Neuron Transfer Function Research Articles

Evaluation of mechanical property of banana HDPE composite through experiment and artificial neural network (ANN)

In this study, two post-curing techniques using microwave and oven were utilized to examine the mechanical characteristics and behavior of composites reinforced with banana fibers with its matrix made of high-density polyethylene polymer. The research involved experimental work through tensile tests and predictive modelling using the artificial neural network (ANN). The first part of the research involved the collection of datasets which were gathered from the tensile tests based on the ASTM D638 using dog bone specimens of various fiber percentages to compare the mechanical properties such as the specific energy and maximum load after post-curing using an oven and microwave at various powers and time. The trend and behavior of the composite due to the fiber percentages and the effects of post curing methods were analyzed. The second part of the research involved the ANN approach to predict the maximum gage length based on the data collected from the experiments using inputs such as the fiber content, maximum load, and displacement. Further optimization was done on the ANN model by fine-tuning variables such as the learning function, transfer function, and number of neurons to predict the stress and strain values. The accuracy of the results was calculated using statistical evaluation equations. The optimized settings were utilized to generate the predicted results and the deviation from the experimental results were evaluated.

Artificial Intelligence and Machine Learning in Dialysis: Ready for Prime Time?

Artificial Intelligence and Machine Learning in Dialysis: Ready for Prime Time?

Bandlimited Signal Reconstruction From Leaky Integrate-and-Fire Encoding Using POCS

Leaky integrate-and-fire (LIF) encoding is a model of neuron transfer function in biology that has recently attracted the attention of the signal processing and neuromorphic computing communities as a technique of event-based sampling for data acquisition. While LIF enables the implementation of analog-circuit signal samplers of lower complexity and higher accuracy simultaneously, the core difficulty of this technique is the retrieval of an input from its LIF-encoded output. In this article, we study this problem in the context of bandlimited inputs, by extracting the most abstract features of an LIF encoder as a generalized nonuniform sampler. In this view, the LIF output is seen as the transformation of the input by a known linear operator. We show that the signal reconstruction method of projection onto convex sets (POCS) converges to a weighted pseudo-inverse of this operator. This allows perfect recovery under uniqueness of reconstruction, minimum-norm reconstruction under incomplete sampling, as well as a noise shaping of time quantization that outperforms standard pseudo-inversion. On the practical side, a single iteration of the POCS method can be used to improve <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">any</i> estimate whose LIF samples are not consistent with those of the input, and a rigorous discrete-time implementation of this iteration is proposed that does not require a Nyquist-rate representation of the signals.

Combination of LF‐NMR and BP‐ANN to monitor the moisture content of rice during hot‐air drying

AbstractIn this study, a rapid real‐time nondestructive method for detecting the moisture content of rice during hot‐air drying was investigated. Intelligent techniques of low‐field nuclear magnetic resonance (LF‐NMR) and back propagation artificial neural network (BP‐ANN) were applied to monitor the moisture content of rice. The effect of different hot‐air temperatures (35, 45, 55, and 65°C) on the moisture content and water migration within rice was studied. The results showed that the drying temperature promoted the diffusion and transfer of water within the rice, and was positively proportional to the drying rate. The binding energy of the different states of water within rice increased with the drying process, and the variation in relaxation time and peak area was consistent for each stage at different temperatures. In addition, the amount of LF‐NMR signals was used as an indicator to build a predictive model for the moisture content of rice during hot‐air drying. A BP‐ANN prediction model optimized by transfer function, training function and number of neurons was used to monitor the moisture content of rice using the amount of LF‐NMR signals of different states of water as input variables. The optimized neural network model had the excellent predictive ability with an MSE of 6.02 × 10−6 and R2 of 0.996. These results provide a reference for combining LF‐NMR and BP‐ANN in the application of intelligent online monitoring of hot‐air drying of rice.Practical ApplicationsThe monitoring of moisture content during hot‐air drying of rice is an essential parameter for optimizing the drying process. The combined approach of LF‐NMR and BP‐ANN for rapid real‐time nondestructive monitoring is well suited to hot‐air drying of rice, allowing for improved product quality and operational processes. In addition, the model developed in this study has the good predictive performance to meet the current industry and production needs, providing new research ideas and technical references for the optimization of the drying process of rice.

Field programmable gate array implementation of an intelligent soft calibration technique for linear variable differential transformers

Background: Displacement is often used as an indirect indicator for monitoring multiple parameters, i.e. force, velocity, acceleration, and weight, making it an important variable for the measurement and control of processes. Sensors such as Linear Variable Differential Transformers (LVDTs) play a primary role in the design of any displacement measuring instrument. Calibration of an instrument is carried out to produce accurate results from the measuring instrument. Methods: The objective of this study is to calibrate the output of LVDT by designing a signal conditioning circuit so as to extend the linearity range of the sensor to 100% of the full scale input range, and also allows the measurement technique to adapt to variations in the physical parameters of the LVDT, the supply frequency, and the temperature. An optimized neural network is trained to produce linear and adaptive output from the raw data obtained from LVDT. Optimization is achieved by choosing the best neural network algorithm, number of hidden layers and transfer function of neurons which produce the least mean square error. The optimized neural network algorithm is implemented on a Field Programmable Gate Array (FPGA) chip for testing and validation in real life. Results: Experimental results show that the proposed technique was able to extend the linearity of LVDT and make the output adaptive for variations in physical parameters of LVDT, supply frequency and temperature. Conclusions: Accurate measurement of displacement is essential in many process applications, and a good calibration technique is required to produce accurate measurement. The presented calibration technique using optimized neural network algorithms has produced reliable measurements as desired.

Effects of dendritic properties on spike train correlations in biophysically-based model neurons

A large number of studies have shown that the correlation of spike trains in paired neurons is a key indicator of functional connectivity. Acting as the important components of neurons, dendrites affect the generation of spike train and the transmission of signals by changing the transfer function of neurons. However, it is not clear how the intrinsic properties of dendrites influence the degree of correlation of neuronal spike trains. This paper used a biophysically-based two-compartment model to simulate the firing output of neurons that receive correlated input currents, quantify the relationship between the active and passive properties of dendrites and the pairwise spike train correlations. We found that increasing the proportion of dendritic area or the coupling conductance between two compartments increases the output correlation. It is also shown that decreasing the intracellular calcium concentration and the conductance of calcium activated outward potassium channel or increasing the conductance of sodium and potassium channel in dendritic compartment increases the output correlation. This results from the increase of the number of dendritic spikes due to changing dendritic intrinsic properties, which causes more internal current flowing into the soma from dendrites, thereby leading to an increase of somatic firing rate. Our results reveal the influence of the intrinsic characteristics of dendrites on firing output of individual neuron, which further influences the correlation between the spike trains of a pairs or population of neurons. This study is of great significance for understanding how neuronal intrinsic properties affect neural network coding.

Individual Travel Knowledge Graph-Based Public Transport Commuter Identification: A Mixed Data Learning Approach

Commuters are the stable travel group for the public transportation (PT) service system. Accurately identifying the PT commuters is conducive to promoting PT service quality and development of urban sustainable transportation. This paper extracts individual PT travel chain information and constructs individual travel knowledge graphs of PT passengers based on the association matching algorithm and the theory of multilayer planning. A mixed dataset is formed by associating individual travel chains with travel survey data. Seven travel characteristic indicators regarding travel performance and spatiotemporal travel characteristics are extracted. The identification model of PT commuters is developed based on a three-layer backpropagation neural network (BPNN). The optimal model structure of neuron node number, transfer function, and learning rate are discussed quantitatively according to the minimization of model errors. The evaluation indexes of overall accuracy and kappa coefficient of the constructed model are 94.5% and 87.9% separately. The results indicate that the model identification accuracy is acceptable, and the proposed characteristic indicators and systematic modelling procedure are effective. Then, the model performance is compared with the other five machine learning models further. The results confirm that the proposed model has a better identification accuracy and viability, and the model performance will improve with the increase of the sample size.

Diagnostic Data Integration Using Deep Neural Networks for Real-Time Plasma Analysis

Recent advances in acquisition equipment are providing experiments with growing amounts of precise, yet affordable sensors. At the same time, an improved computational power, coming from new hardware resources [GPU, field-programmable gate array (FPGA), adaptive compute acceleration platform (ACAP)] has been made available at relatively low costs. This led us to explore the possibility of completely renewing the chain of acquisition for a fusion experiment, where many high-rate sources of data, coming from different diagnostics, can be combined in a wide framework of algorithms. If, on the one hand, adding new data sources with different diagnostics enriches our knowledge about physical aspects, on the other hand, the dimensions of the overall model grow, making relations among variables more and more opaque. A new approach for integrating such heterogeneous diagnostics, based on the composition of deep variational autoencoders, could ease this problem, acting as a structural sparse regularizer. This has been applied to RFX-mod experimental data, integrating the soft X-ray linear images of plasma temperature with the magnetic state. However, to ensure a real-time signal analysis, these algorithmic techniques must be adapted to run in well-suited hardware. In particular, it is shown that, attempting a quantization of neuron transfer functions, such models can be adapted to run in an embedded programmable logic device. The resulting firmware, approximating the deep inference model to a set of simple operations, fits well with the simple logic units that are largely abundant in FPGAs. This is the key factor that permits the use of affordable hardware with complex deep neural topology and operates them in real-time.

A Framework for Analyzing, Designing, and Visualizing Spiking Neural Networks Part I: Linear Response Surfaces

In this first of two closely related papers, we set the foundation for a new framework, called Response Surfaces (RSs), to address fundamental problems of analyzing, designing, and visualizing spiking neurons and networks. An RS is a plot of the direct transfer function between input and output firing times of a spiking neuron and shows all the patterns of input spike times that fire a neuron at a given time. Thus, an RS is a graphical tool for visualizing, analyzing, and designing spiking neural networks. In this paper, we develop a linear RS framework based on triangular, post-synaptic-potential waveforms and apply it to the following problems: graphing the transfer function of a linear spiking neuron, designing an efficient spiking-XOR gate, analyzing phase tracking, and calculating spike times of arbitrarily large recurrent spiking networks. As another fundamental result of the linear RS framework, we show that the output firing time of a spiking neuron is equivalent to the center-of-mass of input spike times (acting as positions) and weights (acting as masses) plus a fixed delay. In the second paper, we explore the possibilities of more biologically realistic post-synaptic-potential waveforms and create a nonlinear RS framework as an extension of the linear RS framework presented here. Although our application examples in both papers focus on design and analysis, we touch on why the RS framework is helpful for understanding the effects of weight changes in learning algorithms.

A Framework for Analyzing, Designing, and Visualizing Spiking Neural Networks—Part II: Nonlinear Response Surfaces

In this second part of a two-part study, we extend to nonlinear synaptic responses a new framework, called Response Surfaces (RSs), for analyzing, designing, and visualizing spiking neurons and networks. An RS is the transfer function between input and output firing times of a spiking neuron and shows all the patterns of input spike times that fire a neuron at a given time. Here in Part II, we present four mathematically tractable functions to model more realistic post-synaptic-potential waveforms, and we build on the linear RS framework of Part I to create a nonlinear RS framework. We then discuss the qualitative differences between the linear and nonlinear RSs frameworks and revisit problems of Part I using the nonlinear RSs framework: graphing the transfer function of a nonlinear spiking neuron, designing an efficient spiking-XOR gate, analyzing phase-tracking in recurrent SNNs, and developing transfer functions for calculating the output firing times of a spiking neuron given a set of input firing times. For the nonlinear RS framework, we show that the output firing time of a nonlinear spiking neuron is equivalent to the center-of-mass of input spike times (acting as positions) and weights (acting as masses) plus a fixed delay plus a second-order correction. The second-order centroid correction is one of the main differences between the linear and nonlinear RS frameworks. For recurrent networks, the nonlinear RS framework also shows a more stable behavior compared to linear SNNs, owing to the smooth shapes of nonlinear post-synaptic-potential waveforms.

High-Order Neural Networks are Equivalent to Ordinary Neural Networks

Equivalence of computational systems can assist in obtaining abstract systems, and thus enable better understanding of issues related their design and performance. For more than four decades, artificial neural networks have been used in many scientific applications to solve classification problems as well as other problems. Since the time of their introduction, multilayer feedforward neural network referred as Ordinary Neural Network (ONN), that contains only summation activation (Sigma) neurons, and multilayer feedforward High-order Neural Network (HONN), that contains Sigma neurons, and product activation (Pi) neurons, have been treated in the literature as different entities. In this work, we studied whether HONNs are mathematically equivalent to ONNs. We have proved that every HONN could be converted to some equivalent ONN. In most cases, one just needs to modify the neuronal transfer function of the Pi neuron to convert it to a Sigma neuron. The theorems that we have derived clearly show that the original HONN and its corresponding equivalent ONN would give exactly the same output, which means; they can both be used to perform exactly the same functionality. We also derived equivalence theorems for several other non-standard neural networks, for example, recurrent HONNs and HONNs with translated multiplicative neurons. This work rejects the hypothesis that HONNs and ONNs are different entities, a conclusion that might initiate a new research frontier in artificial neural network research.

A non-negative artificial neural network configuration for noise predictions

Artificial Neural Networks (ANN) provide an accurate relationship between user-defined inputs and outputs for complex problems. Their basic units are called neurons. The computation that takes place in each neuron is a weighted summation of its inputs. These results are then implemented as an input to an activation or transfer function, which in turn defines the final neuron’s output. The transfer function configuration is very important when defining the final purpose of the ANN. There are many standard activation functions like pure linear, tangent sigmoid, logistic sigmoid, among others. In this study, an ANN capable of predicting tire pavement interaction noise (TPIN) is constructed. Its outputs correspond to sound pressure in narrowband. Thus, the main requirement is that the outputs must be positive. Several neuron transfer functions were investigated for both hidden and output layers. Finally, a customized hybrid linear-sigmoid transfer function was selected. The final ANN configuration is able to capture typical TPIN spectral content within 400 Hz–1600 Hz. In addition, the error is significantly reduced if compared to those obtained after implementing standard transfer functions.

Modeling mesoscopic cortical dynamics using a mean-field model of conductance-based networks of adaptive exponential integrate-and-fire neurons.

Voltage-sensitive dye imaging (VSDi) has revealed fundamental properties of neocortical processing at macroscopic scales. Since for each pixel VSDi signals report the average membrane potential over hundreds of neurons, it seems natural to use a mean-field formalism to model such signals. Here, we present a mean-field model of networks of Adaptive Exponential (AdEx) integrate-and-fire neurons, with conductance-based synaptic interactions. We study a network of regular-spiking (RS) excitatory neurons and fast-spiking (FS) inhibitory neurons. We use a Master Equation formalism, together with a semi-analytic approach to the transfer function of AdEx neurons to describe the average dynamics of the coupled populations. We compare the predictions of this mean-field model to simulated networks of RS-FS cells, first at the level of the spontaneous activity of the network, which is well predicted by the analytical description. Second, we investigate the response of the network to time-varying external input, and show that the mean-field model predicts the response time course of the population. Finally, to model VSDi signals, we consider a one-dimensional ring model made of interconnected RS-FS mean-field units. We found that this model can reproduce the spatio-temporal patterns seen in VSDi of awake monkey visual cortex as a response to local and transient visual stimuli. Conversely, we show that the model allows one to infer physiological parameters from the experimentally-recorded spatio-temporal patterns.

Synchronous spikes are necessary but not sufficient for a synchrony code in populations of spiking neurons

Synchronous activity in populations of neurons potentially encodes special stimulus features. Selective readout of either synchronous or asynchronous activity allows formation of two streams of information processing. Theoretical work predicts that such a synchrony code is a fundamental feature of populations of spiking neurons if they operate in specific noise and stimulus regimes. Here we experimentally test the theoretical predictions by quantifying and comparing neuronal response properties in tuberous and ampullary electroreceptor afferents of the weakly electric fish Apteronotus leptorhynchus These related systems show similar levels of synchronous activity, but only in the more irregularly firing tuberous afferents a synchrony code is established, whereas in the more regularly firing ampullary afferents it is not. The mere existence of synchronous activity is thus not sufficient for a synchrony code. Single-cell features such as the irregularity of spiking and the frequency dependence of the neuron's transfer function determine whether synchronous spikes possess a distinct meaning for the encoding of time-dependent signals.

Integrative properties and transfer function of cortical neurons initiating absence seizures in a rat genetic model.

Absence seizures are accompanied by spike-and-wave discharges in cortical electroencephalograms. These complex paroxysmal activities, affecting the thalamocortical networks, profoundly alter cognitive performances and preclude conscious perception. Here, using a well-recognized genetic model of absence epilepsy, we investigated in vivo how information processing was impaired in the ictogenic neurons, i.e. the population of cortical neurons responsible for seizure initiation. In between seizures, ictogenic neurons were more prone to generate bursting activity and their firing response to weak depolarizing events was considerably facilitated compared to control neurons. In the course of seizures, information processing became unstable in ictogenic cells, alternating between an increased and a decreased responsiveness to excitatory inputs, depending on the spike and wave patterns. The state-dependent modulation in the excitability of ictogenic neurons affects their inter-seizure transfer function and their time-to-time responsiveness to incoming inputs during absences. Epileptic seizures result from aberrant cellular and/or synaptic properties that can alter the capacity of neurons to integrate and relay information. During absence seizures, spike-and-wave discharges (SWDs) interfere with incoming sensory inputs and preclude conscious experience. The Genetic Absence Epilepsy Rats from Strasbourg (GAERS), a well-established animal model of absence epilepsy, allows exploration of the cellular basis of this impaired information processing. Here, by combining in vivo electrocorticographic and intracellular recordings from GAERS and control animals, we investigated how the pro-ictogenic properties of seizure-initiating cortical neurons modify their integrative properties and input-output operation during inter-ictal periods and during the spike (S-) and wave (W-) cortical patterns alternating during seizures. In addition to a sustained depolarization and an excessive firing rate in between seizures, ictogenic neurons exhibited a pronounced hyperpolarization-activated depolarization compared to homotypic control neurons. Firing frequency versus injected current relations indicated an increased sensitivity of GAERS cells to weak excitatory inputs, without modifications in the trial-to-trial variability of current-induced firing. During SWDs, the W-component resulted in paradoxical effects in ictogenic neurons, associating an increased membrane input resistance with a reduction in the current-evoked firing responses. Conversely, the collapse of cell membrane resistance during the S-component was accompanied by an elevated current-evoked firing relative to W-sequences, which remained, however, lower compared to inter-ictal periods. These findings show a dynamic modulation of ictogenic neurons' intrinsic properties that may alter inter-seizure cortical function and participate in compromising information processing in cortical networks during absences.

The mechanics of state-dependent neural correlations.

Simultaneous recordings from large neural populations are becoming increasingly common. An important feature of population activity is the trial-to-trial correlated fluctuation of spike train outputs from recorded neuron pairs. Similar to the firing rate of single neurons, correlated activity can be modulated by a number of factors, from changes in arousal and attentional state to learning and task engagement. However, the physiological mechanisms that underlie these changes are not fully understood. We review recent theoretical results that identify three separate mechanisms that modulate spike train correlations: changes in input correlations, internal fluctuations and the transfer function of single neurons. We first examine these mechanisms in feedforward pathways and then show how the same approach can explain the modulation of correlations in recurrent networks. Such mechanistic constraints on the modulation of population activity will be important in statistical analyses of high-dimensional neural data.

Time Series Forecasting with Neural Networks and Choquet Integral

The objective of this paper is to compare time series forecasting by using three different backpropagation neural networks. A daily time series of Vale Company in the period March 1, 2000-June 10, 2006 is used as a reference against which results from other forecasts are compared. Three types of backpropagation neural networks are constructed with different input layers: the first among those makes use of the real time series data; the second uses the normalized real time series data; and the third uses the normalized real time series data and the Choquet integral in order to fuzzify the input layer. In all of the three backpropagation neural networks a hidden layer with tangent sigmoid transfer function and different numbers of neurons are used. In the output of the three neural networks a linear transfer function with one neuron for obtaining a linear equation after their training is used. The forecasting equations obtained for each neural networks are used with outsample data to forecast Vale's time series data and compare against real data. On the basis of the obtained results we conclude that the use of Choquet integral in this context is powerful enough so that its use must be recommended.

Stability of discrete time recurrent neural networks and nonlinear optimization problems

We consider the method of Reduction of Dissipativity Domain to prove global Lyapunov stability of Discrete Time Recurrent Neural Networks. The standard and advanced criteria for Absolute Stability of these essentially nonlinear systems produce rather weak results. The method mentioned above is proved to be more powerful. It involves a multi-step procedure with maximization of special nonconvex functions over polytopes on every step. We derive conditions which guarantee an existence of at most one point of local maximum for such functions over every hyperplane. This nontrivial result is valid for wide range of neuron transfer functions.

Prediction of carbon dioxide loading capacity in amino acid salt solutions as new absorbents using artificial neural network and Deshmukh–Mather models

The aim of this work is solubility prediction of carbon dioxide (CO2) in amino acid salt solutions as new absorbents over wide ranges of operating conditions, utilizing Artificial Neural Network (ANN) and Deshmukh–Mather models. pH of solutions, overall molar concentration, partial pressure of CO2, apparent molecular weight and temperature was picked as input variables of the proposed ANN. A group of 1364 literature experimental data points for CO2 solubility have been congregated from the literature to build the network. The best architecture of the developed ANN including the numbers of hidden layer, transfer function and number of neurons were attained by utilizing these literature data points. Also CO2 solubility in amino acid salt solution was modeled using Deshmukh–Mather model. Results show that proposed ANN has better performance compared to Deshmukh–Mather model. The ANN was trained by the Levenberg–Marquardt back-propagation algorithm including two hidden layers with 8 and 7 neurons and Tan-sigmoid transfer function for the hidden and output layers. The model results show that proposed ANN that developed with amino acid salt solutions data points has ability to predict accurately the solubility of CO2 and H2S in dissimilar solutions with correlation coefficient (R2) of 0.9982 and Average Relative Deviation (ARD) value of 3.2976.

An Empirical Model for Reliable Spiking Activity.

Understanding a neuron's transfer function, which relates a neuron's inputs to its outputs, is essential for understanding the computational role of single neurons. Recently, statistical models, based on point processes and using generalized linear model (GLM) technology, have been widely applied to predict dynamic neuronal transfer functions. However, the standard version of these models fails to capture important features of neural activity, such as responses to stimuli that elicit highly reliable trial-to-trial spiking. Here, we consider a generalization of the usual GLM that incorporates nonlinearity by modeling reliable and nonreliable spikes as being generated by distinct stimulus features. We develop and apply these models to spike trains from olfactory bulb mitral cells recorded in vitro. We find that spike generation in these neurons is better modeled when reliable and unreliable spikes are considered separately and that this effect is most pronounced for neurons with a large number of both reliable and unreliable spikes.