Neural Computation Research Articles

Levenberg–Marquardt neural network-based intelligent computation for the non-Newtonian polymer during forward roll coating

Scientists and researchers widely recognize the effectiveness of artificial intelligence (AI)-based machine learning and intelligent computing solvers, demonstrating qualities such as resilience, robustness, stability, and rapid convergence. One particularly significant and rapidly growing field within AI is artificial neural networks. This research uses a supervised neural network model based on Levenberg–Marquardt backpropagation (LMB-SNNs) to examine the Sisko fluid model for the forward roll coating process (SFM-FRCP). A suitable transformation is applied to the partial differential equations based SFM-FRCP mathematical model, resulting in a set of nonlinear ordinary differential equations. The perturbation method has been used to find the analytical solutions for the velocity profile, pressure gradient, and pressure profile. A dataset for varying the pertinent parameters is generated, and the LMB-SNNs technique has been used to estimate the velocity profile, pressure gradient, and pressure profile behavior during FRCP for numerous scenarios. The numerical solution for SFM-FRCP in different scenarios, such as the validation, training, and testing procedures of LMB-SNNs, is carried out. Moreover, the state transition index, fitness outline, mean square error, histogram error, and regression presentation also endorse the strength and reliability of the solver LMB-SNNs for SFM-FRCP. The comparative analyses and performance studies through outputs of regression drawings, absolute error, and error histograms validate the effectiveness of the suggested solver LMB-SNNs. The method's precision is verified by the closest numerical outputs of both built and dataset values with similar levels 10−11–10−14. Furthermore, it has been observed that as the non-Newtonian parameter increases, the fluid velocity decreases. The research work carried out in this paper is original and fills a gap in the existing research by showing the rheological properties of the Sisko fluid model and the implementation of the LMB-SNNs during the FRCP.

JPEG Image Enhancement with Pre-Processing of Color Reduction and Smoothing.

JPEG is the international standard for still image encoding and is the most widely used compression algorithm because of its simple encoding process and low computational complexity. Recently, many methods have been developed to improve the quality of JPEG images by using deep learning. However, these methods require the use of high-performance devices since they need to perform neural network computation for decoding images. In this paper, we propose a method to generate high-quality images using deep learning without changing the decoding algorithm. The key idea is to reduce and smooth colors and gradient regions in the original images before JPEG compression. The reduction and smoothing can suppress red block noise and pseudo-contour in the compressed images. Furthermore, high-performance devices are unnecessary for decoding. The proposed method consists of two components: a color transformation network using deep learning and a pseudo-contour suppression model using signal processing. The experimental results showed that the proposed method outperforms standard JPEG in quality measurements correlated with human perception.

Constructing convolutional neural network by utilizing nematode connectome: A brain-inspired method

Convolutional neural networks have achieved impressive results in areas such as computer vision tasks. Recently, more complex architectures have been designed that add additional operations and connections to the standard architecture to enable deeper network training. A large number of trainable parameters are often required to acquire accurate results. As the number of parameters increases, the network structure becomes more intricate and complex, resulting in low model accuracy and wasted computational resources. Motivated by the promising performance of brain-inspired neural computation principles, a new network structure can be constructed using a composition of the biological connectome. The nematode Caenorhabditis elegans (C. elegans), as the only organism whose connectome has been fully mapped, is suitable to be selected for the construction of neural network structures. In this paper, the neuron connectome of C. elegans is encapsulated as a connectome block and used as the connection pattern of the convolutional layer in the convolutional network. Therefore, this modified convolutional neural network structure is called the nematode connectome neural network (NCNN), and the detailed implementation is presented. Moreover, comparative experiments on the CIFAR-100, CIFAR-10, MNIST, and ImageNet datasets are conducted to fully demonstrate the effectiveness and feasibility of the proposed NCNN model on image classification. In addition, the success of the NCNN also provides new ideas for future network structure design. The source code is publicly available at https://github.com/LongJin-lab/Nematode-Connectome-Neural-Network.

Dissecting neural computations in the human auditory pathway using deep neural networks for speech

The human auditory system extracts rich linguistic abstractions from speech signals. Traditional approaches to understanding this complex process have used linear feature-encoding models, with limited success. Artificial neural networks excel in speech recognition tasks and offer promising computational models of speech processing. We used speech representations in state-of-the-art deep neural network (DNN) models to investigate neural coding from the auditory nerve to the speech cortex. Representations in hierarchical layers of the DNN correlated well with the neural activity throughout the ascending auditory system. Unsupervised speech models performed at least as well as other purely supervised or fine-tuned models. Deeper DNN layers were better correlated with the neural activity in the higher-order auditory cortex, with computations aligned with phonemic and syllabic structures in speech. Accordingly, DNN models trained on either English or Mandarin predicted cortical responses in native speakers of each language. These results reveal convergence between DNN model representations and the biological auditory pathway, offering new approaches for modeling neural coding in the auditory cortex.

The autism spectrum disorder risk gene NEXMIF over-synchronizes hippocampal CA1 network and alters neuronal coding.

Mutations in autism spectrum disorder (ASD) risk genes disrupt neural network dynamics that ultimately lead to abnormal behavior. To understand how ASD-risk genes influence neural circuit computation during behavior, we analyzed the hippocampal network by performing large-scale cellular calcium imaging from hundreds of individual CA1 neurons simultaneously in transgenic mice with total knockout of the X-linked ASD-risk gene NEXMIF (neurite extension and migration factor). As NEXMIF knockout in mice led to profound learning and memory deficits, we examined the CA1 network during voluntary locomotion, a fundamental component of spatial memory. We found that NEXMIF knockout does not alter the overall excitability of individual neurons but exaggerates movement-related neuronal responses. To quantify network functional connectivity changes, we applied closeness centrality analysis from graph theory to our large-scale calcium imaging datasets, in addition to using the conventional pairwise correlation analysis. Closeness centrality analysis considers both the number of connections and the connection strength between neurons within a network. We found that in wild-type mice the CA1 network desynchronizes during locomotion, consistent with increased network information coding during active behavior. Upon NEXMIF knockout, CA1 network is over-synchronized regardless of behavioral state and fails to desynchronize during locomotion, highlighting how perturbations in ASD-implicated genes create abnormal network synchronization that could contribute to ASD-related behaviors.

Neural mechanisms for spatial cognition across vertebrates

The ability to navigate the world is a critical cognitive skill that most animals use to find food, shelter, and mates. Understanding the neural basis of navigation requires probing how the brain encodes spatial information through the study of the activity of single neurons and neuronal populations. Classically in vertebrates, studies have centered on the rodent hippocampal formation, which led to the discovery of place, grid, head direction and other cell types. However, since navigation skills are essential to almost all vertebrates, spatial cognition in different species also needs to be explored. In recent years, as a result of advances in technology, new data have emerged on the ways in which space is represented during navigation in the brains of vertebrates other than rodents, including teleost fish, birds, and other mammal species. Here, we review the state of the art on the neural representation of an animal’s position and motion across vertebrates at the level of single neurons. We argue that it is time to pool information across vertebrates to identify the underlying algorithms that lead to successful navigation. Although rodent-based data are important, findings in rodents are unlikely to cover the full spectrum of neural computations supporting navigation strategies in the vertebrate kingdom. Studying other species can shed light on length scales such as in large environments, and different scenarios such as naturalistic environments that are hard to carry out in rodents. In addition, a rodent-centric view may neglect the fact that different species are likely to represent positions in the world in ways that do not exist in mammals. Finally, we provide an outlook for the future which includes prediction about findings in unexplored species, and the opportunities for discoveries and understanding in this field.

Development and research of deep neural network fusion computer vision technology

Abstract Deep learning (DL) has revolutionized advanced digital picture processing, enabling significant advancements in computer vision (CV). However, it is important to note that older CV techniques, developed prior to the emergence of DL, still hold value and relevance. Particularly in the realm of more complex, three-dimensional (3D) data such as video and 3D models, CV and multimedia retrieval remain at the forefront of technological advancements. We provide critical insights into the progress made in developing higher-dimensional qualities through the application of DL, and also discuss the advantages and strategies employed in DL. With the widespread use of 3D sensor data and 3D modeling, the analysis and representation of the world in three dimensions have become commonplace. This progress has been facilitated by the development of additional sensors, driven by advancements in areas such as 3D gaming and self-driving vehicles. These advancements have enabled researchers to create feature description models that surpass traditional two-dimensional approaches. This study reveals the current state of advanced digital picture processing, highlighting the role of DL in pushing the boundaries of CV and multimedia retrieval in handling complex, 3D data.

Neural network execution using nicked DNA and microfluidics.

DNA has been discussed as a potential medium for data storage. Potentially it could be denser, could consume less energy, and could be more durable than conventional storage media such as hard drives, solid-state storage, and optical media. However, performing computations on the data stored in DNA is a largely unexplored challenge. This paper proposes an integrated circuit (IC) based on microfluidics that can perform complex operations such as artificial neural network (ANN) computation on data stored in DNA. We envision such a system to be suitable for highly dense, throughput-demanding bio-compatible applications such as an intelligent Organ-on-Chip or other biomedical applications that may not be latency-critical. It computes entirely in the molecular domain without converting data to electrical form, making it a form of in-memory computing on DNA. The computation is achieved by topologically modifying DNA strands through the use of enzymes called nickases. A novel scheme is proposed for representing data stochastically through the concentration of the DNA molecules that are nicked at specific sites. The paper provides details of the biochemical design, as well as the design, layout, and operation of the microfluidics device. Benchmarks are reported on the performance of neural network computation.

Before-after safety evaluation of part-time protected right-turn signals: An extreme value theory approach by applying artificial intelligence-based video analytics

Extreme value theory models have opened doors for before-after safety evaluation of engineering treatments using traffic conflict techniques. Recent advancements in automated conflict extraction technologies have further expedited conflict-based safety evaluation as a potential alternative to traditional crash-based methods. However, the suitability of extreme value theory models in the before-after evaluation of engineering treatments needs to be rigorously tested. As such, this study proposes a traffic conflict-based before-after evaluation of a novel part-time protected right-turn signal strategy for right-turn or opposing-through crashes at signalised intersections. A part-time protected right-turn signal strategy refers to a signal arrangement where permissive and fully protected right-turn phasings are operated during peak and off-peak hours, respectively. A deep neural network-based computer vision technique was applied to extract the conflicts from a total of 654 h of video recordings (before period: 266 h and after period: 388 h) over seven treated approaches, and four matching control approaches at five signalised intersections in the city of Cairns, Australia. Using post encroachment time and post-collision velocity difference as traffic conflict measures, non-stationary bivariate generalised extreme value models were developed to estimate the severe and non-severe opposing-through crashes at signal cycle levels. The odds ratio analysis of model-predicted crash risks suggests that part-time protected right-turn signals reduce 67% and 81% of severe and non-severe opposing-through crashes at signalised intersections, respectively. Part-time protected right-turn signal strategy offers a good safety solution without precipitating need for capacity upgrades to accommodate queued right turners at signalised intersections.

Adult-born granule cells facilitate remapping of spatial and non-spatial representations in the dentate gyrus

Adult-born granule cells (abGCs) have been implicated in memory discrimination through a neural computation known as pattern separation. Here, using invivo Ca2+ imaging, we examined how chronic ablation or acute chemogenetic silencing of abGCs affects the activity of mature granule cells (mGCs). In both cases, we observed altered remapping of mGCs. Rather than broadly modulating the activity of all mGCs, abGCs promote the remapping of place cells' firing fields while increasing rate remapping of mGCs that represent sensory cues. In turn, these remapping deficits are associated with behavioral impairments in animals' ability to correctly identify new goal locations. Thus, abGCs facilitate pattern separation through the formation of non-overlapping representations for identical sensory cues encountered in different locations. In the absence of abGCs, the dentate gyrus shifts to a state that is dominated by cue information, a situation that is consistent with the overgeneralization often observed in anxiety or age-related disorders.

Accurate Weight Update in an Electrochemical Random‐Access Memory Based Cross‐Point Array Using Channel‐High Half‐Bias Scheme for Deep Learning Accelerator

AbstractRecently cross‐point arrays of synaptic memory devices have been intensively studied to accelerate deep neural network computations. Among various synaptic devices, electrochemical random‐access memory (ECRAM) is emerging as a promising non‐volatile memory candidate owing to its superior synaptic characteristics. However, an optimized update scheme for a three‐terminal ECRAM‐based cross‐point array is yet to be developed. In this study, a metal‐oxide‐based ECRAM (MO‐ECRAM) shows superior synaptic characteristics and the weight update of devices in the MO‐ECRAM cross‐point array is analyzed using the half‐bias (HB) scheme. Additionally, A channel‐high half‐bias (CHB) scheme is proposed to overcome the degraded selectivity of the weight update caused by the three‐terminal configuration of the ECRAM device. In the CHB scheme, the conductance change in the selected device can be increased considerably by applying a calculated additional voltage to the channel. Using the CHB scheme, parallel and selective updates are successfully performed in a 2 × 2 MO‐ECRAM cross‐point array. Finally, an experimental demonstration of the training algorithm shows the impact of selective updates when using the CHB scheme. This new update scheme is expected to improve training accuracy in ECRAM cross‐point array‐based deep learning accelerators.

Neurocomputations on dual-brain signals underlie interpersonal prediction during a natural conversation

Prediction on the partner's speech plays a key role in a smooth conversation. However, previous studies on this issue have been majorly conducted at the single-brain rather than dual-brain level, leaving the interpersonal prediction hypothesis untested. To fill this gap, this study combined a neurocomputational modeling approach with a natural conversation paradigm in which two salespersons persuaded a customer to buy their product with their haemodynamic signals being collected using functional near-infrared spectroscopy hyperscanning. First, the results showed a cognitive hierarchy in a natural conversation, with the lower-level process (i.e., pragmatic representation of the persuasion) in the salesperson interacting with the higher-level process (i.e., value representation of the product) in the customer. Next, we found that the right dorsal lateral prefrontal cortex (rdlPFC) and temporoparietal junction (rTPJ) were associated with the representation of the product's value in the customer, while the right inferior frontal cortex (rIFC) was associated with the representation of the pragmatic processes in the salesperson. Finally, neurocomputational modeling results supported the prediction of the salesperson's lower-level brain activity based on the customer's higher-level brain activity. Moreover, the updating weight of the prediction model based on the neural computation between the rIFC of the salesperson and the rTPJ of the customer was closely associated with the interaction context, whereas that based on the rIFC-rdlPFC was not. In summary, these findings provide initial support for the interpersonal prediction hypothesis at the dual-brain level and reveal a hierarchy for the interpersonal prediction process.

Phase information is conserved in sparse, synchronous population-rate-codes via phase-to-rate recoding

Neural computation is often traced in terms of either rate- or phase-codes. However, most circuit operations will simultaneously affect information across both coding schemes. It remains unclear how phase and rate coded information is transmitted, in the face of continuous modification at consecutive processing stages. Here, we study this question in the entorhinal cortex (EC)- dentate gyrus (DG)- CA3 system using three distinct computational models. We demonstrate that DG feedback inhibition leverages EC phase information to improve rate-coding, a computation we term phase-to-rate recoding. Our results suggest that it i) supports the conservation of phase information within sparse rate-codes and ii) enhances the efficiency of plasticity in downstream CA3 via increased synchrony. Given the ubiquity of both phase-coding and feedback circuits, our results raise the question whether phase-to-rate recoding is a recurring computational motif, which supports the generation of sparse, synchronous population-rate-codes in areas beyond the DG.

Comparative Analysis of PWM AC Choppers with Different Loads with and Without Neural Network Application

In this paper, we focus on the "Artificial Neural Network (ANN) based PWM-AC chopper". This system is based on the PWM AC chopper-encouraged single-phase induction motor. The main purpose of this paper is to design and implement an ideal technique regarding speed control. Here analyzed PWM-based AC-AC converter with resistive load, R-L load and finally, the PWM AC chopper is fed to single phase induction for speed control. Using other soft computing and optimization techniques such as Artificial Neural Networks, Fuzzy Logic, Convolution algorithm, PSO, and Neuro Fuzzy can control the Speed. We used Artificial Neural Network to control the Speed of the PWM-AC Single phase induction motor drive. The Neural Network toolbox has been further used for getting desired responses. Neural system computer programs are executed in MATLAB. The performance of the proposed method of ANN system of PWM AC Chopper fed single phase induction motor drive is better than other traditional and base methods for controlling the Speed, based on the MOSFET.

The Structure of Hippocampal CA1 Interactions Optimizes Spatial Coding across Experience.

Although much is known about how single neurons in the hippocampus represent an animal's position, how circuit interactions contribute to spatial coding is less well understood. Using a novel statistical estimator and theoretical modeling, both developed in the framework of maximum entropy models, we reveal highly structured CA1 cell-cell interactions in male rats during open field exploration. The statistics of these interactions depend on whether the animal is in a familiar or novel environment. In both conditions the circuit interactions optimize the encoding of spatial information, but for regimes that differ in the informativeness of their spatial inputs. This structure facilitates linear decodability, making the information easy to read out by downstream circuits. Overall, our findings suggest that the efficient coding hypothesis is not only applicable to individual neuron properties in the sensory periphery, but also to neural interactions in the central brain.SIGNIFICANCE STATEMENT Local circuit interactions play a key role in neural computation and are dynamically shaped by experience. However, measuring and assessing their effects during behavior remains a challenge. Here, we combine techniques from statistical physics and machine learning to develop new tools for determining the effects of local network interactions on neural population activity. This approach reveals highly structured local interactions between hippocampal neurons, which make the neural code more precise and easier to read out by downstream circuits, across different levels of experience. More generally, the novel combination of theory and data analysis in the framework of maximum entropy models enables traditional neural coding questions to be asked in naturalistic settings.

Complexity Measures for EEG Microstate Sequences: Concepts and Algorithms

EEG microstate sequence analysis quantifies properties of ongoing brain electrical activity which is known to exhibit complex dynamics across many time scales. In this report we review recent developments in quantifying microstate sequence complexity, we classify these approaches with regard to different complexity concepts, and we evaluate excess entropy as a yet unexplored quantity in microstate research. We determined the quantities entropy rate, excess entropy, Lempel–Ziv complexity (LZC), and Hurst exponents on Potts model data, a discrete statistical mechanics model with a temperature-controlled phase transition. We then applied the same techniques to EEG microstate sequences from wakefulness and non-REM sleep stages and used first-order Markov surrogate data to determine which time scales contributed to the different complexity measures. We demonstrate that entropy rate and LZC measure the Kolmogorov complexity (randomness) of microstate sequences, whereas excess entropy and Hurst exponents describe statistical complexity which attains its maximum at intermediate levels of randomness. We confirmed the equivalence of entropy rate and LZC when the LZ-76 algorithm is used, a result previously reported for neural spike train analysis (Amigó et al., Neural Comput 16:717–736, https://doi.org/10.1162/089976604322860677, 2004). Surrogate data analyses prove that entropy-based quantities and LZC focus on short-range temporal correlations, whereas Hurst exponents include short and long time scales. Sleep data analysis reveals that deeper sleep stages are accompanied by a decrease in Kolmogorov complexity and an increase in statistical complexity. Microstate jump sequences, where duplicate states have been removed, show higher randomness, lower statistical complexity, and no long-range correlations. Regarding the practical use of these methods, we suggest that LZC can be used as an efficient entropy rate estimator that avoids the estimation of joint entropies, whereas entropy rate estimation via joint entropies has the advantage of providing excess entropy as the second parameter of the same linear fit. We conclude that metrics of statistical complexity are a useful addition to microstate analysis and address a complexity concept that is not yet covered by existing microstate algorithms while being actively explored in other areas of brain research.

Compressive sensing of functional connectivity maps from patterned optogenetic stimulation of neuronal ensembles

Mapping functional connectivity between neurons is an essential step toward probing the neural computations mediating behavior. Accurately determining synaptic connectivity maps in populations of neurons is challenging in terms of yield, accuracy, and experimental time. Here, we developed a compressive sensing approach to reconstruct synaptic connectivity maps based on random two-photon cell-targeted optogenetic stimulation and membrane voltage readout of many putative postsynaptic neurons. Using a biophysical network model of interconnected populations of excitatory and inhibitory neurons, we characterized mapping recall and precision as a function of network observability, sparsity, number of neurons stimulated, off-target stimulation, synaptic reliability, propagation latency, and network topology. We found that mapping can be achieved with far fewer measurements than the standard pairwise sequential approach, with network sparsity and synaptic reliability serving as primary determinants of the performance. Our results suggest a rapid and efficient method to reconstruct functional connectivity of sparsely connected neuronal networks.

Activeness: A Novel Neural Coding Scheme Integrating the Spike Rate and Temporal Information in the Spiking Neural Network

In neuromorphic computing, the coding method of spiking neurons serves as the foundation and is crucial for various aspects of network operation. Existing mainstream coding methods, such as rate coding and temporal coding, have different focuses, and each has its own advantages and limitations. This paper proposes a novel coding scheme called activeness coding that integrates the strengths of both rate and temporal coding methods. It encompasses precise timing information of the most recent neuronal spike as well as the historical firing rate information. The results of basic characteristic tests demonstrate that this encoding method accurately expresses input information and exhibits robustness. Furthermore, an unsupervised learning method based on activeness-coding triplet spike-timing dependent plasticity (STDP) is introduced, with the MNIST classification task used as an example to assess the performance of this encoding method in solving cognitive tasks. Test results show an improvement in accuracy of approximately 4.5%. Additionally, activeness coding also exhibits potential advantages in terms of resource conservation. Overall, activeness offers a promising approach for spiking neural network encoding with implications for various applications in the field of neural computation.

Understanding visual processing of motion: completing the picture using experimentally driven computational models of MT.

Computational modeling helps neuroscientists to integrate and explain experimental data obtained through neurophysiological and anatomical studies, thus providing a mechanism by which we can better understand and predict the principles of neural computation. Computational modeling of the neuronal pathways of the visual cortex has been successful in developing theories of biological motion processing. This review describes a range of computational models that have been inspired by neurophysiological experiments. Theories of local motion integration and pattern motion processing are presented, together with suggested neurophysiological experiments designed to test those hypotheses.

Information propagated by longitudinal pulses near a van der Waals phase transition.

Longitudinal pulses that propagate in a medium near a van der Waals phase transition have a sigmoidal dependence on the strength of the stimulus arising from the phase structure. This response resembles the all-or-nothing property of action potentials, which raises the question if an acoustic system near a phase transition can be suitable for material-based neuromorphic computation. Herein, we investigate how information about the stimulus is stored within these pulses. We find that (1) the pulse propagates in parallel both digital and analog information about the stimulus amplitude; (2) the pulse encodes the type of stimulus, for instance, mechanical or thermal; and (3) a collision between two pulses stores information about both stimuli and may be used as a fading memory. Our results unravel a rich encoding of information in a phenomenon that is both common in a plethora of materials and mimics neuronal signaling. In addition, we show that these pulses carry more information than is typically considered by models of neural computation. Therefore, this phenomenon is an excellent candidate for in materio computation.