Narrow Results

The embedded system of intelligent reservoir dam achieves the integration and efficient utilization of water conservancy dam system data through multi-channel data collection and analysis calculated by computer technology, CNC system, and neural network. Compared with traditional data collection and processing methods, both timeliness and accuracy have been greatly improved. This study aims to develop a multi-channel sensor data acquisition device for reservoir dams based on embedded system technology. This device can collect real-time and efficient data from sensors in various parts of the dam, ensuring the safe operation of the reservoir dam. By using advanced embedded system technology, this device has advantages such as low power consumption, high stability, and real-time data transmission. The Analytic Hierarchy Process (AHP) was used to study the embedded multi-channel sensor data acquisition device for reservoir dams in multiple directions and factors. The AHP method provides an effective means for problem decision-making in complex situations. Referring to the AHP method, the factors that affect reservoir dams can be divided into different levels. Compare the importance of two random factors in each level to obtain a specific quantitative expression of the relative important factors on a scale. Then repeat this step to obtain the weight ranking for different levels. At the same time, the device monitors key parameters such as temperature, humidity, displacement, and pressure in various parts of the dam through multiple sensors, providing strong support for early warning and decision-making of reservoir dams. The results of this study have important practical significance and application value for improving the safety and stability of reservoir dams.

Information resources have become a very important source of wealth, and due to differences in language and writing, technological information exchange between countries has become difficult. In this article, we propose a data augmentation method based on statistical models. A framework for a multimodal output model is proposed, taking into account the semantic relevance and importance between different languages. This framework is based on text sequence to sequence framework, decoupled, and a network architecture based on dual stream attention mechanism is designed. A multimodal interactive neural network layer was added between the encoder and decoder, achieving the fusion of multilingual domain information. It compensates for the influence of text on traditional translation generation systems in the process of English Chinese translation, and solves the problems of inaccurate translation results and low similarity with the original text in English Chinese machine translation. The system proposed in this article mainly consists of preprocessing module, lexical analysis and segmentation module, part of speech tagging and phrase analysis module, translation rule construction module, decoding module, translation generation module, etc. The experimental results show that using an improved generation system, compared with traditional generation systems, improves the accuracy of translation generation, has certain advantages, and is more practical.

This research focuses on the generation of urban public space color design schemes using Artificial Intelligence (AI). The designed AI-based system predicts colors for the environment, culture, and users and designs a color palette for smart cities. Neural networks and clustering algorithms are used to determine the best hues and shades, depending on the input parameters like climate, architectural style, and general looks in a specific region. It evokes sophisticated colors in varied urban contexts, URBAN FABRICS proposes a specific color answer to the aesthetic and performative beautification of the public realm. In addition to adopting elements of AI binding into the design principles, the system enables dynamic colors to adapt to the changing needs and external conditions. Testing the model in multiple cities demonstrated its ability to generate unique, context-sensitive designs, improving both aesthetic value and user satisfaction. This research highlights the role of AI in modern urban planning, presenting an innovative approach to color design that balances artistic creativity with data-driven insights. The findings offer practical implications for architects, urban planners, and designers seeking to enhance urban public spaces through personalized, AI-assisted color schemes.

The simulation of vocal performance has emerged as an important means of training singers in a more engaging manner than standard methods of vocal coaching. The purpose of this research is to present a new vocal performance simulation and training system developed using neural networks (NN) in a virtual reality (VR) environment and to improve the training of vocal performance using advanced NN models and VR. For this purpose, the researcher’s audio is primarily to analyze, collect and process samples recording performance as well as specific performers’ training audio-video material. Preprocessing techniques such as noise reduction, and normalization, and applied to prepare the data. Key features like pitch, tone, and breath control were extracted using the Mel-frequency cepstral coefficients (MFCC) algorithm, enabling effective feature representation. We propose the Refined Fruit Fly Optimized Intelligent Long-Short Term Memory (RFF-ILSTM) model, a recurrent neural network (RNN)-based approach optimized for handling sequential vocal data with high precision. The model incorporates the RFF optimization technique in order to enhance the tuning of the LSTM architecture thereby increasing its speed of convergence and training. The information collected was already processed by the proposed framework in an effort to enhance vocal performance simulation. VR integration enhances the enjoyment and effectiveness of the interface as it allows singers to receive immediate feedback as if they were performing in front of an audience. Users engage in performing activities based on 3D images, thereby allowing the practice of different types of vocations without difficulty. Simulation results showed dramatic improvements in terms of ability to control vocals and tones, as well as consistency in performance. The system successfully integrates VR and NN enhancement for an improved, interactive system of training for vocal performances.

This study presents a novel framework for assessing physical education (PE) performance by integrating principal component analysis (PCA) and neural networks (NN). A comprehensive sports measurement system was developed, incorporating diverse indicators such as physical endurance, skill proficiency, and teamwork. PCA was applied to extract key performance indicators, reducing data dimensionality while retaining critical information. These refined indicators were used as inputs for an NN model, which provided detailed and objective evaluations of student performance. The proposed framework demonstrated superior accuracy, F1-scores, and recall compared to traditional methods and advanced machine learning models, such as SVM, RF, and GBM. Furthermore, the insights generated by the model enabled the design of personalized training plans tailored to individual student needs. This data-driven approach offers a significant advancement in PE assessment, ensuring objective evaluations and fostering effective, individualized teaching strategies.

Prediction of stock prices has always been a hot research topic. Internet of Things (IoT) devices can collect a large amount of data in real time, including market trends, consumer behavior, economic indicators, etc., providing valuable references for stock price prediction. The stock market is the barometer of the national economy, and the prediction of stock price fluctuation can not only provide a reference for investors and enterprises to set reasonable goals, make decisions and risk management, but also have great reference significance for the government to supervise and manage the market and formulate policies. In this paper, we use the quantum swarm intelligence algorithm to optimize the neural network method to predict the stock closing prices of two listed artificial intelligence companies, Dawning Digital Creation and Yunchuang Data and then analyze the data through IoT devices. In the stock price prediction results, the quantum particle swarm has the highest prediction efficiency and accuracy value among the four quantum optimization models.

This study aims to exploit the Artificial intelligence (AI)-based computing paradigm to analyze the economic system to define the price movements of unsustainable expansion, rapid collapse, and eventual equilibrium that characterize financial bubbles represented with differential equations to portray the role of societal contagion and group mentality from a behavioral viewpoint with the market population classified as bull, i.e., optimistic, neutrals, bear, i.e., pessimistic, and quitter categories. The concept of the financial bubble is characterized as an unexpected rise in prices that is rapidly followed by a shrill decline and retrospectively appears as a consequence of such uncertainty in price value. AI-based applications facilitate financial analysts with innovative computational paradigms for gaining deep insights, improving predictive accuracy, and creating sustained vigorous risk management stratagems for the financial bubble framework and solving with supervised nonlinear autoregressive exogenous networks with optimized Bayesian regularization algorithm to accomplish the reasonable predictive accuracy and malleability for the solution of financial bubble behavioral dynamics. The Adams numerical solver accomplishes the acquisition of synthetic data for the execution of a multi-layer structure of exogenous networks to solve for financial bubble parameters termed as contagion rate of optimistic behavior, bearish behavior, pessimist’s average time of staying in the bearish group, the pessimist’s population effect on the autonomous supply and the rate of optimists conversion to pessimists group while assuming other parameters values to be fixed for demand and supply functions. A consistent overlap between proposed results and synthetic numerical values of a financial bubble model is indicated by negligible error value that verifies the exogenous network’s effectiveness and is verified by the enclosure of several evaluation measures of the precision and efficiency, through mean square error objective functions, adaptive amendable parameters, error dispersal, and input-error correlation analyses.

Human activity recognition (HAR) employs a broad range of sensors that generate massive volumes of data. Traditional server-based and cloud computing methods require all sensor data to be sent to servers or clouds for processing, which leads to high latency and bandwidth costs. The long-term data transfer between servers and sensors maximizes the cost of latency and bandwidth. Real-time processing is, nevertheless, highly required for human action identification. By bringing processing and quick data storage to the sensors instead of depending on a central database, edge computing is rapidly emerging as a solution to this issue. Artificial intelligence is responsible for most HAR, which demands a lot of processing power and calculation. Artificial intelligence (AI) needs more computation which is not allowed by edge computing. So Edge intelligence, which allows AI to operate at the network edge for actual-time applications, has been made possible by the advent of binarized neural networks. To provide less latency and less memory for human activity identification at the edge network, we construct a hybrid deep learning-based binarized neural network (HDL-Binary Dilated DenseNet) in this research. Fractal HAR optimization algorithms could be applied to these algorithms. For example, fractal-HAR optimization techniques might be used to provide less latency and less memory human activity identification at the edge network. Using three sensors-based human activity detection datasets such as Radar HAR dataset, UCI HAR dataset and UniMib-SHAR dataset, we implemented the Hybrid Binary Dilated Dense Net. It is then assessed using four criteria. Comparatively, the Hybrid Binary Dilated DenseNet performs better with 99.6% radar HAR dataset which is highest than other models like CNN-BiLSTM and GoogLeNet.

Predicting a client’s affluence is essential in financial services. This task is the unity of the most important danger factors in groups and additional economic institutions. Typically, credit risk evaluation relies on black box models. However, these models often need to clarify the hidden information within the data. Moreover, few clear models focus on being easy to understand and accessible. This paper proposes a fractal credit risk assessment model that uses deep techniques like self-attention generative adversarial networks (SA-GAN) and deep multi-layer perceptron (DMLP). We use blockchain technology with the Brakerski–Gentry–Vaikuntanathan (BGV) encryption method to bolster safekeeping. Additionally, the scheme is designed for the Edge-of-things network, enabling communication through a LoRaWAN server. The proposed solution was tested on the German retail credit dataset. We assessed its performance using accuracy, F1 score, precision, and recall as metrics. Notably, our hybrid deep model, which combines SA-GAN with DMLP, achieved an impressive accuracy of 97.8% — outperforming existing methods in works.

Cryptocurrency (CRP) has grown in popularity over the last decade. Since there is no central body to control the Bitcoin (BTC) markets, they are extremely volatile. However, several similar variables that cause price volatility in traditional markets also affect cryptocurrencies. Several bubble phases have taken place in BTC prices, mostly during the years 2013 and 2017. Other digital currencies of primary importance, such as Ethereum and Litecoin, also exhibited several bubble phases. Among traditional methods of analysis for this volatile market, only a small number of studies focused on Machine Learning (ML) techniques. The present study objective is to get an in-depth knowledge of the time series properties of CRP data and combine volatility models with ML models. In the hybrid method, we first apply the Nonlinear Generalized Autoregressive Conditional Heteroskedasticity (NGARCH) model with asymmetric distribution to calculate standardized returns, then forecast the UP and DOWN movement of standardized returns through ML models such as Logistic Regression (LR), Linear Discrimination Analysis (LDA), Quadratic Discrimination Analysis (QDA), Artificial Neural Networks (ANNs), K-Nearest Neighbors (KNN), and Support Vector Machine (SVM). The findings show that the proposed hybrid approach of time series models and ML accurately predicts prices; specifically, the KNN model reveals that the scheme can be applicable to CRP market prediction. It is deduced that ML methods combined with volatility models have the tendency to better forecast this volatile market.

The development of modern Internet communication technology and the popularization of multimedia technology have brought convenience to the sharing and storage of multimedia information such as images, videos, and audio. However, at the same time, it has brought about the problem of copyright theft of multimedia information, causing serious information security risks. Digital watermarking technology embeds copyright information in multimedia information in an invisible way, which can effectively realize copyright protection and traceability of infringement. Aiming at the problem that the existing learning model-based methods cannot fully extract and fuse the features of carrier images and watermark images, a robust digital watermarking method based on reversible information hiding is proposed. First, a watermark embedding model based on reversible information hiding is established, and the features of the download volume image and the watermark image in different dimensions are fully extracted and fused to generate a dense image with excellent visual quality. Then, a watermark extraction model based on reversible information hiding is established, and a noise layer is added between the embedding and the extraction model, and the attacked dense image is input to the watermark extraction model to extract the watermark. Under the constraint of the loss function, the network model learns to embed watermark information in the area that is more robust to the attack and is not easy to cause visual quality degradation, so as to optimize the comprehensive performance of the method. Experimental results show that the proposed method effectively improves the imperceptibility and robustness.

Chemistry evaluation is a bottleneck to computational fluid dynamics (CFD) simulations of many real-life problems such as propulsion system design, engine diagnostics, and atmospheric modeling. In this work, we study approach for accelerating chemical kinetics calculations using artificial neural networks (ANNs) on the example of combustion of a hydrogen–air mixture. This work carries out a detailed exploratory study of the optimal design of a fully connected neural network, including the number of network parameters, number of layers as well as used activation function. Part of the work is also dedicated to investigation and optimization of network training process itself. Comparison with the results of other works, bringing some unification to the widely disparate reported results, is also performed. Links to the used datasets and the resulting neural network are provided.

Artificial intelligence has become the most widely used and trusted component of research in almost all disciplines of science and technology, starting from engineering, online businesses, and industry, to biotechnology and agriculture. Successful rice crops with maximum yield and weed management are the target set by several developed as well as developing countries, based on a combination of cultural and chemical control methods. In this paper, the weed control strategy through the competition model is documented with the aid of the time-series forecasting tool of artificial intelligence. A time-dependent computational framework is built based on the real data, and by utilizing supervised learning algorithm, incorporated with delay. It is emphasized during this research that the accurate precision of time delays in competition models can help in developing the weed control strategies more efficiently and can further support in implementing these strategies as precision models for other crop protection challenges.

The advent of the Connection Machine profoundly changes the world of supercomputers. Its highly nontraditional architecture makes possible the exploration of algorithms that were impractical for standard Von Neumann architectures. Kanerva’s sparse distributed memory (SDM) is an example of such an algorithm.

Sparse distributed memory is a particularly simple and elegant formulation for an associative memory. In this paper I describe the foundations for sparse distributed memory, and give some simple examples of using the memory. I continue by showing the relationship of sparse distributed memory to random-access memory. Finally, I discuss the implementation of the algorithm for sparse distributed memory on the Connection Machine.

The Artificial Neural Networks (ANNs) are being used to solve a variety of problems in pattern recognition, robotic control, VLSI CAD and other areas. In most of these applications, a speedy response from the ANNs is imperative. However, ANNs comprise a large number of artificial neurons, and a massive interconnection network among them. Hence, implementation of these ANNs involves execution of computer-intensive operations. The usage of multiprocessor systems therefore becomes necessary.

In this article, we have presented the implementation of ART1 and ART2 ANNs on ring and mesh architectures. The overall system design and implementation aspects are presented. The performance of the algorithm on ring, 2-dimensional mesh and n-dimensional mesh topologies is presented. The parallel algorithm presented for implementation of ART1 is not specific to any particular architecture. The parallel algorithm for ART2 is more suitable for a ring architecture.

We prove that the modulus function MOD_m of n variables can be computed by a threshold circuit C of energy e and size s = O(e(n/m)^{1/(e − 1)}) for any integer e ≥ 2, where the energy e is defined to be the maximum number of gates outputting "1" over all inputs to C, and the size s to be the number of gates in C. Our upper bound on the size s almost matches the known lower bound s = Ω(e(n/m)^1/e). We also consider an extreme case where threshold circuits have energy 1, and prove that such circuits need at least 2^{(n − m)/2} gates to compute MOD_m of n variables.

Backpropagation, which is frequently used in Neural Network training, often takes a great deal of time to converge on an acceptable solution. Momentum is a standard technique that is used to speed up convergence and maintain generalization performance. In this paper we present the Windowed Momentum algorithm, which increases speedup over Standard Momentum. Windowed Momentum is designed to use a fixed width history of recent weight updates for each connection in a neural network. By using this additional information, Windowed Momentum gives significant speedup over a set of applications with same or improved accuracy. Windowed Momentum achieved an average speedup of 32% in convergence time on 15 data sets, including a large OCR data set with over 500,000 samples. In addition to this speedup, we present the consequences of sample presentation order. We show that Windowed Momentum is able to overcome these effects that can occur with poor presentation order and still maintain its speedup advantages.

The organization of video information in video databases requires automatic temporal segmentation with minimal user interaction. As neural networks are capable of learning the characteristics of various video segments and clustering them accordingly, in this paper, a neural network based technique is developed to segment the video sequence into shots automatically and with a minimum number of user-defined parameters. We propose to employ growing neural gas (GNG) networks and integrate multiple frame difference features to efficiently detect shot boundaries in the video. Experimental results are presented to illustrate the good performance of the proposed scheme on real video sequences.

Bayesian belief networks (BBN) are a widely studied graphical model for representing uncertainty and probabilistic interdependence among variables. One of the factors that restricts the model's wide acceptance in practical applications is that the general inference with BBN is NP-hard. This is also true for the maximum a posteriori probability (MAP) problem, which is to find the most probable joint value assignment to all uninstantiated variables, given instantiation of some variables in a BBN . To circumvent the difficulty caused by MAP's computational complexity, we suggest in this paper a neural network approximation approach. With this approach, a BBN is treated as a neural network without any change or transformation of the network structure, and the node activation functions are derived based on an energy function defined over a given BBN. Three methods are developed. They are the hill-climbing style discrete method, the simulated annealing method, and the continuous method based on the mean field theory. All three methods are for BBN of general structures, with the restriction that nodes of BBN are binary variables. In addition, rules for applying these methods to noisy-or networks are also developed, which may lead to more efficient computation in some cases. These methods' convergence is analyzed, and their validity tested through a series of computer experiments with two BBN of moderate size and complexity. Although additional theoretical and empirical work is needed, the analysis and experiments suggest that this approach may lead to effective and accurate approximation for MAP problems.

In this paper multilayer neural networks (MNNs) are used to control the balancing of a class of inverted pendulums. Unlike normal inverted pendulums, the pendulum discussed here has two degrees of rotational freedom and the base-point moves randomly in three-dimensional space. The goal is to apply control torques to keep the pendulum in a prescribed position in spite of the random movement at the base-point. Since the inclusion of the base-point motion leads to a non-autonomous dynamic system with time-varying parametric excitation, the design of the control system is a challenging task. A feedback control algorithm is proposed that utilizes a set of neural networks to compensate for the effect of the system's nonlinearities. The weight parameters of neural networks updated on-line, according to a learning algorithm that guarantees the Lyapunov stability of the control system. Furthermore, since the base-point movement is considered unmeasurable, a neural inverse model is employed to estimate it from only measured state variables. The estimate is then utilized within the main control algorithm to produce compensating control signals. The examination of the proposed control system, through simulations, demonstrates the promise of the methodology and exhibits positive aspects, which cannot be achieved by the previously developed techniques on the same problem. These aspects include fast, yet well-maintained damped responses with reasonable control torques and no requirement for knowledge of the model or the model parameters. The work presented here can benefit practical problems such as the study of stable locomotion of human upper body and bipedal robots.