Interconnected Neural Networks Research Articles

Advanced Low-Cost Technology for Assessing Metal Accumulation in the Body of a Metropolitan Resident Based on a Neural Network Model

This study is devoted to creating a neural network technology for assessing metal accumulation in the body of a metropolis resident with short-term and long-term intake from anthropogenic sources. Direct assessment of metal retention in the human body is virtually impossible due to the many internal mechanisms that ensure the kinetics of metals and the wide variety of organs, tissues, cellular structures, and secretions that ensure their functional redistribution, transport, and cumulation. We have developed an intelligent multi-neural network model capable of calculating the content of metals in the human body based on data on their environmental content. The model is two interconnected neural networks trained on actual measurement data. Since metals enter the body from the environment, the predictors of the model are metal content in drinking water and soil. In this case, water characterizes the short-term impact on the organism, and drinking water, combined with metal contents in soil, is a depository medium that accumulates metals from anthropogenic sources—the long-term impact. In addition, human physiological characteristics are taken into account in the calculations. Each period of exposure is taken into account by its neural network. Two variants of the model are proposed: open loop, where the calculation is performed by each neural network separately, and closed loop, where neural networks work together. The model built in this way was trained and tested on the data of real laboratory studies of 242 people living in different districts of Kazan. As a result, the accuracy of the neural network block for calculating long-term impact was 90% and higher, and the accuracy of the block for calculating short-term impact was 92% and higher. The closed double-loop model showed an accuracy of at least 96%. Conclusions: Our proposed method of assessing and quantifying metal accumulation in the body has high accuracy and reliability. It does not require expensive laboratory tests and allows quantifying the body’s metal accumulation content based on readily available information. The calculation results can be used as a tool for clinical diagnostics and operational and planned management to reduce the levels of polymetallic contamination in urban areas.

Thermodynamics-Informed Neural Networks for the Design of Solar Collectors: An Application on Water Heating in the Highland Areas of the Andes

This study addresses the challenge of optimizing flat-plate solar collector design, traditionally reliant on trial-and-error and simplified engineering design methods. We propose using physics-informed neural networks (PINNs) to predict optimal design conditions in a range of data that not only characterized the highlands of Ecuador but also similar geographical locations. The model integrates three interconnected neural networks to predict global collector efficiency by considering atmospheric, geometric, and physical variables, including overall loss coefficient, efficiency factors, outlet fluid temperature, and useful heat gain. The PINNs model surpasses traditional simplified thermodynamic equations employed in engineering design by effectively integrating thermodynamic principles with data-driven insights, offering more accurate modeling of nonlinear phenomena. This approach enhances the precision of solar collector performance predictions, making it particularly valuable for optimizing designs in Ecuador’s highlands and similar regions with unique climatic conditions. The ANN predicted a collector overall loss coefficient of 5.199 W/(m2·K), closely matching the thermodynamic model’s 5.189 W/(m2·K), with similar accuracy in collector useful energy gain (722.85 W) and global collector efficiency (33.68%). Although the PINNs model showed minor discrepancies in certain parameters, it outperformed traditional methods in capturing the complex, nonlinear behavior of the data set, especially in predicting outlet fluid temperature (55.05 °C vs. 67.22 °C).

Semi-supervised abdominal multi-organ segmentation by object-redrawing.

Multi-organ segmentation is a critical task in medical imaging, with wide-ranging applications in both clinical practice and research. Accurate delineation of organs from high-resolution 3D medical images, such as CT scans, is essential for radiation therapy planning, enhancing treatment outcomes, and minimizing radiation toxicity risks. Additionally, it plays a pivotal role in quantitative image analysis, supporting various medical research studies. Despite its significance, manual segmentation of multiple organs from 3D images is labor-intensive and prone to low reproducibility due to high interoperator variability. Recent advancements in deep learning have led to several automated segmentation methods, yet many rely heavily on labeled data and human anatomy expertise. In this study, our primary objective is to address the limitations of existing semi-supervised learning (SSL) methods for abdominal multi-organ segmentation. We aim to introduce a novel SSL approach that leverages unlabeled data to enhance the performance of deep neural networks in segmenting abdominal organs. Specifically, we propose a method that incorporates a redrawing network into the segmentation process to correct errors and improve accuracy. Our proposed method comprises three interconnected neural networks: a segmentation network for image segmentation, a teacher network for consistency regularization, and a redrawing network for object redrawing. During training, the segmentation network undergoes two rounds of optimization: basic training and readjustment. We adopt the Mean-Teacher model as our baseline SSL approach, utilizing labeled and unlabeled data. However, recognizing significant errors in abdominal multi-organ segmentation using this method alone, we introduce the redrawing network to generate redrawn images based on CT scans, preserving original anatomical information. Our approach is grounded in the generative process hypothesis, encompassing segmentation, drawing, and assembling stages. Correct segmentation is crucial for generating accurate images. In the basic training phase, the segmentation network is trained using both labeled and unlabeled data, incorporating consistency learning to ensure consistent predictions before and after perturbations. The readjustment phase focuses on reducing segmentation errors by optimizing the segmentation network parameters based on the differences between redrawn and original CT images. We evaluated our method using two publicly available datasets: the beyond the cranial vault (BTCV) segmentation dataset (training: 44, validation: 6) and the abdominal multi-organ segmentation (AMOS) challenge 2022 dataset (training:138, validation:16). Our results were compared with state-of-the-art SSL methods, including MT and dual-task consistency (DTC), using the Dice similarity coefficient (DSC) as an accuracy metric. On both datasets, our proposed SSL method consistently outperformed other methods, including supervised learning, achieving superior segmentation performance for various abdominal organs. These findings demonstrate the effectiveness of our approach, even with a limited number of labeled data. Our novel semi-supervised learning approach for abdominal multi-organ segmentation addresses the challenges associated with this task. By integrating a redrawing network and leveraging unlabeled data, we achieve remarkable improvements in accuracy. Our method demonstrates superior performance compared to existing SSL and supervised learning methods. This approach holds great promise in enhancing the precision and efficiency of multi-organ segmentation in medical imaging applications.

Dynamics of memristor coupled discrete fractional symmetric neural network model

Neural networks, also known as artificial neural networks (ANNs), represent innovative computational models inspired by the functioning of biological neurons in the human body. Currently, neural network models are employed to address intricate real-world challenges. This article focuses on constructing and dynamically analyzing interconnected symmetric neural network models inspired by the Hopfield type. These models incorporate discrete fractional memristor elements featuring quadratic memductance. The primary objective is to comprehend system characteristics by introducing nonlinearity to the inter-neuron weights in a functional manner. Additionally, the study explores the influence of electromagnetic radiation on neurons. Bifurcation diagrams, considering fractional order and Lyapunov exponents, illustrate the chaotic nature of the interconnected neural network model. The investigation delves into coexisting system states, examining coexisting bifurcations and attractors. The article offers potential for developing bio-inspired, energy-efficient, and adaptive neural network architectures, thereby contributing to advancements in artificial intelligence and neuromorphic applications.

Self‐triggered control for linear systems based on hierarchical reinforcement learning

AbstractIn this article, we propose a model‐free self‐triggered control approach for linear systems using a hierarchical policy framework. This framework splits the self‐triggered control approach into top and bottom‐level policies. The top‐level policy generates a triggering time interval based on an initial control strategy, while the bottom‐level policy creates a control inputs guided by the top‐level sub‐goal. This division ensures that both strategies have their own independent tasks and optimization goals, facilitating a model‐free iterative design process for self‐triggered control. The present structure integrates with a dual‐Actor Critic algorithm, utilizing two interconnected neural networks to approximate control and trigger policies. It reflects the framework of hierarchical reinforcement learning, wherein top‐level policies guide bottom‐level decision‐making. It fosters the model‐free design of self‐triggered controller, thereby enhancing the efficiency of the learning process. To validate the effectiveness of our proposed method, we conduct a series of numerical simulations.

Data-Driven Self-Triggered Control for Networked Motor Control Systems Using RNNs and Pre-Training: A Hierarchical Reinforcement Learning Framework.

This paper introduces a novel data-driven self-triggered control approach based on a hierarchical reinforcement learning framework in networked motor control systems. This approach divides the self-triggered control policy into higher and lower layers, with the higher-level policy guiding the lower-level policy in decision-making, thereby reducing the exploration space of the lower-level policy and improving the efficiency of the learning process. The data-driven framework integrates with the dual-actor critic algorithm, using two interconnected neural networks to approximate the hierarchical policies. In this framework, we use recurrent neural networks as the network architecture for the critic, utilizing the temporal dynamics of recurrent neural networks to better capture the dependencies between costs, thus enhancing the critic network's efficiency and accuracy in approximating the multi-time cumulative cost function. Additionally, we have developed a pre-training method for the control policy networks to further improve learning efficiency. The effectiveness of our proposed method is validated through a series of numerical simulations.

Neuronal plasticity features are independent of neuronal holding membrane potential

Dynamical reversible neuronal features in vitro are typically examined using a fixed holding membrane potential, imitating the physiological conditions of intact brains in an awake state. Here, a set of neuronal plasticity features in synaptic blocked cultures are found to be independent of the holding membrane potential in the range [−95, −50] mV. Specifically, dendritic maximal firing frequency and its absolute refractory period are independent of the holding membrane potential. In addition, the stimulation threshold is also independent of the holding membrane potential in neurons that do not show membrane depolarization in response to sub-threshold stimulations. These robust dendritic plasticity features are a prerequisite for neuronal modeling and for their utilization in interconnected neural networks to realize higher-order functionalities.

Microfluidic Bi-Layer Platform to Study Functional Interaction between Co-Cultured Neural Networks with Unidirectional Synaptic Connectivity

The complex synaptic connectivity architecture of neuronal networks underlies cognition and brain function. However, studying the spiking activity propagation and processing in heterogeneous networks in vivo poses significant challenges. In this study, we present a novel two-layer PDMS chip that facilitates the culturing and examination of the functional interaction of two interconnected neural networks. We utilized cultures of hippocampal neurons grown in a two-chamber microfluidic chip combined with a microelectrode array. The asymmetric configuration of the microchannels between the chambers ensured the growth of axons predominantly in one direction from the Source chamber to the Target chamber, forming two neuronal networks with unidirectional synaptic connectivity. We showed that the local application of tetrodotoxin (TTX) to the Source network did not alter the spiking rate in the Target network. The results indicate that stable network activity in the Target network was maintained for at least 1–3 h after TTX application, demonstrating the feasibility of local chemical activity modulation and the influence of electrical activity from one network on the other. Additionally, suppression of synaptic activity in the Source network by the application of CPP and CNQX reorganized spatio-temporal characteristics of spontaneous and stimulus-evoked spiking activity in the Target network. The proposed methodology and results provide a more in-depth examination of the network-level functional interaction between neural circuits with heterogeneous synaptic connectivity.

Three-dimensional highly porous hydrogel scaffold for neural circuit dissection and modulation.

Biomimetic brain structures and artificial neural networks have provided a simplified strategy for quantitatively investigating the complex structural and functional characteristics of highly interconnected neural networks. To achieve this, three-dimensional (3D) cell culture approaches have attracted much attention, which can mimic cell-cell interactions at the organism level and help better understand the function of specific neurons and neuronal networks than traditional two-dimensional cell culture methods. However, 3D scaffolds similar to the natural extracellular matrix to support the culturing, recording, and manipulation of neurons have long been an unresolved challenge. To resolve this, 3D hydrogel scaffolds can be fabricated via an innovative thermal treatment followed by an esterification process. A highly porous microstructure was formed within the bulk hydrogel scaffold, which showed a high porosity of 91% and a low Young's modulus of 6.11 kPa. Due to the merits of the fabricated hydrogel scaffolds, we constructed 3D neural networks and detected spontaneous action potentials in vitro. We successfully induced seizure-like waveforms in 3D cultured neurons and suppressed hyperactivated discharges by selectively activating γ-aminobutyric acid-ergic (GABAergic) interneurons. These results prove the advantages of our hydrogel scaffolds and demonstrate their application potential in the accurate dissection of neural circuits, which may help develop effective treatments for various neurological disorders. STATEMENT OF SIGNIFICANCE: While 3D cell culture approaches have attracted much attention and offer more advantages than two-dimensional cell culture methods, 3D scaffolds similar to the natural extracellular matrix to support the culturing, recording, and manipulation of neurons have long been an unresolved challenge. Herein, we developed a simplified and low-cost strategy for fabricating highly porous and cytocompatible hydrogel scaffolds for the construction of three-dimensional (3D) neural networks in vitro. The cultured 3D neural networks can mimic the in vivo connection among different neuron subgroups and help accurately dissect and manipulate the structure and function of specific neural circuits.

Embodied Metaphor Processing: A Study of the Priming Impact of Congruent and Opposite Gestural Representations of Metaphor Schema on Metaphor Comprehension

ABSTRACT This study examined the performances of three groups of participants in interpreting metaphors in three different conditions: congruent gesture-prime conditions, opposite gesture-prime conditions, and no-prime conditions. In congruent gesture-prime conditions, each metaphor was preceded by the congruent gestural representation of metaphor schema. In opposite gesture-prime conditions, each metaphor was preceded by the opposite gestural representation of metaphor schema. The results showed that participants of congruent gesture-prime conditions had the best performance in interpreting metaphors, while participants of opposite gesture-prime conditions had the worst performance. It is suggested that metaphor schema is an important part of metaphorical meaning of a metaphor. Therefore, when this part of meaning is activated by a gestural prime, metaphor comprehension is facilitated. Furthermore, when a schema is used to metaphorically describe a concept or an event, that schema and the neural network that represents it become one part of a larger interconnected neural network that represents the meaning of the metaphor. It could mean that the activation of one part of it could facilitate the activation of the whole of the network. Finally, it is emphasized that a gesture that depicts the schema of a metaphor is in fact the embodied realization of that metaphor.

Functional interactions among neurons within single columns of macaque V1.

Recent developments in high-density neurophysiological tools now make it possible to record from hundreds of single neurons within local, highly interconnected neural networks. Among the many advantages of such recordings is that they dramatically increase the quantity of identifiable, functional interactions between neurons thereby providing an unprecedented view of local circuits. Using high-density, Neuropixels recordings from single neocortical columns of primary visual cortex in nonhuman primates, we identified 1000s of functionally interacting neuronal pairs using established crosscorrelation approaches. Our results reveal clear and systematic variations in the synchrony and strength of functional interactions within single cortical columns. Despite neurons residing within the same column, both measures of interactions depended heavily on the vertical distance separating neuronal pairs, as well as on the similarity of stimulus tuning. In addition, we leveraged the statistical power afforded by the large numbers of functionally interacting pairs to categorize interactions between neurons based on their crosscorrelation functions. These analyses identified distinct, putative classes of functional interactions within the full population. These classes of functional interactions were corroborated by their unique distributions across defined laminar compartments and were consistent with known properties of V1 cortical circuitry, such as the lead-lag relationship between simple and complex cells. Our results provide a clear proof-of-principle for the use of high-density neurophysiological recordings to assess circuit-level interactions within local neuronal networks.

Accelerating the Layup Sequences Design of Composite Laminates via Theory-Guided Machine Learning Models.

Experimental and numerical investigations are presented for a theory-guided machine learning (ML) model that combines the Hashin failure theory (HFT) and the classical lamination theory (CLT) to optimize and accelerate the design of composite laminates. A finite element simulation with the incorporation of the HFT and CLT were used to generate the training dataset. Instead of directly mapping the relationship between the ply angles of the laminate and its strength and stiffness, a multi-layer interconnected neural network (NN) system was built following the logical sequence of composite theories. With the forward prediction by the NN system and the inverse optimization by genetic algorithm (GA), a benchmark case of designing a composite tube subjected to the combined loads of bending and torsion was studied. The ML models successfully provided the optimal layup sequences and the required fiber modulus according to the preset design targets. Additionally, it shows that the machine learning models, with the guidance of composite theories, realize a faster optimization process and requires less training data than models with direct simple NNs. Such results imply the importance of domain knowledge in helping improve the ML applications in engineering problems.

Deep Learning Applications in Surgery: Current Uses and Future Directions.

Deep learning (DL) is a subset of machine learning that is rapidly gaining traction in surgical fields. Its tremendous capacity for powerful data-driven problem-solving has generated computational breakthroughs in many realms, with the fields of medicine and surgery becoming increasingly prominent avenues. Through its multi-layer architecture of interconnected neural networks, DL enables feature extraction and pattern recognition of highly complex and large-volume data. Across various surgical specialties, DL is being applied to optimize both preoperative planning and intraoperative performance in new and innovative ways. Surgeons are now able to integrate deep learning tools into their practice to improve patient safety and outcomes. Through this review, we explore the applications of deep learning in surgery and related subspecialties with an aim to shed light on the practical utilization of this technology in the present and near future.

Prefrontal Input to the Periaqueductal Gray Controls Placebo Analgesia

Placebo analgesia is associated with activity in a highly interconnected neural network distributed throughout the brain. However, this is based on a large body of functional imaging work that primarily relies on correlative measures of neural activity and neurochemical signaling obtained in human subjects, making it challenging to ascribe causal roles to the underlying neural pathways. To address this gap, we developed a mouse model of placebo analgesia and used chemogenetic loss-of-function manipulations to identify the underlying neural circuits. We specifically addressed the role of the descending pain modulatory system (DPMS) and its upstream cortical circuits. We found that chemogenetic inactivation of a subpopulation of excitatory neurons in the ventrolateral periaqueductal gray (vlPAG) abolishes both morphine analgesia and morphine-conditioned placebo analgesia. Placebo, but not morphine analgesia, depends on vlPAG-projecting neurons in the medial prefrontal cortex (mPFC) and anterior cingulate cortex (ACC), but not the anterior insula (AI). For placebo to occur, mPFC-to-vlPAG neurons must be active during conditioning, consistent with a critical role for activity-dependent plasticity in this pathway. Our findings suggest that the vlPAG implements cognitive pain modulation by integrating top-down commands, at least some of which are conveyed directly from the cortex. Grant support from Rita Allen Foundation (non-profit) NIDA grant (R00 DA 034648) Klingenstein-Simons Foundation (non-profit).

Studying Cardiac Neural Network Dynamics: Challenges and Opportunities for Scientific Computing.

Neural control of the heart involves continuous modulation of cardiac mechanical and electrical activity to meet the organism’s demand for blood flow. The closed-loop control scheme consists of interconnected neural networks with central and peripheral components working cooperatively with each other. These components have evolved to cooperate control of various aspects of cardiac function, which produce measurable “functional” outputs such as heart rate and blood pressure. In this review, we will outline fundamental studies probing the cardiac neural control hierarchy. We will discuss how computational methods can guide improved experimental design and be used to probe how information is processed while closed-loop control is operational. These experimental designs generate large cardio-neural datasets that require sophisticated strategies for signal processing and time series analysis, while presenting the usual large-scale computational challenges surrounding data sharing and reproducibility. These challenges provide unique opportunities for the development and validation of novel techniques to enhance understanding of mechanisms of cardiac pathologies required for clinical implementation.

Prediction-error neurons in circuits with multiple neuron types: Formation, refinement, and functional implications

SignificanceAn influential idea in neuroscience is that neural circuits do not only passively process sensory information but rather actively compare them with predictions thereof. A core element of this comparison is prediction-error neurons, the activity of which only changes upon mismatches between actual and predicted sensory stimuli. While it has been shown that these prediction-error neurons come in different variants, it is largely unresolved how they are simultaneously formed and shaped by highly interconnected neural networks. By using a computational model, we study the circuit-level mechanisms that give rise to different variants of prediction-error neurons. Our results shed light on the formation, refinement, and robustness of prediction-error circuits, an important step toward a better understanding of predictive processing.

Recognition of dynamic objects from UGVs using Interconnected Neural network-based Computer Vision system

In this study, moving object recognition is performed by using images from a camera mounted on an unmanned ground vehicle. A GPS coordinate-based algorithm has been developed to obtain moving object silhouettes. In order to classify these silhouettes, an interconnected artificial neural network (ICANN) architecture consisting of two stages has been developed. The method consists of two phases. In the first phase, real-time images are converted to binary images at the end of the GPS-assisted image registration process. Then, the silhouettes are extracted from the background of the images using connected component labelling. In the second phase, two interconnected neural networks are used. The first neural network classifies silhouettes as objects or noise. The second neural network divides objects into seven subclasses as pedestrians, potholes, cars, etc. Compared to CNN-based techniques, a simpler NN architecture was employed in the research, and better accuracy rates were achieved with fewer samples. Another contribution of the research is simultaneous localization and mapping (SLAM) applications can be performed in non-GPS environments using pre-recorded images containing GPS information. In experimental studies, maximum success rates of 96,1% in object classification were obtained. The results obtained were compared to YOLO, the recently popular algorithm for object recognition.

Multilayer Network Approach in EEG Motor Imagery with an Adaptive Threshold.

The brain has been understood as an interconnected neural network generally modeled as a graph to outline the functional topology and dynamics of brain processes. Classic graph modeling is based on single-layer models that constrain the traits conveyed to trace brain topologies. Multilayer modeling, in contrast, makes it possible to build whole-brain models by integrating features of various kinds. The aim of this work was to analyze EEG dynamics studies while gathering motor imagery data through single-layer and multilayer network modeling. The motor imagery database used consists of 18 EEG recordings of four motor imagery tasks: left hand, right hand, feet, and tongue. Brain connectivity was estimated by calculating the coherence adjacency matrices from each electrophysiological band (, , and ) from brain areas and then embedding them by considering each band as a single-layer graph and a layer of the multilayer brain models. Constructing a reliable multilayer network topology requires a threshold that distinguishes effective connections from spurious ones. For this reason, two thresholds were implemented, the classic fixed (average) one and Otsu’s version. The latter is a new proposal for an adaptive threshold that offers reliable insight into brain topology and dynamics. Findings from the brain network models suggest that frontal and parietal brain regions are involved in motor imagery tasks.

Synchronization in Finite/Fixed Time for Markovian Complex-Valued Nonlinear Interconnected Neural Networks With Reaction–Diffusion Terms

This paper considers a novel class of nonlinear interconnected neural networks (INNs) where the Markovian jump parameters and reaction–diffusion terms are both involved. The nonlinear INNs are studied in the complex domain, which is a meaningful attempt in the research on INNs. Then, by designing a suitable nonlinear controller and employing the Lyapunov functional and algebraic inequality technologies, a finite/fixed-time synchronization criterion in probability can be obtained for the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$N$</tex-math></inline-formula> subsystems. Note that the criteria for synchronization in finite time and fixed time are obtained with only one controller and integrated into a unified theorem. A secret communication example is presented, such that the validity and practicability of this paper can be verified.

PREDICTION OF THERMAL ERRORS IN MACHINE TOOLS THROUGH DECOUPLED SIMULATIONS USING GENETIC ALGORITHM AND ARTIFICIAL NEURAL NETWORKS

Thermal errors are one of the major contributors towards positioning discrepancies in machine tools in precision machining. Along with friction and waste heat generated from production processes and internal heat sources, environmental influences around the machine tool create considerable thermal gradients followed by non-linear structural deformations. Efficient quantification of these three contributing sources of thermal errors are required in order to formulate a reliable thermal-error compensation system. The creation of all possible thermal configurations, which a machine tool could be subjected to, is experimentally infeasible and requires complex and time-consuming coupled flow and thermo-structural simulations. This paper presents a new approach in thermal error prediction by using CFD and finite element (FE) simulations to train a three-level interconnected neural network system. The first level essentially decouples flow simulations from thermo-structural simulations using optimal FE node points found using a Genetic Algorithm (GA), which significantly reduces the required training data. The boundary convection data obtained from this level is used in the second level to predict possible thermal configurations of the machine tool, after careful consideration of parameters related to internal heat sources and production processes. The third level maps these thermal configurations onto displacements on the machine tool.