Search

In the era of wireless communication, the MIMO technology contributes the communication among users with high data rate and spectral efficiency. The interference of data transmission along the multipath channel is tackled by the OFDM of the MIMO system. However, the existing works didn’t focus on efficient signal modulation, thus resulting in sub-optimal beam forming and inaccurate channel estimation in the MIMO system. Hence, in this paper, an efficient CE using the Universal Approximation Layer-based P-ReLU DNN (UAL-P-DNN), Line Sweep-Multi-Variate Kernel Density estimation (LS-MVKD), and two-dimensional Gray Codes Quadratic Amplitude Modulation (GCQAM) is proposed. Here, in this paper, the UAL-P-DNN is mainly used for the CE, LS-MVKD is used for the prior estimation of the channel, and GCQAM is adopted for the signal modulation process. In the proposed methodology, the process begins with the transmission. In the transmission phase, the data bits of input text, audio, and video are randomly generated and encoded using Montgomery Curve-based Elliptic Curve Cryptography (MC-ECC). Then, the exact data bits to be transmitted are extracted by the Rate matching. Further, the extracted bits are mapped using GCQAM for modulating the signal amplitude. After that, the modulated signal along with the pilot symbol is given to the IFrFT for converting signal characteristics to the time domain. Subsequently, the message bits are transmitted along the Rayleigh fading channel, and the prior channel is estimated using LS-MVKD. Based on the obtained signal matrix, input data, and target data, the channel is estimated using the UAL-P-DNN model. In the receiver phase, the original data is retrieved by inversely performing the operations, which are done on the transmitter. The executed results show that the proposed system estimated a channel with 54.485db PSNR and 0.29% BER. Also, the input data is encoded by the proposed method within 2130 ms and 98.46% security level.

One of the main issues facing in the future wireless communications is ultra-reliable and low-latency communication. Polar codes are well-suited for such applications, and recent advancements in deep learning have shown promising results in enhancing polar code decoding performance. We propose a robust decoder based on a bidirectional long short-term memory (Bi-LSTM) network, which processes sequences in both forward and backward directions simultaneously. This approach leverages the strengths of bidirectional recurrent neural networks to improve the decoding of polar-coded short packets. Our study focuses on packet transmission over frequency-flat quasi-static Rayleigh fading channels, using a simple codebook originally designed for additive white Gaussian noise channels. We evaluate the packet error rate for various signal-to-noise ratio levels using different modulation schemes. The simulation results demonstrate that the proposed Bi-LSTM-based decoder closely approaches the theoretical outage performance and achieves significant coding gains in fading channels. Furthermore, the proposed decoder outperforms convolutional neural network and deep neural network-based decoders, validating its superiority in decoding polar codes for short packet transmission in challenging wireless environments.

In this paper, we propose a method based on deep neural networks for the statistical design of flip-flops, taking into account nonlinear performance constraints. Flip-flop design and manufacturing are influenced by random variations in the technological process, making deterministic design approaches inadequate for achieving high yields. The conventional yield maximization method using Monte Carlo (MC) simulation is a time-consuming process. Also, for many performance constraints, either there are no analytical formulations or if they exist, they are not sufficiently accurate to be used in circuit optimization. To address these challenges, we approximated the nonlinear constraints with linearized ones (polyhedral approximation) and performed a yield maximization process which was done by developing our first proposed method. Then in the second proposed method, we used deep neural networks to generate precise nonlinear closed-form models for circuit performance metrics and also replaced MC simulation with an analytical yield formula. The combination of these techniques significantly enhances the speed and accuracy of statistical circuit design by employing powerful gradient-based optimization methods that converge quickly to the optimal solution. Experimental results demonstrate that our proposed approach enables the design of circuits with various performance constraints under process variation, and achieves more optimum results with much fewer iterations and less CPU time compared to the conventional simulation-based yield maximization methods.

Automatic Speech Recognition (ASR) systems have improved and eased how humans interact with devices. ASR system converts an acoustic waveform into the relevant text form. Modern ASR inculcates deep neural networks (DNNs) to provide faster and better results. As the use of DNN continues to expand, there is a need for examination against various adversarial attacks. Adversarial attacks are synthetic samples crafted carefully by adding particular noise to legitimate examples. They are imperceptible, yet they prove catastrophic to DNNs. Recently, adversarial attacks on ASRs have increased but previous surveys lack generalization of the different methods used for attacking ASR, and the scope of the study is narrowed to a particular application, making it difficult to determine the relationships and trade-offs between the attack techniques. Therefore, this survey provides a taxonomy illustrating the classification of the adversarial attacks on ASR based on their characteristics and behavior. Additionally, we have analyzed the existing methods for generating adversarial attacks and presented their comparative analysis. We have clearly drawn the outline to indicate the efficiency of the adversarial techniques, and based on the lacunae found in the existing studies, we have stated the future scope.

By 2050, zero-carbon electric power systems will rely heavily on innumerable distributed energy resources (DERs), such as wind and solar. Accurate estimation of the aggregate connected DER capacity becomes pivotal in such a landscape. However, forecasting, power flow analysis, and optimization of feeders for operational decision-making by individually modeling each of these numerous renewables in the absence of complete information are operationally challenging and technically impractical. In response, we introduce a method to accurately estimate the aggregate capacities of the connected DERs on distribution feeders and a near-term forecasting method. Our proposal comprises: 1) ovel deep learning-based architecture with a few convolutional neural network and long short-term memory (CNN-LSTM) modules to represent feeder connected aggregate models of DERs and loads and associated training algorithms; 2) method for estimating aggregate capacities of connected renewables and loads; and 3) method for short-term (hourly) high-resolution forecasting. This step of estimation of the aggregate capacities of connected DERs, is a sequel to solving feeder hosting capacity problem. The method is tested using a North American utility feeder data, achieving an average accuracy of 95.56% for forecasting aggregate load power, 93.70% for feeder flow predictions, and 97.53% for estimating the aggregate capacity of DERs.

This study compares the performance of five selected machine learning models regarding precipitation climatology during the warm season in 2022 and 2023 over the continental U.S. Input features included retrieved products from the microwave integrated retrieval system (MiRS) based on NOAA-20 ATMS data. The radar-based instantaneous multiradar multisensor system precipitation was used for model training and validation. Among the models, three used a U-Net architecture and two used a deep neural network (DNN) architecture. The U-Net models all significantly outperformed the DNN models for the evaluated metrics. While the DNN architecture can only learn from local inputs, the U-Net also has the capability to learn from neighborhood spatial patterns. As such, the DNN overcorrected the precipitation amounts that MiRS had overestimated, leading to net underestimation, but also failed to improve the overall performance relative to the original MiRS estimates. The U-Net not only corrected MiRS overestimation in the central U.S., but also improved the MiRS dry bias over the Southeast. Among the five experiments, the one that used the MiRS retrieved column-integrated hydrometeors of graupel water path, rainwater path, cloud liquid water, total precipitable water, and geolocation information demonstrated the best performance, improving the MiRS spatial correlation coefficient from 0.75 to 0.89 and reducing the mean bias percentage from 11.95% to −6.33% for 2022 accumulated precipitation. This suggests that applying an appropriate architecture and input features provides an opportunity to determine more accurate physical and statistical relationships which can include spatial and regional dependence, leading to improved microwave-based precipitation estimates.

This paper presents an efficient deep neural network (DNN) accelerator designed for radio frequency (RF) signal modulation recognition. A novel DNN design optimized for mobile applications is demonstrated by combining MobileNetV3-based DNN with a ternary weight quantization. We also propose a new training method called decaying weight training to overcome the performance degradation due to quantization. The effect of the ternary weight quantization is demonstrated with a co-analysis of the classification accuracy and the physical design. The physical design analysis is based on the Application Specific Integrated Circuit (ASIC), and the results show that the ternary weight quantization with the proposed training method minimizes the impact of the quantization while increasing the allowable clock frequency and reducing hardware cost significantly. We also implement the hardware design dedicated to the ternary weight networks to reduce the required number of the multiply and accumulate (MAC) engines. The hardware design is verified on FPGA and the ternary weight-based DNN shows the feasibility of reducing the hardware cost significantly.

In the network structure analysis, we explore an underestimated key metric, the Relative Size of Largest Connected Component (RSLCC) and demonstrate its importance in post-disaster network connectivity assessment. RSLCC was first investigated in the study of complex network structures but remains largely unexplored in terms of analysis within a specific application domain such as scenarios in transportation networks, wireless networks, communication networks, power networks, etc. Through the research presented in this paper, we not only prove that this metric is underestimated, but also design 7 methods to predict the value of this metric, with a Deep Neural Network (DNN) prediction accuracy of more than 99%. This study focuses on the assessment and analysis of post-disaster network connectivity, by exploring how the RSLCC, a key metric of network connectivity, can be used to efficiently predict and assess network connectivity in a disaster scenario, specifically, the approximate network connectivity value can be predicted simply by knowing the number of connected edges in the pre-disaster network and the number of connected edges in the post-disaster network. To achieve this, firstly, a sufficiently large-scale 100,000 datasets containing the values of attributes related to the network structure is prepared. Secondly, based on the preprocessing of the data, principal component analysis and variance contribution analysis are carried out, and the metric with the highest contribution to the principal component is approximated as the network connectivity. The next step is the prediction process, Network Disruption Degree (NDD) is chosen as the independent variable. since it is best to choose an extremely simple metric as the independent variable for prediction, rather than all network structure-related metrics, this paper demonstrates that it is possible to get satisfactory prediction results with this metric. It is found that NDD prediction methods have the highest prediction accuracy but take the longest run time and require training data of a sufficiently large size. If the prediction is done in small-size data, then Random Forest Regression (RFR) is proven to have the highest prediction accuracy. Although the network connectivity metric proposed in this paper is only an approximation, it provides good directions for simplifying the network connectivity analysis and the use of this metric for the study of practical modelling problems is also highly interpretable.

Synthetic aperture radar automatic target recognition (SAR-ATR) models based on deep neural networks (DNNs) are vulnerable to attacks of adversarial examples. Universal adversarial attack algorithms can help evaluate and improve the robustness of the SAR-ATR models and have become a research hotspot. However, current universal adversarial attack algorithms have limitations. First, considering the difficulty in obtaining information on the attacking SAR-ATR models, there is an urgent need to design a universal adversarial attack algorithm under a black-box scenario. Second, given the difficulty of acquiring synthetic aperture radar images, the effectiveness of attacks under small-sample conditions requires improvement. To address these limitations, this study proposed a black-box universal adversarial attack algorithm: transferable universal adversarial network (TUAN). Based on the idea of the generative adversarial network, we implemented the game of generator and attenuator to improve the transferability of universal adversarial perturbation (UAP). We designed loss functions for the generator and the attenuator, respectively, which can effectively improve the success rate of black-box attacks and the stealthiness of attacks. In addition, U-Net was used as a network structure of the generator and the attenuator to fully learn the distribution of examples, thereby enhancing the attack success rate under small-sample conditions. The TUAN attained a higher black-box attack success rate and superior stealthiness than up-to-date UAP algorithms in non-targeted and targeted attacks.

The emergence of deep-learning technology has significantly improved the performance of super-resolution algorithms for a single remote sensing image; however, the number of deep-learning model parameters is large, which limits its real-time deployment. In addition, the reconstructed image quality still needs improvement. To deal with the aforementioned problems, a lightweight multiscale residual attention information distillation network is proposed in this article. It achieves high-quality and fast super-resolution processing of remote sensing images. First, a high- and low-frequency (HF and LF) separation reconstruction strategy is adopted that enables the network to improve the reconstructed details of HF components while keeping the number of model parameters low. Second, a novel multiscale residual attention information distillation group is designed as the key component to further extract richer regional features with different perceptual fields and HF information while reducing the number of network parameters. This is achieved by combining a multiscale residual information distillation block that consists of multiple residual convolutional sub-blocks and an HF channel aware attention block. Last, the experimental results show that, compared with existing mainstream methods, such as the MHAN, the number of model parameters can be reduced by 75%, and the edge details of the reconstructed images are richer and more complete. The corresponding peak signal-to-noise ratio and SSIM can reach 31.59 dB and 0.824, respectively, under the condition of a $ \times 4$ magnification factor, and 27.39 dB and 0.668, respectively, under the condition of a $ \times 8$ magnification factor.

The ability to constantly monitor urban changes is of significant socio-economic interest, such as detecting trends in urban expansion or tracking the vitality of urban areas. Especially in present conflict zones or disaster areas, such insights provide valuable information to keep track of the current situation. However, they are often subject to limited data availability in space and time. We built on our previous work, which used a transferred deep neural network operating on multimodal Sentinel 1 and 2 data. In the current study, we have demonstrated and discussed its applicability in monitoring the present conflict zone of Mariupol, Ukraine, with high-temporal resolution Sentinel time series for the years 2022/23. A transfer to that conflict zone was challenging due to the limited availability of recent very high resolution (VHR) data. The current work had two objectives. First, transfer learning with older and publicly available VHR data was shown to be sufficient. That guaranteed the availability of more and less expensive data as time constraints were relaxed. Second, in an ablation study, we analyzed the effects of loss of observations to demonstrate the resiliency of our method. That was of particular interest due to the malfunctioning of Sentinel 1B shortly before the selected conflict. Our study demonstrated that urban change monitoring is possible for present conflict zones after transferring with older VHR data. It also indicated that, despite the multimodal input, our method was more dependent on optical multispectral than synthetic aperture radar observations but resilient to loss of observations.

This article presents a length-dependent deep neural network (LD-DNN) based channel modeling methodology to predict the frequency response of high-speed channels. The proposed method significantly enhances the model accuracy and design efficiency while considering the channel length dependence that was neglected in previous modeling approaches. We define the concept of the electrical length to model the length and frequency dependence, then further leverage the activation function to capture the multiple reflection effects to improve accuracy. Additionally, we model the insertion loss resonance induced by crosstalk that can seriously deteriorate signal integrity. As a result, by adopting the proposed model which can predict the S-parameters as a function of length, the need for performing additionally 3D electromagnetic simulations when adjusting the channel length can be eliminated. Various high-speed channel cases are tested to validate the accuracy of the proposed method. The modeling accuracy is less than 4% for different high-speed channel structures with run times of less than 1.4 second per design.

In this work, different deep learning approaches based on deep neural networks for person re-identification were analyzed. Both identification and re-identification of people are frequently required in various fields of human life. Some of the most common applications are in various security systems where it is necessary to identify and track a particular person. In the case of person identification, the identity of a particular person needs to be established. In the case of re-identification, the main task is to match the identity of a particular person across different, non-overlapping cameras or even with the same camera at different times. In this work, three different deep neural networks were used for the purpose of person re-identification. Two of them were user-defined, while one of them is a pre-trained neural network adapted to work with a specific dataset. Two neural networks used were Convolutional Neural Networks (CNN). For the defined experiment, it was used own dataset with 13 subjects in gait.

Abstract Deep neural networks (DNN) are highly effective in a number of tasks related to machine learning across different domains. It is quite challenging to apply the information gained to textual data because of its graph representation structure. This article applies innovative graph structures and protection techniques to secure wireless systems and mobile computing applications. We develop an Intrusion Detection System (IDS) with DNN and Support Vector Machine (SVM) to identify adversarial inversion attacks in the network system. It employs both normal and abnormal adversaries. It constantly generates signatures, creates attack signatures, and refreshes the IDS signature repository. In conclusion, the assessment indicators, including latency rates and throughput, are used to evaluate the effectiveness and efficiency of the recommended framework with Random Forest. The results of the proposed model (SVM with DNN) based on adversarial inversion attacks were better and more efficient than traditional models, with a detection rate of 93.67% and 95.34% concerning latency rate and throughput. This article also compares the proposed model (SVM with DNN) accuracy with other classifiers and the accuracy comparison for feature datasets of 90.3% and 90%, respectively.

Abstract Background Plant defense activators offer advantages over pesticides by avoiding the emergence of drug-resistant pathogens. However, only a limited number of compounds have been reported. Reactive oxygen species (ROS) act as not only antimicrobial agents but also signaling molecules that trigger immune responses. They also affect various cellular processes, highlighting the potential ROS modulators as plant defense activators. Establishing a high-throughput screening system for ROS modulators holds great promise for identifying lead chemical compounds with novel modes of action (MoAs). Results We established a novel in silico screening system for plant defense activators using deep learning-based predictions of ROS accumulation combined with the chemical properties of the compounds as explanatory variables. Our screening strategy comprised four phases: (1) development of a ROS inference system based on a deep neural network that combines ROS production data in plant cells and multidimensional chemical features of chemical compounds; (2) in silico extensive-scale screening of seven million commercially available compounds using the ROS inference model; (3) secondary screening by visualization of the chemical space of compounds using the generative topographic mapping; and (4) confirmation and validation of the identified compounds as potential ROS modulators within plant cells. We further characterized the effects of selected chemical compounds on plant cells using molecular biology methods, including pathogenic signal-triggered enzymatic ROS induction and programmed cell death as immune responses. Our results indicate that deep learning-based screening systems can rapidly and effectively identify potential immune signal-inducible ROS modulators with distinct chemical characteristics compared with the actual ROS measurement system in plant cells. Conclusions We developed a model system capable of inferring a diverse range of ROS activity control agents that activate immune responses through the assimilation of chemical features of candidate pesticide compounds. By employing this system in the prescreening phase of actual ROS measurement in plant cells, we anticipate enhanced efficiency and reduced pesticide discovery costs. The in-silico screening methods for identifying plant ROS modulators hold the potential to facilitate the development of diverse plant defense activators with novel MoAs.

Explorative simulation experiment is an important mean to understand and study warfare, but it often faces problems such as high complexity of scenario sample space and explosion of space dimensions. To solve all above problems, a qualitative and quantitative method for screening experimental factors based on multi-label learning(MLL)is proposed. First, based on qualitative analysis, a pre-simulation experiment is designed and implemented, experimental data are collected and processed, and the problem of missing machine learning sample data is solved. Then, the input control layer is introduced to build a deep neural network, and sparse regularization is introduced to achieve feature selection during multi-label model training. Then, regression qualitative analysis is used to supplement and improve the experimental factors. Finally, the simulation of stereoscopic projection action is used for experimental verification. The experimental result shows that the method is feasible and valuable in military.

Additive Manufacturing (AM) is a pivotal technology for transforming complex geometries with minimal tooling requirements. Among the several AM techniques, Wire Arc Additive Manufacturing (WAAM) is notable for its ability to produce large metal components, which makes it particularly appealing in the aerospace sector. However, precise control of the bead geometry, specifically bead width and height, is essential for maintaining the structural integrity of WAAM-manufactured parts. This paper introduces a methodology using a Deep Neural Network (DNN) model for forecasting the bead geometry in the WAAM process, focusing on gas metal arc welding cold metal transfer (GMAW-CMT) WAAM. This study addresses the challenges of bead geometry prediction by developing a robust predictive framework. Key process parameters, such as the wire travel speed, wire feed rate, and bead dimensions of the previous layer, were monitored using a Coordinate Measuring Machine (CMM) to ensure precision. The collected data were used to train and validate various regression models, including linear regression, ridge regression, regression, polynomial regression (Quadratic and Cubic), Random Forest, and a custom-designed DNN. Among these, the Random Forest and DNN models were particularly effective, with the DNN showing significant accuracy owing to its ability to learn complex nonlinear relationships inherent in the WAAM process. The DNN model architecture consists of multiple hidden layers with varying neuron counts, trained using backpropagation, and optimized using the Adam optimizer. The model achieved mean absolute percentage error (MAPE) values of 0.014% for the width and 0.012% for the height, and root mean squared error (RMSE) values of 0.122 for the width and 0.153 for the height. These results highlight the superior capability of the DNN model in predicting bead geometry compared to other regression models, including the Random Forest and traditional regression techniques. These findings emphasize the potential of deep learning techniques to enhance the accuracy and efficiency of WAAM processes.

The upcoming deep neural architectures overpower other previously proposed techniques for named entity recognition by involving fewer efforts but improving accuracy. In the present study, an important variant of recurrent neural network namely, a Bidirectional Long Short-Term Memory which is specially used for sequence classification problems has been used for the NER task. Word contextual features play an important role in the correct prediction of named entities. The present work contributes to developing novel word embeddings i.e. boosted word embeddings which can learn contextual features efficiently. Boosted word embeddings are the combination of character-based convolutional embeddings, part of speech embeddings, and word length embeddings. The results obtained using boosted word embeddings include f-score values of 73.99%, 66.94% and 77.95% respectively for Hindi, Punjabi and bilingual Hindi and Punjabi named entity recognition system. Boosted word embeddings are found effective in raising the accuracy of the named entity recognition model as compared to other models already present in the literature. [ABSTRACT FROM AUTHOR]

In this paper, we explore the problem of direction-of-arrival (DOA) estimation for a non-uniform linear array (NULA) under strong noise. The compressed sensing (CS)-based methods are widely used in NULA DOA estimations. However, these methods commonly rely on the tuning of parameters, which are hard to fine-tune. Additionally, these methods lack robustness under strong noise. To address these issues, this paper proposes a novel DOA estimation approach using a deep neural network (DNN) for a NULA in a low SNR. The proposed network is designed based on the denoising convolutional neural network (DnCNN) and the alternating direction method of multipliers (ADMM), which is dubbed as LDnADMM-Net. First, we construct an unfolded DNN architecture that mimics the behavior of the iterative processing of an ADMM. In this way, the parameters of an ADMM can be transformed into the network weights, and thus we can adaptively optimize these parameters through network training. Then, we employ the DnCNN to develop a denoising module (DnM) and integrate it into the unfolded DNN. Using this DnM, we can enhance the anti-noise ability of the proposed network and obtain a robust DOA estimation in a low SNR. The simulation and experimental results show that the proposed LDnADMM-Net can obtain high-accuracy and super-resolution DOA estimations for a NULA with strong robustness in a low signal-to-noise ratio (SNR). [ABSTRACT FROM AUTHOR]

The non-orthogonal multiple-access (NOMA) technique affords massive user scaling, industrial use for mission-critical communication, and multi-user shared in a single resource block. Because of these merits, it is a scheme that future technologies will prefer over orthogonal multiple access (OMA). This research study focuses on a comparative analysis of several power allocation algorithms (PAA) to identify the PAA that is most suited for the "MIMO-NOMA" technology. The investigation is done using two different methods. First, an analysis is conducted to take into account direct approaches, followed by PAA and the Deep Neural Network (DNN) and the Depth Limited Search Algorithm (DLSA). Both techniques are tested considering two-user and multi-user scenarios in a MIMO-NOMA network. Optimum power allocation (OPA) to the weaker user terminal is a complex task in a real environment; moreover, the appropriate distribution of access point (AP) source power is constrained by the successive interference cancellation (SIC) technique. The above issues can be resolved with the help of the Deep Neural Network (DNN) and DLSA, where the stronger signal user gets minimal power and the poorer signal user gets optimal power. Based on the DLSA, the "DNN-MIMO-NOMA" system helps user terminals to receive their signals more accurately and without noise. For the successful implementation of SIC, the DLSA combines MIMO-NOMA with deep learning. The location of the user terminal is predicted by the DLS and depending on where the user terminal is located; optimal power allocation is done. The simulation results show that the PAA, combining DNN and the DLSA, outperforms traditional power allocation technique. [ABSTRACT FROM AUTHOR]

Healthcare is the foremost concern for any country. UN has also fixed 'good health' and well‐being' as its third sustainable development goal. CoVID‐19 era has shown the vulnerability of health infrastructure worldwide. As per WHO, aging population of the world would put enormous pressure on health infrastructure in coming decades. With lack of healthcare professionals and limited budget to achieve UNSDG goal‐3, IoMT with its huge umbrella of medical services and applications can address these issues. 5G has been deployed in many countries and vision for 6G have been finalized by different research groups. 6G will help Internet of Medical Things to realize its full potential. In this work, an open '6G enabled IoMT' architecture has been presented. It can integrate a wide number of services. The limitations of 5G in catering to such a network are also highlighted. The challenges and open issues are presented in the light of services and applications that can be provided by the purposed architecture. A brief overview of 6G is also provided. [ABSTRACT FROM AUTHOR]

The accurate quantification of concentration fluctuations is crucial when evaluating the exposure to toxic, infectious, reactive, flammable, or explosive substances, as well as for the estimation of odor nuisance. However, in the field of Computational Fluid Dynamics (CFD), the industry currently relies predominantly on steady-state RANS turbulence models for simulating near-field pollutant dispersion, which are only capable of producing the time-averaged concentration field. This paper presents a regression relationship for calculating the standard deviation of the local concentration based on the mean concentration and the downstream distance from a point source, over a city-like surface, in the case of the wind direction perpendicular to the streets. The desired peak values and other statistical characteristics can be predicted by assuming a gamma distribution which is fitted based on the average and standard deviation. To obtain the regression function, a deep neural network model was used. The model was trained using timeresolved concentration data obtained from wind tunnel experiments. The validation results show that the concentration fluctuations predicted by the DNN-based model are in satisfactory agreement with the measurement data in terms of the skewness, the kurtosis, the median, and the peak concentrations. Furthermore, the present paper suggests a workflow for estimating the concentration fluctuations based on RANS CFD results, as well as recommendations for generating further training data for specific applications. [ABSTRACT FROM AUTHOR]

Predicting indoor airflow distribution in multi-storey residential buildings is essential for designing energy-efficient natural ventilation systems. The indoor environment significantly impacts human health and well-being, considering the substantial time spent indoors and the potential health and safety risks faced daily. To ensure occupants' thermal comfort and indoor air quality, airflow simulations in the built environment must be efficient and precise. This study proposes a novel approach combining Computational Fluid Dynamics (CFD) simulations with machine learning techniques to predict indoor airflow. Specifically, we investigate the viability of employing a Deep Neural Network (DNN) model for accurately forecasting indoor airflow dispersion. The quantitative results reveal the DNN's ability to faithfully reproduce indoor airflow patterns and temperature distributions. Furthermore, DNN approaches to investigate indoor airflow in the residential building achieved an 80% reduction in the time required to anticipate testing scenarios compared with CFD simulation, underscoring the potential for efficient indoor airflow prediction. This research underscores the feasibility and effectiveness of a data-driven approach, enabling swift and accurate indoor airflow predictions in naturally ventilated residential buildings. Such predictive models hold significant promise for optimizing indoor air quality, thermal comfort, and energy efficiency, thereby contributing to sustainable building design and operation. [ABSTRACT FROM AUTHOR]

Accurate temperature forecasting is critical for various sectors, yet traditional methods struggle with complex atmospheric dynamics. Deep neural networks (DNNs), especially transformer-based DNNs, offer potential advantages, but face challenges with domain adaptation across different geographical regions. We evaluated the effectiveness of DNN-based domain adaptation for daily maximum temperature forecasting in experimental low-resource settings. We used an attention-based transformer deep learning architecture as the core forecasting framework and used kernel mean matching (KMM) for domain adaptation. Domain adaptation significantly improved forecasting accuracy in most experimental settings, thereby mitigating domain differences between source and target regions. Specifically, we observed that domain adaptation is more effective than exclusively training on a small amount of target-domain training data. This study reinforces the potential of using DNNs for temperature forecasting and underscores the benefits of domain adaptation using KMM. It also highlights the need for caution when using small amounts of target-domain data to avoid overfitting. Future research includes investigating strategies to minimize overfitting and to further probe the effect of various factors on model performance. [ABSTRACT FROM AUTHOR]

Background: Plant defense activators offer advantages over pesticides by avoiding the emergence of drug-resistant pathogens. However, only a limited number of compounds have been reported. Reactive oxygen species (ROS) act as not only antimicrobial agents but also signaling molecules that trigger immune responses. They also affect various cellular processes, highlighting the potential ROS modulators as plant defense activators. Establishing a high-throughput screening system for ROS modulators holds great promise for identifying lead chemical compounds with novel modes of action (MoAs). Results: We established a novel in silico screening system for plant defense activators using deep learning-based predictions of ROS accumulation combined with the chemical properties of the compounds as explanatory variables. Our screening strategy comprised four phases: (1) development of a ROS inference system based on a deep neural network that combines ROS production data in plant cells and multidimensional chemical features of chemical compounds; (2) in silico extensive-scale screening of seven million commercially available compounds using the ROS inference model; (3) secondary screening by visualization of the chemical space of compounds using the generative topographic mapping; and (4) confirmation and validation of the identified compounds as potential ROS modulators within plant cells. We further characterized the effects of selected chemical compounds on plant cells using molecular biology methods, including pathogenic signal-triggered enzymatic ROS induction and programmed cell death as immune responses. Our results indicate that deep learning-based screening systems can rapidly and effectively identify potential immune signal-inducible ROS modulators with distinct chemical characteristics compared with the actual ROS measurement system in plant cells. Conclusions: We developed a model system capable of inferring a diverse range of ROS activity control agents that activate immune responses through the assimilation of chemical features of candidate pesticide compounds. By employing this system in the prescreening phase of actual ROS measurement in plant cells, we anticipate enhanced efficiency and reduced pesticide discovery costs. The in-silico screening methods for identifying plant ROS modulators hold the potential to facilitate the development of diverse plant defense activators with novel MoAs. [ABSTRACT FROM AUTHOR]

Speech enhancement is an important method for improving speech quality and intelligibility in noisy environments. An effective speech enhancement model depends on precise modelling of the long-range dependencies of noisy speech. Several recent studies have examined ways to enhance speech by capturing the long-term contextual information. For speech enhancement, the time–frequency (T–F) distribution of speech spectral components is also important, but is usually ignored in these studies. The multi-stage learning method is an effective way to integrate various deep learning modules at the same time. The benefit of multi-stage training is that the optimization target can be iteratively updated stage by stage. In this paper, speech enhancement is investigated by multi-stage learning using a multi-stage structure in which time–frequency attention (TFA) blocks are followed by stacks of squeezed temporal convolutional networks (S-TCN) with exponentially increasing dilation rates. To reinject original information into later stages, a feature fusion block (FB) is inserted at the input of later stages to reduce the possibility of speech information being lost in the early stages. The S-TCN blocks are responsible for temporal sequence modelling tasks. The time–frequency attention (TFA) is a simple but effective network module that explicitly exploits position information to generate a 2D attention map to characterize the salient T–F distribution of speech by using two branches, time-frame attention and frequency attention in parallel. Extensive experiments have demonstrated that the proposed model consistently improves the performance over existing baselines across two widely used objective metrics such as PESQ and STOI. A significant improvement in system robustness to noise is also shown by our evaluation results using the TFA module. [ABSTRACT FROM AUTHOR]

Distributed denial of service (DDoS) attack is the most common attack that obstructs a network and makes it unavailable for a legitimate user. We proposed a deep neural network (DNN) model for the detection of DDoS attacks in the Software-Defined Networking (SDN) paradigm. SDN centralizes the control plane and separates it from the data plane. It simplifies a network and eliminates vendor specification of a device. Because of this open nature and centralized control, SDN can easily become a victim of DDoS attacks. We proposed a supervised Developed Deep Neural Network (DDNN) model that can classify the DDoS attack traffic and legitimate traffic. Our Developed Deep Neural Network (DDNN) model takes a large number of feature values as compared to previously proposed Machine Learning (ML) models. The proposed DNN model scans the data to find the correlated features and delivers high-quality results. The model enhances the security of SDN and has better accuracy as compared to previously proposed models. We choose the latest state-of-the-art dataset which consists of many novel attacks and overcomes all the shortcomings and limitations of the existing datasets. Our model results in a high accuracy rate of 99.76% with a low false-positive rate and 0.065% low loss rate. The accuracy increases to 99.80% as we increase the number of epochs to 100 rounds. Our proposed model classifies anomalous and normal traffic more accurately as compared to the previously proposed models. It can handle a huge amount of structured and unstructured data and can easily solve complex problems. [ABSTRACT FROM AUTHOR]

Deep neural networks (DNN) are highly effective in a number of tasks related to machine learning across different domains. It is quite challenging to apply the information gained to textual data because of its graph representation structure. This article applies innovative graph structures and protection techniques to secure wireless systems and mobile computing applications. We develop an Intrusion Detection System (IDS) with DNN and Support Vector Machine (SVM) to identify adversarial inversion attacks in the network system. It employs both normal and abnormal adversaries. It constantly generates signatures, creates attack signatures, and refreshes the IDS signature repository. In conclusion, the assessment indicators, including latency rates and throughput, are used to evaluate the effectiveness and efficiency of the recommended framework with Random Forest. The results of the proposed model (SVM with DNN) based on adversarial inversion attacks were better and more efficient than traditional models, with a detection rate of 93.67% and 95.34% concerning latency rate and throughput. This article also compares the proposed model (SVM with DNN) accuracy with other classifiers and the accuracy comparison for feature datasets of 90.3% and 90%, respectively. [ABSTRACT FROM AUTHOR]

Abstract Efficient and exact traffic flow forecasting is critical for intelligent transportation systems. The number of model generations increases the computational complexity of deep neural network (DNN) models. Overfitting occurs when hyperparameters are used excessively to train neural network models, and this has a major influence on prediction accuracy. To address these limitations, this approach employed H2O grid-based stochastic gradient descent with ridge regression in deep neural network (GSGDRR-DNN). This technique efficiently distributes memory across several clusters, runs independently, and simultaneously creates numerous DNN models. To remove multicollinearity and achieve better computational efficiency and lower variance, GSGDRR-DNN utilizes stochastic gradient descent (SGD) with ridge regression in the H2O cluster. Finally, we evaluate the performance of the recommended GSGDRR-DNN approach against several DNN methods, including LSTM, Bi-LSTM, GRU, and current state-of-the-art methods. Additionally, the run-time performance of the parallel GSGDRR-DNN model was compared with the run-time performance of the sequential GSGDRR-DNN model. The suggested system has a minimum MSE, a minimum RMSE, a minimum MAE, a minimum RMSLE, and maximum R2 values of 0.012, 0.108, 0.096, 0.015, and 0.99. This demonstrates that the GSGDRR-DNN model of traffic flows outperforms other state-of-the-art approaches in terms of prediction accuracy.

Stepwise inertial control (SIC) provides a step-increase of power after load fluctuation, which can effectively prevent system frequency decline and ensure the safety of grid frequency. However, in the power recovery stage, secondary frequency drop (SFD) is easy to occur. Therefore, it is necessary to optimize SIC to obtain a better frequency regulation effect. The traditional method has the disadvantages of high calculation dimension and long consuming time, which is difficult to meet the requirements of providing the optimal control effect in different scenarios. In order to realize the optimal stepwise inertial fast control of wind power frequency regulation in load disturbance events, this paper introduces the deep learning algorithm and proposes a stepwise inertial intelligent control of wind power for frequency regulation based on stacked denoising autoencoder(SDAE) and deep neural network(DNN). First, sparrow search algorithm (SSA) is used to obtain the optimal parameters, and SDAE is used to extract the data features efficiently. Then, DNN is used to learn the data features, and the accelerated adaptive moment estimation is introduced to optimize the network parameters to improve the global optimal parameters of the network. Finally, the stepwise inertial online control of wind power frequency regulation after disturbance event is realized according to SDAE-DNN. The simulation analysis is conducted for a single wind turbine and a wind farm in the IEEE 30-bus test system. Compared with the results obtained by the traditional method, shallow BP neural network and original DNN network, it is found that the proposed network structure has a better prediction accuracy and generalization ability, and the proposed method can achieve a great effect of stepwise inertia frequency regulation.

In the future, mixed traffic flow will consist of human-driven vehicles (HDVs) and connected autonomous vehicles (CAVs). Effective traffic management is a global challenge, especially in urban areas with many intersections. Much research has focused on solving this problem to increase intersection network performance. Reinforcement learning (RL) is a new approach to optimising traffic signal lights that overcomes the disadvantages of traditional methods. In this paper, we propose an integrated approach that combines the multi-agent advantage actor-critic (MA-A2C) and smart navigation (SN) to solve the congestion problem in a road network under mixed traffic conditions. The A2C algorithm combines the advantages of value-based and policy-based methods to stabilise the training by reducing the variance. It also overcomes the limitations of centralised and independent MARL. In addition, the SN technique reroutes traffic load to alternate paths to avoid congestion at intersections. To evaluate the robustness of our approach, we compare our model against independent-A2C (I-A2C) and max pressure (MP). These results show that our proposed approach performs more efficiently than others regarding average waiting time, speed and queue length. In addition, the simulation results also suggest that the model is effective as the CAV penetration rate is greater than 20%.

Dental age estimation is extensively employed in forensic medicine practice. However, the accuracy of conventional methods fails to satisfy the need for precision, particularly when estimating the age of adults. Herein, we propose an approach for age estimation utilizing orthopantomograms (OPGs). We propose a new dental dataset comprising OPGs of 27,957 individuals (16,383 females and 11,574 males), covering an age range from newborn to 93 years. The age annotations were meticulously verified using ID card details. Considering the distinct nature of dental data, we analyzed various neural network components to accurately estimate age, such as optimal network depth, convolution kernel size, multi-branch architecture, and early layer feature reuse. Building upon the exploration of distinctive characteristics, we further employed the widely recognized method to identify models for dental age prediction. Consequently, we discovered two sets of models: one exhibiting superior performance, and the other being lightweight. The proposed approaches, namely AGENet and AGE-SPOS, demonstrated remarkable superiority and effectiveness in our experimental results. The proposed models, AGENet and AGE-SPOS, showed exceptional effectiveness in our experiments. AGENet outperformed other CNN models significantly by achieving outstanding results. Compared to Inception-v4, with the mean absolute error (MAE) of 1.70 and 20.46 B FLOPs, our AGENet reduced the FLOPs by 2.7×. The lightweight model, AGE-SPOS, achieved an MAE of 1.80 years with only 0.95 B FLOPs, surpassing MobileNetV2 by 0.18 years while utilizing fewer computational operations. In summary, we employed an effective DNN searching method for forensic age estimation, and our methodology and findings hold significant implications for age estimation with oral imaging.

This comprehensive review explores the advancements in processing-in-memory (PIM) techniques and chiplet-based architectures for deep neural networks (DNNs). It addresses the challenges of monolithic chip architectures and highlights the benefits of chiplet-based designs in terms of scalability and flexibility. This review emphasizes dataflow-awareness, communication optimization, and thermal considerations in PIM-enabled manycore architectures. It discusses tailored dataflow requirements for different machine learning workloads and presents a heterogeneous PIM system for energy-efficient neural network training. Additionally, it explores thermally efficient dataflow-aware monolithic 3D (M3D) NoC architectures for accelerating CNN inferencing. Overall, this review provides valuable insights into the development and evaluation of chiplet and PIM architectures, emphasizing improved performance, energy efficiency, and inference accuracy in deep learning applications.