic modeling of synaptically driven single hippocampal neuron intracellular activity

A high-order nonlinear dynamic model of the input-output properties of single hippocampal CA1 pyramidal neurons was developed based on synaptically driven intracellular activity. The purpose of this study is to construct a model that: 1) can capture the nonlinear dynamics of both subthreshold activities [postsynaptic potentials (PSPs)] and suprathreshold activities (action potentials) in a single formalism; 2) is sufficiently general to be applied to any spike-input and spike-output neurons (point process input and point process output neural systems); and 3) is computationally efficient. The model consisted of three major components: 1) feedforward kernels (up to third order) that transform presynaptic action potentials into PSPs; 2) a constant threshold, above which action potentials are generated; and 3) a feedback kernel (first order) that describes spike-triggered after-potentials. The model was applied to CA1 pyramidal cells, as they were electrically stimulated with broadband Poisson random impulse trains through the Schaffer collaterals. The random impulse trains used here have physiological properties similar to spiking patterns observed in CA3 hippocampal neurons. PSPs and action potentials were recorded from the soma of CA1 pyramidal neurons using whole-cell patch-clamp recording. We evaluated the model performance separately with respect to PSP waveforms and the occurrence of spikes. The average normalized mean square error of PSP prediction is 14.4%. The average spike prediction error rate is 18.8%. In summary, although prediction errors still could be reduced, the model successfully captures the majority of high-order nonlinear dynamics of the single-neuron intracellular activity. The model captures the general biophysical processes with a small set of open parameters that are directly constrained by the intracellular recording, and thus, can be easily applied to any spike-input and spike-output neuron.

Dynamic nonlinear modeling of interactions between neuronal ensembles using principal dynamic modes

We present a novel methodology for modeling the interactions between neuronal ensembles that utilizes the concept of Principal Dynamic Modes (PDM) and their associated nonlinear functions (ANF). This new approach seeks to reduce the complexity of the multi-input/multi-output (MIMO) model of the interactions between neuronal ensembles--an issue of critical practical importance in scaling up the MIMO models to incorporate hundreds (or even thousands) of input-output neurons. Global PDMs were extracted from the data using estimated first-order and second-order kernels and singular value decomposition (SVD). These global PDMs represent an efficient "coordinate system" for the representation of the MIMO model. The ANFs of the PDMs are estimated from the histograms of the combinations of PDM output values that lead to output spikes. For initial testing and validation of this approach, we applied it to a set of data collected at the pre-frontal cortex of a non-human primate during a behavioral task (Delayed Match-to-Sample). Recorded spike trains from Layer-2 neurons were viewed as the "inputs" and from Layer-5 neurons as the outputs. Model prediction performance was evaluated by means of computed Receiver Operating Characteristic (ROC) curves. The results indicate that this methodology may greatly reduce the complexity of the MIMO model without significant degradation of performance.

A hardware-based computational platform for Generalized Laguerre-Volterra MIMO model for neural activities

A parallelized and pipelined architecture based on FPGA and a higher-level Self Reconfiguration Platform are proposed in this paper to model Generalized Laguerre-Volterra MIMO system essential in identifying the time-varying neural dynamics underlying spike activities. Our proposed design is based on the Xilinx Virtex-6 FPGA platform and the processing core can produce data samples at a speed of 1.33 × 10(6)/s, which is 3.1 × 10(3) times faster than the corresponding C model running on an Intel i7-860 Quad Core Processor. The ongoing work of the construction of the advanced Self Reconfiguration Platform is presented and initial test results are provided.

Tracking the changes of hippocampal population nonlinear dynamics in rats learning a memory-dependent task

Neurobiological processes associated with learning are known to be highly nonlinear, dynamical, and time-varying. Characterizing the time-varying functional input-output properties of neural systems is a critical step to understand the neurobiological basis of learning. In this paper, we present a study on tracking of the changes of neural dynamics in rat hippocampus during learning of a memory-dependent delayed nonmatch-to-sample (DNMS) task. The rats were first trained to perform the DNMS task without a delay between the sample and response events. After reaching a performance level, they were subjected to the DNMS task with variable delays with a 5s mean duration. Spike trains were recorded from hippocampal CA3 (input) and CA1 (output) regions during all training sessions and constitute the input-output data for modeling. We applied the time-varying Generalized Laguerre-Volterra Model to study the changes of the CA3-CA1 nonlinear dynamics using these data. Result showed significant changes in the Volterra kernels after the introduction of delays. This result suggests that the CA3-CA1 nonlinear dynamics established in the initial training sessions underwent a functional reorganization as animals were learning to perform the task that now requires delays.

Estimation and statistical validation of event-invariant nonlinear dynamic models of hippocampal CA3-CA1 population activities

To develop hippocampal prosthetic devices that can restore the memory-dependent cognitive functions lost in diseases or injuries, it is essential to build a computational model that sufficiently captures the transformations of multiple memories performed by hippocampal sub-regions. A universal model with a single set of coefficients for all memories is desirable, since it can transform the memories without explicitly knowing what those memories represent and thus avoids switching between multiple models for multiple memories in implementation. In this study, we test the feasibility of such universal models of hippocampal CA3-CA1 by estimating the multi-input, multi-output (MEMO) nonlinear dynamic models using input (CA3) and output (CA1) spike trains recorded during multiple behavioral events representing multiple memories from rats performing a delayed nonmatch-to-sample task. We further statistically evaluated the model performances of the MEMO models on the different events. Results show that the models accurately replicate the output spike patterns during those events, and thus can be used as event-invariant nonlinear dynamic models that continuously predict the ongoing CA1 spatio-temporal patterns as the ongoing CA3 spatio-temporal patterns unfold.

A two-stage cascade nonlinear dynamical model of single neurons for the separation and quantification of pre- and post-synaptic mechanisms of synaptic transmission

Neurons receive pre-synaptic spike trains and transform them into post-synaptic spike trains. This spike train to spike train temporal transformation underlies all cognitive functions performed by neurons, e.g., learning and memory. The transformation is a highly nonlinear dynamical process that involves both pre- and post-synaptic mechanisms. The ability to separate and quantify the nonlinear dynamics of pre- and post-synaptic mechanism is needed to gain insights into this transformation. In this study, we developed a Volterra kernel based two-stage cascade model of synaptic transmission using synaptically-driven intracellular activities, to which broadband stimulation conditions were imposed. The first stage of the model represents the pre-synaptic mechanisms and describes the nonlinear dynamical transformation from pre-synaptic spike trains to transmitter vesicle release strengths. The vesicle release strengths were obtained from the intracellularly recorded excitatory post-synaptic currents (EPSCs). The second stage of the model represents the post-synaptic mechanisms and describes the nonlinear dynamical transformation from release strengths to excitatory post-synaptic potentials (EPSPs). One application of this cascade model is to analyze the pre- and post-synaptic mechanism change induced by long-term potentiation (LTP). This future application is expected to shed new light on the expression locus of LTP.

Computational studies of NMDA receptors: differential effects of neuronal activity on efficacy of competitive and non-competitive antagonists

N-Methyl-D-Aspartate receptors (NMDARs) play important physiological as well as pathological roles in the central nervous system (CNS). While NMDAR competitive antagonists, such as D-2-Amino-5-Phosphopentanoic acid (AP5) have been shown to impair learning and memory, the non-competitive antagonist, memantine, is paradoxically beneficial in mild to moderate Alzheimer's disease (AD) patients. It has been proposed that differences in kinetic properties could account for antagonist functional differences. Here we present a new elaborated kinetic model of NMDARs that incorporates binding sites for the agonist (glutamate) and co-agonist (glycine), channel blockers, such as memantine and magnesium (Mg(2+)), as well as competitive antagonists. We first validated and optimized the parameters used in the model by comparing simulated results with a wide range of experimental data from the literature. We then evaluated the effects of stimulation frequency and membrane potential (Vm) on the characteristics of AP5 and memantine inhibition of NMDARs. Our results indicated that the inhibitory effects of AP5 were independent of Vm but decreased with increasing stimulation frequency. In contrast, memantine inhibitory effects decreased with both increasing Vm and stimulation frequency. They support the idea that memantine could provide tonic blockade of NMDARs under basal stimulation conditions without blocking their activation during learning. Moreover they underline the necessity of considering receptor kinetics and the value of the biosimulation approach to better understand mechanisms of drug action and to identify new ways of regulating receptor function.

Noise-robust acoustic signature recognition using nonlinear Hebbian learning

We propose using a new biologically inspired approach, nonlinear Hebbian learning (NHL), to implement acoustic signal recognition in noisy environments. The proposed learning processes both spectral and temporal features of input acoustic data. The spectral analysis is realized by using auditory gammatone filterbanks. The temporal dynamics is addressed by analyzing gammatone-filtered feature vectors over multiple temporal frames, which is called a spectro-temporal representation (STR). Given STR features, the exact acoustic signatures of signals of interest and the mixing property between signals of interest and noises are generally unknown. The nonlinear Hebbian learning is then employed to extract representative independent features from STRs, and to reduce their dimensionality. The extracted independent features of signals of interest are called signatures. In the meantime of learning, the synaptic weight vectors between input and output neurons are adaptively updated. These weight vectors project data into a feature subspace, in which signals of interest are selected, while noises are attenuated. Compared with linear Hebbian learning (LHL) which explores the second-order moment of data, the applied NHL involves the higher-order statistics of data. Therefore, NHL can capture representative features that are more statistically independent than LHL can. Besides, the nonlinear activation function of NHL can be chosen to refer to the implicit distribution of many acoustic sounds, and thus making the learning optimized in an aspect of mutual information. Simulation results show that the whole proposed system can more accurately recognize signals of interest than other conventional methods in severely noisy circumstances. One applicable project is detecting moving vehicles. Noise-contaminated vehicle sound is recognized while other non-vehicle sounds are rejected. When vehicle is contaminated by human vowel, bird chirp, or additive white Gaussian noise (AWGN) at SNR=0 dB, the proposed system dramatically decreases the error rate over normally used acoustic feature extraction method, mel-frequency cepstral computation (MFCC), by 26%, 36.3%, and 60.3%, respectively; and, over LHL by 20%, 2.3%, and 15.3%, respectively. Another applicable project is vehicle type identification. The proposed system achieves better performance than LHL, e.g., 40% improvement when gasoline heavy wheeled car is contaminated by AWGN at SNR=5 dB. More importantly, the proposed system is implemented in real-time field testing for months. The purpose is to detect vehicle with any make or model moving on the street with speed 10-35 mph. The missing rate is 1-2%, when vehicle is contaminated by any surrounding noises (human conversation, animal sound, airplane, wind, etc.) at SNR=0-20 dB. The false alarm rate is around 1%. To summarize, this study not only provides an efficient approach to extract representative independent features from high-dimensional data, but also offers robustness against severe noises.

Modeling of nonlinear nonstationary dynamic systems with a novel class of artificial neural networks

This paper introduces a novel neural-network architecture that can be used to model time-varying Volterra systems from input-output data. The Volterra systems constitute a very broad class of stable nonlinear dynamic systems that can be extended to cover nonstationary (time-varying) cases. This novel architecture is composed of parallel subnets of three-layer perceptrons with polynomial activation functions, with the output of each subnet modulated by an appropriate time function that gives the summative output its time-varying characteristics. The paper shows the equivalence between this network architecture and the class of time-varying Volterra systems, and demonstrates the range of applicability of this approach with computer-simulated examples and real data. Although certain types of nonstationarities may not be amenable to this approach, it is hoped that this methodology will provide the practical tools for modeling some broad classes of nonlinear, nonstationary systems from input-output data, thus advancing the state of the art in a problem area that is widely viewed as a daunting challenge.

Sparse generalized Laguerre-Volterra model of neural population dynamics

To understand the function of a brain region, e.g., hippocampus, it is necessary to model its input-output property. Such a model can serve as the computational basis of the development of cortical prostheses restoring the transformation of population neural activities performed by the brain region. We formulate a sparse generalized Laguerre-Volterra model (SGLVM) for the multiple-input, multiple-output (MIMO) transformation of spike trains. A SGLVM consists of a set of feedforward Laguerre-Volterra kernels, a feedback Laguerre-Volterra kernel, and a probit link function. The sparse model representation involving only significant self and cross terms is achieved through statistical model selection and cross-validation methods. The SGLVM is applied successfully to the hippocampal CA3-CA1 population dynamics.