Neural Network Platform Research Articles

Оптимизация тихоходного синхронного генератора модульного типа и принцип реализации системы управления напряжением генератора на основе нейронной сети

A method for optimizing the parameters of a modular half-speed synchronous generator with permanent magnets (PMSG) and the generator voltage control system with a neural network-based algorithm are proposed. The basic design scheme of the modular half-speed PMSG is considered, which features a compact layout of the generator main parts, thereby ensuring the optimal use of the working volume, smaller sizes of the magnetic system, and smaller mass of the active materials used in manufacturing the machine. Owing to the simple and reliable design of the generator, its output parameters can be varied in a wide range with using standard electrical circuits for voltage stabilization and current rectification along with an additional voltage regulation unit. Owing to this feature, the design scheme of the considered generator has essential advantages over the existing analogs with a common cylindrical magnetic core. In view of these circumstances, the development of a high-efficient modular half-speed PMSG as an autonomous DC power source is of both scientific and practical interest; this generator can be used to supply power to a large range of electricity consumers located in rural areas, low-rise residential areas, military communities, allotments etc. In solving the problem of optimizing the generator’s magnetic system, the main electrical machine analysis equation is obtained. The optimal ratios of the winding wire mass to the mass of permanent magnets and of the PM height to the air gap value for achieving the maximum specific useful power output have been determined. An analytical correlation between the optimal design parameters of a half-speed modular PMSG and its power performance parameters has been established. The expediency to develop a neural network-based control system is shown. The number of load-bearing modules of the half-speed PMSG is determined depending on the wind velocity, load factor and the required output voltage. The neural network was trained on the examples of a training sample using a laboratory test bench. The neural network was implemented in the MatLab 2019b environment by constructing a synchronous generator simulation model in the Simulink software extension. The possibility of using the voltage control system of a half-speed modular PMSG with a microcontroller for operation of the neural network platform of the Arduino family (ArduinoDue) independently of the PC is shown.

Deep Learning-Based Nuclear Lobe Count Method for Differential Count of Neutrophils.

Differentiating neutrophils based on the count of nuclear lobulation is useful for diagnosing various hematological disorders, including megaloblastic anemia, myelodysplastic syndrome, and sepsis. It has been reported that one-fifth of sepsis-infected patients worldwide died between 1990 and 2017. Notably, fewer nuclear-lobed and stab-formed neutrophils develop in the peripheral blood during sepsis. This abnormality can serve as an early diagnostic criterion. However, testing this feature is a complex and time-consuming task that is rife with human error. For this reason, we apply deep learning to automatically differentiate neutrophil and nuclear lobulation counts and report the world's first small-scale pilot. Blood films are prepared using venous peripheral blood taken from four healthy volunteers and are stained with May-Grünwald Giemsa stain. Six-hundred 360 × 363-pixel images of neutrophils having five different nuclear lobulations are automatically captured by Cellavision DM-96, an automatic digital microscope camera. Images are input to an original architecture with five convolutional layers built on a deep learning neural-network platform by Sony, Neural Network Console. The deep learning system distinguishes the four groups (i.e., band-formed, two-, three-, and four- and five- segmented) of neutrophils with up to 99% accuracy, suggesting that neutrophils can be automatically differentiated based on their count of segmented nuclei using deep learning.

Machine Learning Outperforms Standard Force-Field Calibration Techniques

The neural network platform, called DeepCalib, streamlines the calibration process even in cases where no standard method exists.

A Novel Deep Learning and Polar Transformation Framework for an Adaptive Automatic Modulation Classification

Automatic Modulation Classification (AMC) is an approach to identify an observed signal's most likely modulation scheme without any a priori knowledge of the intercepted signal. In this research, the authors present a new direction for both stages of feature-based (FB) approach. In the feature extraction stage, the authors design a new architecture that 1) removes the bias issue for the estimator of fourth-order cumulants, and 2) extracts polar-transformed information of the received $IQ$ symbols, and finally 3) forms a unique dataset to be used in the labeling stage. Furthermore, the authors contribute to increasing the classification accuracy in low signal-to-noise ratio (SNR) conditions by employing the deep belief network (DBN) platform in addition to the spiking neural network (SNN) platform to overcome execution latency concerns associated with deep learning architectures. For this research, the authors first study each individual FB AMC classifier to derive their respective upper and lower performance bounds and then propose an adaptive framework that is built and developed with these findings. This framework aims to efficiently classify the modulation scheme by intelligently switching between these different FB classifiers to achieve an optimal balance between accuracy and execution latency for any observed channel conditions derived from the main receiver's equalizer. Subsequently, a performance analysis is conducted using the standard RadioML dataset to achieve a realistic evaluation. Numerical results indicate a notably higher classification accuracy, by 16.02% on average, when DBN is employed, whereas SNN requires significantly lower execution latency to label the modulation scheme when compared against two other modified FB classifiers that are built upon convolutional and recurrent neural networks, shown to be reduced by 34.31%.

Artificial van der Waals hybrid synapse and its application to acoustic pattern recognition

Brain-inspired parallel computing, which is typically performed using a hardware neural-network platform consisting of numerous artificial synapses, is a promising technology for effectively handling large amounts of informational data. However, the reported nonlinear and asymmetric conductance-update characteristics of artificial synapses prevent a hardware neural-network from delivering the same high-level training and inference accuracies as those delivered by a software neural-network. Here, we developed an artificial van-der-Waals hybrid synapse that features linear and symmetric conductance-update characteristics. Tungsten diselenide and molybdenum disulfide channels were used selectively to potentiate and depress conductance. Subsequently, via training and inference simulation, we demonstrated the feasibility of our hybrid synapse toward a hardware neural-network and also delivered high recognition rates that were comparable to those delivered using a software neural-network. This simulation involving the use of acoustic patterns was performed with a neural network that was theoretically formed with the characteristics of the hybrid synapses.

Guidance Mechanism for Flexible-Wing Aircraft Using Measurement-Interfaced Machine-Learning Platform

The autonomous operation of flexible-wing aircraft poses theoretical and technical challenges not yet addressed in the literature. The lack of an exact modeling framework is due to the complex nonlinear aerodynamics driven by the deformations of the flexible-wings, which in turn complicates the controls and instrumentation setup of the navigation system. This urges for innovative approaches to interface affordable instrumentation platforms to autonomously control this type of aircraft. This article leverages the ideas from instrumentation and measurements, machine learning, and optimization fields in order to develop an autonomous navigation system for a flexible-wing aircraft. A novel machine-learning process based on a guiding search mechanism is developed to interface real-time measurements of wing-orientation dynamics into control decisions. This process is realized using an online value iteration algorithm based on two improved and interacting model-free control strategies in real time. The first strategy is concerned with achieving the tracking objectives, whereas the second supports the stability of the system. A neural network platform that employs adaptive critics is utilized to approximate the control strategies while approximating the assessments of their values. An experimental actuation system is utilized to test the validity of the proposed platform. The experimental results are shown to be aligned with the stability features of the proposed model-free adaptive learning approach.

Automated Surgical Instrument Detection from Laparoscopic Gastrectomy Video Images Using an Open Source Convolutional Neural Network Platform

The common use of laparoscopic intervention produces impressive amounts of video data that are difficult to review for surgeons wishing to evaluate and improve their skills. Therefore, a need exists for the development of computer-based analysis of laparoscopic video to accelerate surgical training and assessment. We developed a surgical instrument detection system for video recordings of laparoscopic gastrectomy procedures. This system, the use of which might increase the efficiency of the video reviewing process, is based on the open source neural network platform, YOLOv3. A total of 10,716 images extracted from 52 laparoscopic gastrectomy videos were included in the training and validation data sets. We performed 200,000 iterations of training. Video recordings of 10 laparoscopic gastrectomies, independent of the training and validation data set, were analyzed by our system, and heat maps visualizing trends of surgical instrument usage were drawn. Three skilled surgeons evaluated whether each heat map represented the features of the corresponding operation. After training, the testing data set precision and sensitivity (recall) was 0.87 and 0.83, respectively. The heat maps perfectly represented the devices used during each operation. Without reviewing the video recordings, the surgeons accurately recognized the type of anastomosis, time taken to initiate duodenal and gastric dissection, and whether any irregular procedure was performed, from the heat maps (correct answer rates ≥ 90%). A new automated system to detect manipulation of surgical instruments in video recordings of laparoscopic gastrectomies based on the open source neural network platform, YOLOv3, was developed and validated successfully.

Нейросетевые платформенные решения для снижения аварийности процесса строительства нефтяных и газовых скважин

Нейросетевые платформенные решения для снижения аварийности процесса строительства нефтяных и газовых скважин

Research on Automatic Operation Control Algorithm of High Speed Train Based on Artificial Neural Network

With the development of social economy, the demand for transportation is also growing. High-speed railway has become an important part of the transportation system. It is not only an inevitable requirement for the development of the national economy, but also an important support for the implementation of the national strategy. Therefore, whether the high-speed train can operate smoothly and safely is related to the passenger’s life and health and property safety. This requires accurate control of the automatic operation of the high-speed train during the driving process. At present, China’s research on the automatic operation control of high-speed trains is still in its infancy. The problems of traction and brake system actuator failures that may occur during long-term running of trains can be solved by artificial neural network-based modeling methods. The automatic operation control algorithm of high-speed train running process is established. The self-correcting control strategy based on artificial neural network is used to realize the speed tracking of high-speed trains. The high-speed train visual simulation system is established by virtual reality technology and verified on the artificial neural network platform. The effectiveness of the proposed automatic operation control method.

Abstract 2447: Accurate prediction of the signaling pathway activity elicited by variants of uncertain significance (VUS) and their drug sensitivity using deep convolutional neural networks on florescent microscopy images of cells

Abstract Sequencing based methods to tailor the treatment of cancer patients are at the cutting edge of personalized medicine and have remarkably advanced in the past few years. However, a large number of mutations that are found when sequencing tumors are poorly characterized in the literature and are thus labeled as VUS. Because of their uncertain significance, it is not possible to assign them a function and hence are overlooked when designing treatment regimens, and this can have deleterious effects. Here we report the development of a Deep Convolutional Neural Network (DCNN) platform trained using Deep Learning algorithms, that is able to accurately predict the activity elicited by specific mutations. The DCNN is based on a high-throughput platform to predict the activation of signaling pathways elicited by the specific mutations via changes to the intra-cellular localization of fluorescently labeled protein reporters. For our training set we used images of 6 billion cells expressing either WT or mutations with known activity of the analyzed genes. After training, the network was presented with images of cells transfected with VUS that the network has never seen before. We show that the network was able to correctly predict the level of activity of a large number of VUS in KRAS, BRAF and cKIT and on multiple signaling pathways, specifically the MAPK and JAK-STAT pathways. These findings were further confirmed by both biochemical assays as well as literature search. Furthermore, images of cells expressing known mutations of BRAF and cKIT and treated with increasing doses of the targeted treatments Vemurafenib and Imatinib, respectively, were analyzed by the network and we were able to generate high resolution dose response curves in specific signaling pathways. Lastly, we present the predicted dose response curves in specific signaling pathways of a variety of VUS which have never been characterized before and show that different VUS respond differently to the tested drugs. Overall, we provide evidence that this novel tool enables robust and highly accurate calculation of activity profiles of a wide range of genes and signaling pathways. Each mutation is highly specific in the magnitude of activity it elicits. The high degree of resolution allows the investigation of the efficacy of different targeted therapies on a wide range of mutations. Citation Format: Lior Zimmerman, Arie Aizenman, Yutti Daitsh, Zohar Barbash, Gabi Tarcic, Michael Vidne. Accurate prediction of the signaling pathway activity elicited by variants of uncertain significance (VUS) and their drug sensitivity using deep convolutional neural networks on florescent microscopy images of cells [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 2447.

Optimized Real-Time Biomimetic Neural Network on FPGA for Bio-hybridization.

Neurological diseases can be studied by performing bio-hybrid experiments using a real-time biomimetic Spiking Neural Network (SNN) platform. The Hodgkin-Huxley model offers a set of equations including biophysical parameters which can serve as a base to represent different classes of neurons and affected cells. Also, connecting the artificial neurons to the biological cells would allow us to understand the effect of the SNN stimulation using different parameters on nerve cells. Thus, designing a real-time SNN could useful for the study of simulations of some part of the brain. Here, we present a different approach to optimize the Hodgkin-Huxley equations adapted for Field Programmable Gate Array (FPGA) implementation. The equations of the conductance have been unified to allow the use of same functions with different parameters for all ionic channels. The low resources and high-speed implementation also include features, such as synaptic noise using the Ornstein–Uhlenbeck process and different synapse receptors including AMPA, GABAa, GABAb, and NMDA receptors. The platform allows real-time modification of the neuron parameters and can output different cortical neuron families like Fast Spiking (FS), Regular Spiking (RS), Intrinsically Bursting (IB), and Low Threshold Spiking (LTS) neurons using a Digital to Analog Converter (DAC). Gaussian distribution of the synaptic noise highlights similarities with the biological noise. Also, cross-correlation between the implementation and the model shows strong correlations, and bifurcation analysis reproduces similar behavior compared to the original Hodgkin-Huxley model. The implementation of one core of calculation uses 3% of resources of the FPGA and computes in real-time 500 neurons with 25,000 synapses and synaptic noise which can be scaled up to 15,000 using all resources. This is the first step toward neuromorphic system which can be used for the simulation of bio-hybridization and for the study of neurological disorders or the advanced research on neuroprosthesis to regain lost function.

Development of an artificial neural network based virtual sensing platform for the simultaneous prediction of emission-performance-stability parameters of a diesel engine operating in dual fuel mode with port injected methanol

The present work explores the potential of an artificial neural network platform to emulate the performance, emissions and stability indices of an existing single cylinder diesel engine operating in dual-fuel mode with methanol port injection under varying fuel injection pressure. This investigation is further augmented by hydrous methanol injection strategies. Brake power, fuel injection pressure, diesel specific fuel consumption, methanol specific fuel consumption, air flow rate, exhaust oxygen and temperature have been chosen as the model inputs while oxides of nitrogen, unburned hydrocarbon, carbon monoxide, carbon dioxide, soot have been chosen as the emission responses to be modelled along with equivalent brake specific fuel consumption as the performance response and coefficient of variance of indicated mean effective pressure as the stability parameter to be estimated. Absolute, relative and percentage-based statistical error metrics have been employed for model evaluation. The developed model shows an excellent agreement with the experimental data as evident from its extremely low normalized mean square error, symmetric mean absolute percentage error, Normalized root mean square error, mean squared relative error footprint coupled with high coefficient of determination which was observed to be within a range of 0.983–0.9999 and a corresponding Nash Sutcliffe coefficient of efficiency of 85%–99.6%. Furthermore, low Theil uncertainty evaluation and Kullback-Leibler Divergence values imparted a commendable credence of robustness to the estimation capability of the developed model. The present study manifests a computationally efficient and reliable virtual sensing platform to simultaneously emulate the emission-performance and stability parameters of a diesel-methanol partially premixed dual fuel operational paradigms in real time engine control strategies.

Analysis of the bystander effect in cone photoreceptors via a guided neural network platform.

The mammalian retina system consists of a complicated photoreceptor structure, which exhibits extensive random synaptic connections. To study retinal development and degeneration, various experimental models have been used previously, but these models are often uncontrollable, are difficult to manipulate, and do not provide sufficient similarity or precision. Therefore, the mechanisms in many retinal diseases remain unclear because of the limited capability in observing the progression and molecular driving forces. For example, photoreceptor degeneration can spread to surrounding healthy photoreceptors via a phenomenon known as the bystander effect; however, no in-depth observations can be made to decipher the molecular mechanisms or the pathways that contribute to the spreading. It is then necessary to build dissociated neural networks to investigate the communications with controllability of cells and their treatment. We developed a neural network chip (NN-Chip) to load single neurons into highly ordered microwells connected by microchannels for synapse formation to build the neural network. By observing the distribution of apoptosis spreading from light-induced apoptotic cones to the surrounding cones, we demonstrated convincing evidence of the existence of a cone-to-cone bystander killing effect. Combining the NN-Chip with microinjection technology, we also found that the gap junction protein connexin 36 (Cx36) is critical for apoptosis spreading and the bystander effect in cones. In addition, our unique NN-Chip platform provides a quantitative, high-throughput tool for investigating signaling mechanisms and behaviors in neurons and opens a new avenue for screening potential drug targets to cure retinal diseases.

Detection and Analysis of Distributed Denial-of-service in the Internet of Things—Employing an Artificial Neural Network and Apache Spark Platform

Detection and Analysis of Distributed Denial-of-service in the Internet of Things—Employing an Artificial Neural Network and Apache Spark Platform

Application of a computational neural network to optimize the fluorescence signal from a receptor-ligand interaction on a microfluidic chip.

We describe the use of a computational neural network platform to optimize the fluorescence upon binding 5-carboxyfluorescein-d-Ala-d-Ala-d-Ala (5-FAM(DA)3 ) (1) to the antibiotic teicoplanin covalently attached to a glass slide. A three-level response surface experimental design was used as the first stage of investigation. Subsequently, three defined experimental parameters were examined by the neural network approach: (i) the concentration of teicoplanin used to derivatize a glass platform on the microfluidic device, (ii) the time required for the immobilization of teicoplanin on the platform, and (iii) the length of time 1 is allowed to equilibrate with teicoplanin in the microfluidic channel. Optimal neural structure provided a best fit model, both for the training set (r(2) = 0.961) and test set (r(2) = 0.934) data. Model simulated results were experimentally validated with excellent agreement (% difference) between experimental and predicted fluorescence shown, thus demonstrating efficiency of the neural network approach.

Implementation of a genetically tuned neural platform in optimizing fluorescence from receptor-ligand binding interactions on microchips.

This paper describes the use of a genetically tuned neural network platform to optimize the fluorescence realized upon binding 5-carboxyfluorescein-D-Ala-D-Ala-D-Ala (5-FAM-(D-Ala)(3) ) (1) to the antibiotic teicoplanin from Actinoplanes teichomyceticus electrostatically attached to a microfluidic channel originally modified with 3-aminopropyltriethoxysilane. Here, three parameters: (i) the length of time teicoplanin was in the microchannel; (ii) the length of time 1 was in the microchannel, thereby, in equilibrium with teicoplanin, and; (iii) the amount of time buffer was flushed through the microchannel to wash out any unbound 1 remaining in the channel, are examined at a constant concentration of 1, with neural network methodology applied to optimize fluorescence. Optimal neural structure provided a best fit model, both for the training set (r(2) = 0.985) and testing set (r(2) = 0.967) data. Simulated results were experimentally validated demonstrating efficiency of the neural network approach and proved superior to the use of multiple linear regression and neural networks using standard back propagation.

Biologically driven neural platform invoking parallel electrophoretic separation and urinary metabolite screening

This work reveals a computational framework for parallel electrophoretic separation of complex biological macromolecules and model urinary metabolites. More specifically, the implementation of a particle swarm optimization (PSO) algorithm on a neural network platform for multiparameter optimization of multiplexed 24-capillary electrophoresis technology with UV detection is highlighted. Two experimental systems were examined: (1) separation of purified rabbit metallothioneins and (2) separation of model toluene urinary metabolites and selected organic acids. Results proved superior to the use of neural networks employing standard back propagation when examining training error, fitting response, and predictive abilities. Simulation runs were obtained as a result of metaheuristic examination of the global search space with experimental responses in good agreement with predicted values. Full separation of selected analytes was realized after employing optimal model conditions. This framework provides guidance for the application of metaheuristic computational tools to aid in future studies involving parallel chemical separation and screening. Adaptable pseudo-code is provided to enable users of varied software packages and modeling framework to implement the PSO algorithm for their desired use.

A Method of Genetic Algorithm (GA) for FIR Filter Construction: Design and Development with Newer Approaches in Neural Network Platform

The main focus of this paper is to describe a developed and dynamic method of designing finite impulse response filters with automatic, rapid and less computational complexity by an efficient Genetic approach. To obtain such efficiency, specific filter coefficient coding scheme has been studied and implemented. The algorithm generates a population of genomes that represents the filter coefficient where new genomes are generated by crossover, mutation operations methods. Our proposed genetic technique has able to give better result compare to other method

Analog neural network-based helicopter gearbox health monitoring system.

The development of a reliable helicopter gearbox health monitoring system (HMS) has been the subject of considerable research over the past 15 years. The deployment of such a system could lead to a significant saving in lives and vehicles as well as dramatically reduce the cost of helicopter maintenance. Recent research results indicate that a neural network-based system could provide a viable solution to the problem. This paper presents two neural network-based realizations of an HMS system. A hybrid (digital/analog) neural system is proposed as an extremely accurate off-line monitoring tool used to reduce helicopter gearbox maintenance costs. In addition, an all analog neural network is proposed as a real-time helicopter gearbox fault monitor that can exploit the ability of an analog neural network to directly compute the discrete Fourier transform (DFT) as a sum of weighted samples. Hardware performance results are obtained using the Integrated Neural Computing Architecture (INCA/1) analog neural network platform that was designed and developed at The Charles Stark Draper Laboratory. The results indicate that it is possible to achieve a 100% fault detection rate with 0% false alarm rate by performing a DFT directly on the first layer of INCA/1 followed by a small-size two-layer feed-forward neural network and a simple post-processing majority voting stage.

Hybrid architecture for understanding motion sequences

The aim of the research introduced in this paper is the development of a hybrid vision system architecture based on a unified neural network platform and a query-based database that shares their noise-tolerant, learning and interactive properties. The function of the system is to model and interpret the behaviour of a class of cancerous cells by (a) extracting cell features from the images, (b) classifying the temporal and social behaviour of the cells, and (c) driving the interpretation search according to the needs of the user. This paper gives a description of the application and the proposed system and presents the results drawn from two neural network architectures used for the extraction of cell centroid areas from images. Both networks are implemented on a Distributed Array of Processors (DAP) and trained using the backpropagation learning algorithm.