Phase-gradient information from vortex encoders

This paper reports results with ABI, a portable noninvasive brain-computer interface. It uses 8 scalp electrodes to measure spontaneous EEG signals from which we extract simple power spectral features. The features are fed to a simple local neural network that recognizes reliably 3 different mental states. We compare the performance of this local classifier to more complex time-processing neural networks. We also illustrate the control of a complex brain-actuated device; i.e., a robot moving along smooth and safe paths between rooms.

House price forecasting with neural networks

The author considers a discrete-time neural network which consists of only two neurons with the sigmoidal nonlinear function as the neuron activation function and has no external inputs and no time delay. He treats the simple network as a one-parameter family of two-dimensional maps with the neuron gain as the parameter, and mathematically proves the existence of period-doublings to chaos in the network with an excitatory neuron and an inhibitory neuron. Specifically, it is proved that, for a certain class of singular connection weight matrices, the simple neural network is dynamically equivalent to a one-parameter full family of (one-dimensional) S-unimodal maps on the interval which is well-known to become chaotic through the period-doubling route as the parameter varies.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Leaky integrate-and-fire neurons based on perovskite memristor for spiking neural networks

We have investigated selective learning techniques for improving the ability of back-propagation neural networks to predict large changes. The prediction of daily stock prices was taken as an example of a noisy real-world problem. We previously proposed the selective-presentation and selective-learning-rate approaches and applied them into feed-forward neural networks. This paper applies the selective-learning-rate approach into three types of simple recurrent neural networks. We evaluated their performances through experimental stock-price prediction. Using selective-learning-rate approach, the network can learn the large changes well and profit per trade was improved in all of simple recurrent neural networks.

In today's volatile financial market the demand for an accurate option price forecaster has been a focal point for researchers. The purpose of this study is to forecast option prices using neural networks. Initially simple neural network was implemented using twenty year period data from S&P 500 index call option prices. The prediction result was better than that of traditional Black-Scholes model. A hybrid neural network was developed that utilized aspects of Black- Scholes model into the neural network and tested against the traditional approach and simple neural network. The hybrid neural network outperformed performance of the tradition forecasting model and improved prediction results of simple neural networks.

In the 21st century marine navigation has become dominated by satellite positioning systems and automated navigational processes. Today, global navigation satellite systems (GNSS) play a central role in the process of carrying out basic navigational tasks, e.g. determining the coordinates of a vessel’s position at sea. Since satellite systems are being used increasingly more often in everyday life, the signals they send are becoming more and more prone to jamming. Therefore there is a need to search for other positioning systems and methods that would be as accurate and fast as the existing satellite systems. On the other hand, the automation process makes it possible to conduct navigational tasks more quickly. Due to the development of this technology, all kinds of navigation equipment can be used in the process of automating navigation. This also applies to marine radars, which are characterised by a relatively high accuracy that allows them to replace satellite systems in performing classic navigational tasks. By employing M-estimation methods that are used in geodesy as well as simple neural networks, a software package can be created that will aid in automating navigation and will provide highly accurate information about a given object’s position at sea by making use of radar in comparative navigation. This paper presents proposals for automating the process of determining a vessel’s position at sea by using comparative navigation methods that are based on simple neural networks and geodetic M-estimation methods.

Heterogeneous graph neural networks (HGNNs) have the powerful capability to embed rich structural and semantic information of a heterogeneous graph into node representations. Existing HGNNs inherit many mechanisms from graph neural networks (GNNs) designed for homogeneous graphs, especially the attention mechanism and the multi-layer structure. These mechanisms bring excessive complexity, but seldom work studies whether they are really effective on heterogeneous graphs. In this paper, we conduct an in-depth and detailed study of these mechanisms and propose the Simple and Efficient Heterogeneous Graph Neural Network (SeHGNN). To easily capture structural information, SeHGNN pre-computes the neighbor aggregation using a light-weight mean aggregator, which reduces complexity by removing overused neighbor attention and avoiding repeated neighbor aggregation in every training epoch. To better utilize semantic information, SeHGNN adopts the single-layer structure with long metapaths to extend the receptive field, as well as a transformer-based semantic fusion module to fuse features from different metapaths. As a result, SeHGNN exhibits the characteristics of a simple network structure, high prediction accuracy, and fast training speed. Extensive experiments on five real-world heterogeneous graphs demonstrate the superiority of SeHGNN over the state-of-the-arts on both accuracy and training speed.

Hepatic steatosis, commonly known as fatty liver is more prevalent in current era owing to lifestyle changes. Fatty liver does not exhibit any symptoms in the early stages. Computer aided diagnosis systems for fatty liver may identify the subtle changes in the liver tissues which go invisible to the clinicians. This study explored the classification of fatty liver disease from ultrasound images using a simple convolutional neural network and a deep neural network. This study highlighted that the simple network with adam optimizer performed well on the small ultrasound dataset compared to a deep neural network.

Thermocouple is one of the widely used temperature sensor used in process industries. The biggest challenge of thermocouple measurements is its inherent nonlinearity. Software-based linearization techniques are more flexible and offer easy implementation using a microcontroller. The published works on software-based linearization have considered the current measurement sample, while the nonlinearity varies across the segments. The present work uses a simple 3-layer recurrent neural network (RNN) based model to consider a short history of immediately preceding predictions (temperatures) along with the current measurement (millivolt) to predict the current output (temperature). The RNN performance was evaluated with Type J, K and T-type thermocouple data from NIST database. The blind test accuracy was evaluated by computing the mean square error (MSE) and mean absolute error (MAE). The MSE (MAE) of 0.15 (0.33), 0.18 (0.34) and 0.11 (0.31) respectively was found from J, K and T-type thermocouples, using a total of 1791 samples from the three datasets. The proposed software-based linearizer offered improved performance over a piecewise linear interpolation and a simple 2-layer multilayer perceptron-based neural network model.

Specimen x-ray imaging provides important information on the margin of surgically excised tissue as well as radiologic and pathologic correlation of the lesion. Similar to breast imaging, where mammograms are digitally processed to enhance readability and lesion conspicuity, specimen images are also processed and enhanced. However, specimen image processing is made challenging by the diversity of specimen containers that are commercially available, compounded by variations in specimen size. In this work, we demonstrate our specimen container and size classification system based on a simple convolutional neural network (CNN), trained to identify the container type. This system allows for automated image processing of the supported container types. A dataset consisting of 1428 HIPAA and IRB-complaint anonymized specimen images were collected. We prepared a simple CNN for image classification with 3 convolutional and 3 fully connected layers, and evaluated the performance based on three comparison metrics. Each network was analyzed in terms of accuracy, multi-class AUC, and via a confusion matrix. The best performing classifier, determined via cross validation, was then used for testing, and evaluated with the same three metrics. The results of training and tuning within cross validation showed that the specimen classes are easily differentiable with this simple convolutional neural network structure. During testing, the network was able to achieve an accuracy of 95.8±4.0%, and an AUC of 0.9763±0.0001.

This paper focuses on implementations of two modified Aihara's chaotic neuron models and a simple chaotic neural network constructed with two chaotic neurons in a programmable and reconfigurable manner with an analog programmable device, FPAA (Field Programmable Analog Array). After testing the chaotic behaviors of two chaotic neuron models and a simple chaotic neural network through numerical analyses that consist of time domain responses, phase portrait illustrations and bifurcation diagrams, the experimental setup is constructed with a FPAA device. The parametric adjustments of chaotic neural structures are possible with the proposed flexible design methodology and different chaotic neuron models are constructed on the same reconfigurable device without any hardware changes. Experimental results verify the dynamic behaviors of these chaotic neural structures and demonstrate the efficiency of programmable implementations.

Due to their synaptic-like characteristics and memory properties, memristors are often used in neuromorphic circuits, particularly neural network circuits. However, most of the existing neural network circuits that can generate complex dynamics have high dimensions and excessive connections, which is not conducive to implementation. This article introduces a memristor containing an arctangent function into a simple cyclic neural network (SCNN) circuit to design a simple cyclic memristive neural network (SCMNN) circuit capable of generating complex multiscroll chaotic attractors. The designed SCMNN contains an external stimulus current and generates multiscroll attractors, with the number of scrolls expanding as the switches in the memristor equivalent circuit are activated. By varying the parameters, the multiscroll attractors can be broken into different numbers of coexisting attractors, which also depends on the switch, and it can achieve multivariable amplitude control when there is only one scroll. The anti-interference ability of the circuit is tested. A low-cost circuit-based microcontroller suitable for engineering applications is designed for it, and multiscroll attractors are successfully captured in an oscilloscope. The National Institute of Standards and Technology (NIST) test is carried out to verify its application value.

In this paper, we propose and demonstrate a low-complexity fiber nonlinearity impairments compensation (NLC) scheme based on simple recurrent neural network (SRNN), combining with the triplets derived from the first-order perturbation solution of the coupled nonlinear Schrodinger equation. Benefitting from the time memory of SRNN, the proposed NLC can reduce the computational complexity without sacrificing the compensation performance compared with the schemes based on memoryless feedforward neural networks. To verify the feasibility of the proposed NLC, the simulation systems of single channel and three channel 30 GBaud polarization division multiplexing (PDM) 16 quadrature amplitude modulation (QAM) over 2400 km standard single mode fiber transmission are constructed respectively. Comprehensive numerical simulation results demonstrate that although the NLC performance of proposed scheme is inferior to that of digital back propagation (DBP) with much higher complexity and impracticality, it can still achieve the competitive NLC performance with lower computational complexity compared to the schemes based on deep neural network (DNN) and artificial neural network (ANN). Moreover, it is further proved that using triplets as the feature required for NLC can cause the pattern of pseudo random bit sequence to be destroyed, resulting in the neural network-based NLC scheme not suffering from the problem of overestimation for performance. Additionally, the time memory of SRNN enables the number of triplets in training set required by the SRNN-based NLC to be reduced by 169% compared to that of DNN and ANN. The experimental results of 16 GBaud PDM-16QAM wavelength division multiplexing coherent systems demonstrate that compared to with ANN and DNN, the comparable NLC performance can also be achieved by the proposed scheme with lower complexity.

The ability for people and organizations to connect in the digital age has allowed the growth of networks that cover an increasing proportion of human interactions. The research community investigating networks asks a range of questions such as which participants are most central, and which community label to apply to each member. This paper deals with the question on how to label nodes based on the features (attributes) they contain, and then how to model the changes in the label assignments based on the influence they produce and receive in their networked neighborhood. The methodological approach applies the simple graph convolutional neural network in a novel setting. Primarily that it can be used not only for label classification, but also for modeling the spread of the influence of nodes in the neighborhoods based on the length of the walks considered. This is done by noticing a common feature in the formulations in methods that describe information diffusion which rely upon adjacency matrix powers and that of graph neural networks. Examples are provided to demonstrate the ability for this model to aggregate feature information from nodes based on a parameter regulating the range of node influence which can simulate a process of exchanges in a manner which bypasses computationally intensive stochastic simulations.