Search

With the development of emerging information technology, the traditional management methods of marine fishes are slowly replaced by new methods due to high cost, time-consumption and inaccurate management. The update of marine fishes management technology is also a great help for the creation of smart cities. However, some new methods have been studied that are too specific, which are not applicable for the other marine fishes, and the accuracy of identification is generally low. Therefore, this paper proposes an ecological Internet of Things (IoT) framework, in which a lightweight Deep Neural Networks model is implemented as a image recognition model for marine fishes, which is recorded as Fish-CNN. In this study, multi-training and evaluation of Fish-CNN is accomplished, and the accuracy of the final classification can be fixed to 89.89%–99.83%. Moreover, the final evaluation compared with Rem-CNN, Linear Regression and Multilayer Perceptron also verify the stability and advantage of our method. [ABSTRACT FROM AUTHOR]

In this paper, an example of an intelligent security system based on deep learning approach has been analyzed and described. Security is an important element in many aspects of human life. Depending on customer requirements, different levels of security may be required and implemented. Nowadays, artificial intelligence is becoming an indispensable part in many areas of human life and accordingly security systems are also based on the new achievements in this field. In this paper, an intelligent security system based on two methods for identifying persons, gait recognition and face recognition, was described. The two methods mentioned above can be used to give persons access to different buildings, facilities, institutions, etc. or to different indoor or outdoor areas. In light of the above, the aforementioned system uses face and gait recognition to identify persons who should be granted access to a facility or a specific area within the facility. Accordingly, two deep learning models for gait and face recognition have been developed and described. The analysis, the settings defined and the results obtained in relation to the models developed were also described. [ABSTRACT FROM AUTHOR]

In this paper, we analyzed the deep neural network (DNN) model and proposed a DNN-based method to build the training model for professional farmers. This paper aims to construct the vocational farmer cultivation model and seek a better path for transforming farmers into new experienced farmers by analyzing each training body involved in the cultivation of vocational farmers and studying their respective problems and reasons. The conceptual and logical structures of the system database were designed, and MySQL was selected for database implementation to complete the information of professional farmers. The system network topology and logical architecture are created, and the functions of view, control, business logic and data access layers are divided. This paper combines the DNN with the vocational farmer training enhancement decision tree. The experimental results of this model are most intuitive and accurate. This paper reconstructs the neural network model using the global average pooling layer to better model the vocational farmer training model to replace the fully connected layer in the original convolutional neural network. At the same time, to make the network model produce a lower probability of overfitting, a dropout layer is added to the layer after the fully connected layer to improve the efficiency of the neural network further to enhance vocational farmer training. The experimental content of this paper provides a new research direction for the estimation of vocational farmer cultivation. The above modeling is used to improve the vocational farmer cultivation model and accelerate the process of vocational farmer cultivation. [ABSTRACT FROM AUTHOR]

This paper introduces the use of an extra X current conveyor (EXCCII) for realizing logic operations, suited for mixed signal design, without necessitating separate digital blocks. The NOR and OR logic functions are thus shown to be realized using the current-mode analog building block operable at a low voltage. The use of ± 0. 7 5 V supply makes the new circuits operable at low voltage. The EXCCII bias current used in the design is 25 μ A. The input currents for defining logic 0 and logic 1 are, respectively, taken as 0 mA and 0.5 mA. The proposed circuits' delay is found to be 10 ns for low-to-high transition, while 2 ns for high-to-low transition for NOR gate under capacitive loading conditions. The use of the current-mode approach makes the new proposed circuit apt for current-mode signal processing. The new proposed logic function circuits are compatible with analog circuits, thus providing an easy merger for mixed signal systems design, paving the way to a modular design approach. As an extended application in artificial neural networks, the proposed circuit is shown to generate a sigmoid activation function with convincing results. Several results are included by varying the reference current, and the circuit's robustness aspect is also tested in the presence of fluctuation in the bias current of EXCCII. The novel approach of utilizing an analog building block for realizing digital functions is, therefore, verified with promising future applications. [ABSTRACT FROM AUTHOR]

Deep neural networks are vulnerable to adversarial examples. While there are many methods for generating adversarial examples using neural networks, creating such examples with high perceptual quality and improved training remains an area of active research. In this paper, we propose the Iterative Training Attack (ITA), a black-box attack based on a perturbation generative network for generating adversarial examples. ITA generates such examples by randomly initializing the perturbation generative network multiple times, iteratively training and optimizing a refined loss function. Compared to other neural network-based attacks, our proposed method generates adversarial examples with higher attack rates and within a small perturbation range even when the advanced defense is employed. Despite being a black-box attack, ITA outperforms gradient-based white-box attacks even under basic standards. The authors evaluated their method on a TRADES robust model trained with the MNIST dataset and achieved a robust accuracy of 92.46%, the highest among the evaluated methods. [ABSTRACT FROM AUTHOR]

Intelligent forecasting of economic indexes has been an important demand for sustainable management of smart cities. Existing methods for this purpose were mostly established upon the basis of economic mechanism. Econometric models are the most general technical means in this area. However, in era of digital economy, increasing amount of big data has brought great change to traditional production. It is becoming more difficult for conventional technological forecasting methods to deal with multi-dimensional economic indexes. To deal with such challenge, this paper introduces the artificial intelligence algorithms to implement automatic information processing, and proposes a deep neural network-based intelligent forecasting method for multi-dimensional economic indexes in smart cities. Specifically, a deep neural network with three-layer structure is developed as the backbone methodology. For empirical analysis, the real-world data from "Chengdu–Chongqing Economic Circle" in China from 2012 to 2022 are selected as the main simulation scenario. Four major indexes are selected as the main research object: gross product (GDP), per capita GDP, GDP growth rate and the proportion of tertiary industry in GDP. The experimental results show that the proposal can well deal with such forecasting problem from a data-driven perspective, with a proper forecasting effect on historical data. [ABSTRACT FROM AUTHOR]

Automatic fault diagnosis for power system equipment has always been an essential concern in this industry. Conventionally, such works are conducted by manual patrol inspection, which consumes much human labor and expert knowledge. Fortunately, infrared images can present diagnosis areas inside the equipment via the thermal sensing function. In such context, this work utilizes deep neural network to construct a specific infrared image processing framework that can realize automatic fault diagnosis. Thus in this paper, a deep learning-based fault diagnosis approach for power system equipment via infrared image sensing is proposed. First, a pulse-coupled neural network structure is employed to enhance feature representation for infrared images of the equipment. Next, a fuzzy C -means (FCM)-based segmentation method is developed to filter diagnosis areas from the infrared images. Finally, a convolution operation-based fault diagnosis operator is adopted to identify the diagnosis types. After that, some simulation experiments are conducted on real-world infrared images on the power system equipment, in order to make the performance evaluation of the proposed approach. The proposal realizes the end-to-end process of feature extraction and fault detection and identification, and avoids the problems of single feature. It is due to manual extraction of fault features, and the inability to detect and identify faults effectively in specific situations and scenarios. [ABSTRACT FROM AUTHOR]

Financial fraud detection has been an urgent technical demand in cyberspace. It highly relies on clear extraction and deep representation toward complex relationships inside financial social networks. As consequence, this study combines both knowledge graph and deep learning to deal with such issue. Thus, an intelligent financial fraud detection model based on knowledge graph guidance and deep neural network is proposed in this paper. First, a new knowledge graph based on financially related systems is constructed, which includes multiple entities and relationships. Then, an adversarial learning-based neural network structure is formulated to extract financial attributes. Finally, the detection results can be output according to the extracted factors. Empirically, the proposal is implemented on a real-world dataset for performance evaluation. The experimental results show that it has higher accuracy and effectiveness compared to traditional fraud detection methods. The proposed detection model can not only identify known fraudulent behaviors, but also predict potential fraud patterns based on consumer habits, thereby improving the security and reliability of financial transactions. It can also update the knowledge graph in real-time, enabling it to cope with emerging fraud methods and variants. [ABSTRACT FROM AUTHOR]

Financial risk warning is a crucial technical issue for enterprises. Traditionally, it relied on modeling analysis of a single data type, which does not fully capture the diverse characteristics of financial risk activities. To address this, this paper introduces a multimodal deep learning framework driven by semantic analysis and image processing for intelligent warning of enterprise financial risks. Initially, natural language processing algorithms analyze textual data such as financial statements, news reports, and social media comments within the industry. Concurrently, two convolutional neural network models, M-CNN and M-RNN, extract features from images and chart data. These textual and visual feature representations are then fused to create a multimodal deep neural network framework. Extensive experimental evaluation and comparative analysis of the proposed framework were conducted. The results indicate that the financial risk rate of consumer fraud graph analysis varies significantly, displaying a fluctuating state with values ranging from 2.1% to 16.8%. Compared to other methods, the proposed approach demonstrates superior performance in financial risk warning tasks, with the risk rate increasing from 1.2% to 26.5% as it iterates from 1 to 6. [ABSTRACT FROM AUTHOR]

Monitoring and managing litter on the roadways is important not only for preserving the cleanliness and esthetic appeal of our cities but also for safeguarding the overall health and sanitary environment of citizens. With the development of artificial intelligence, it is now possible to design algorithms capable of autonomously assessing the sanitation status of roadways. In this paper, we propose a method for detecting garbage on roads as our solution to tackle the issue. The proposed model is a deep learning model with an attention mechanism introduced to detect garbage in real-time scenarios. Furthermore, the object recognition capability of the model is enhanced through transfer learning to mitigate the influence of irrelevant content during training. The refined network can detect rubbish on roads with higher efficiency than the existing models. Also, compared with other methods, it consumes less amount of data for training. The experimental results demonstrate the efficiency of our proposed model, with a performance boost of 7.62% over the state-of-the-art methods. [ABSTRACT FROM AUTHOR]

In this paper, we propose a novel transformer-based deep neural network model to learn semantic bug patterns from a corpus of buggy/fixed codes, then generate correct ones automatically. Transformer is a deep learning model relying entirely on attention mechanism to model global dependencies between input and output. Although there are a few endeavors to repair programs by learning neural language models (NLM), many special program properties, such as structure and semantics of an identifier, are not considered in embedding input sequence and designing model effectively, which results in undesired performance. In the proposed Bug-Transformer, we design a novel context abstraction mechanism to better support neural language models. Specifically, it is capable of 1) compressing code information but preserving the key structure and semantics, which provides more thorough information for NLM models, 2) renaming identifiers and literals based on their lexical scopes, structural and semantic information, to reduce code vocabulary size and 3) reserving keywords and selected idioms (domain- or developer-specific vocabularies) for better understanding code structure and semantics. Hence, Bug-Transformer adequately embeds code structural and semantic information into input data and optimize attention-based transformer neural network to well handle code features in order to improve learning tasks for bug repair. We evaluate the performance of the proposed work comprehensively on three datasets (Java code corpora) and generate patches to buggy code using a beam search decoder. The experimental results show that our proposed work outperforms the-state-of-art techniques: Bug-Transformer can successfully predict 54.81%, 34.45%, and 42.40% of the fixed code in these three datasets, respectively, which outperform the baseline models. These success rates steadily increase along with the increase of beam size. Besides, the overall syntactic correctness of all patches remains above 97%, 96%, and 50% on the three benchmarks, respectively, regardless of the beam size. [ABSTRACT FROM AUTHOR]

The neuristor based on memristors can be used to mimic synapse and neurons of biological neural systems, and it is the key unit of spiking neural networks. However, the resistance states of realistic memristors are nonvolatile, which is not conducive to mimicking the forgetting function of the brain. Given that the resistance states of memristor emulators are volatile after power down, this paper exhibits a scalable neuristor built with a half-wave memristor emulator. The proposed neuristor demonstrates four critical features for action-potential-based computing: the all-or-nothing spiking of an action potential, threshold-driven spiking, diverse periodic spiking and symmetric anti-Hebbian learning rule of spike-timing-dependent plasticity. Particularly, there are no complex shape and duration constraints on pre- and post-spikes for implementing the symmetric anti-Hebbian learning rule. [ABSTRACT FROM AUTHOR]

Neuromorphic systems are the future computing systems to overcome the von Neumann's power consumption and latency wall between memory and processing units. The two main components of any neuromorphic computing system are neurons and synapses. Synapses carry the weight of the system to be multiplied by the neuromorphic attributes, which represent the features of the task to be solved. Memristor (memoryresistor) is the most suitable circuit element to act as a synapse. Its ability to store, update and do matrix multiplication in nanoscale die area makes it very useful in neuromorphic synapses. One of the most popular memristor synapse configurations is the two-transistor–one-memristor (2T1M) synapse. This configuration is very useful in neuromorphic synapses for its ability to control reading and updating the weight on a chip by signals. The main problem with this synapse is that the reading operation is destructive, which results in changing the stored weight value. In this paper, a novel refreshment circuit is proposed to restore the correct weight in case of any destructive reading operations. The circuit makes a small interrupt time during operation without disconnecting the memristor, which makes the circuit very practical. The circuit has been simulated by using hardware-calibrated CMOS TSMC 130 nm technology on Cadence Virtuoso and linear ion drift memristor Verilog-A model. The proposed circuit achieves the refreshment task accurately for several error types. It is used to refresh 2T1M synapse with any destructive reading signal shape. [ABSTRACT FROM AUTHOR]

The price of carbon emission rights in the market fluctuates greatly due to various factors from economy, finance, and climate. For an enterprise that needs to conduct carbon trading activities, it is extremely necessary to fully grasp the price of carbon trading in the future period. Thus in this paper, a deep neural network-based assistive decision model for financial risk prediction in carbon trading market is proposed for this purpose. Specifically, the dynamic risk spillover effects of domestic and international carbon trading markets are studied, and a frontier time-varying model is utilized to measure the risk spillover effects. Then, the deep neural network is used for quantitative research and to construct an intelligent decision scheme that outputs financial risk prediction results. Four perspectives, energy price, climate environment, carbon market price, and macroeconomics, are selected as the input to analyze the influence factors of carbon emission rights price. Finally, several linear regression models are adopted as the baseline methods for comparison, and experimental results show that the proposed method can achieve better prediction performance compared with baseline methods. [ABSTRACT FROM AUTHOR]

Several hardware Trojan (HT) detection techniques are available today to ensure the security of hardware systems. However, the existing pre-silicon HT detection techniques have problems such as difficulties in capturing HT path features and poor applicability. To address these challenges, this paper proposes a gate-level HT detection scheme based on a deep learning model. We parse the circuit gate-level netlist and develop an algorithm to extract circuit path sentences based on the signal propagation rule. Path sentences consisting of gate names are extracted as experimental datasets. We apply the theory of natural language processing (NLP) to the task of HT detection and use three neural networks to filter the length of path sentences. Then, based on the deep learning model text convolutional neural network (TextCNN), we propose PS-TextCNN for HT detection. Our approach is verified on seven benchmark circuits of the RS232-series and eight benchmark circuits of the s-series. We achieve an average true positive rate (TPR) of 88.9%. The TPR of the RS232-series reaches a high score of 99.5%. The TPR of the s-series is 79.5%, which is significantly higher than that of the existing gate-level HT detection techniques. [ABSTRACT FROM AUTHOR]

The credit risk assessment acts as an important part in daily affairs for financial institutions. But in the era of big data, the growing business volume makes it an urgent demand to develop digital ways of credit risk assessment. Currently, the machine learning is universally employed to establish various data-driven models for this purpose. However, machine learning models generally suffer from limited ability of feature representation and robustness, and cannot deal with more complex financial security scenarios. To deal with this issue, this work introduces ensemble learning to construct a stronger credit risk prediction model via integration of several basic machine learning models. Thus, an ensemble learning-enhanced smart prediction model for financial credit risk is proposed in this paper. Three classification-based machine learning models (support vector machine, artificial neural network and radial basis function) are selected as the basic classifiers, and "voting" strategy is utilized to integrate them into a novel strong classifier. A real-world financial credit dataset released by a Chinese commercial bank was selected as the experimental scenario. The obtained results show that the proposal has better prediction accuracy compared with basic machine learning models without ensemble learning. [ABSTRACT FROM AUTHOR]

Deep neural network has made surprising achievements in natural language processing, image pattern classification recognition, and other domains in the last few years. It is still tough to apply to hardware-constrained or mobile equipment because of the huge number of parameters, high storage as well as computing costs. In this paper, a new sparse iteration neural network architecture is proposed. First, the pruning method is used to compress the model size and make the network sparse. Then the architecture is iterated on the sparse network model, and the network performance is improved without adding additional parameters. Finally, the hybrid deep learning model was carried out on CV tasks and NLP tasks on ANN, CNN, and Transformer. Compared with the sparse network architecture, we finally found that the accuracy of the MINST, CIFAR10, PASCAL VOC 2012, and SQuAD datasets is improved by 0.47%, 0.64%, 3.75%, and 15.06%, respectively. [ABSTRACT FROM AUTHOR]

Recent advances in speaker embeddings promote a great development of speaker diarization. However, determining 'who spoke when' in the meeting scenarios is still challenging due to similar speaker voices and unknown speaker quantity. In this paper, this research proposes enhanced discriminative features for speaker diarization, including discriminative speaker-specific features based on Siamese networks, and a speaker re-verification method. With Siamese architecture, SiamTDNN, this research first explores latent representations which is capable of modeling intra-class and inter-class differences between speakers, by training with audio pairs. Then, the re-verification method is introduced with a local-global strategy to identify speakers in a multi-person talking scene. Our method provides a novel speaker embedding with enhanced discriminative power for disambiguated speakers and achieves an elevated upper bound on the number of speakers. The proposed speaker embedding achieved an EER of 1.1% and a minDCF of 0.1192 on VoxCeleb1 for the speaker verification task. Extensive experiments on AiShell-4, ICSI, AMI and VoxConverse demonstrate the effectiveness of the proposed method with an average DER reduction of 3% and an RTF of 0.0792. [ABSTRACT FROM AUTHOR]

This paper aims to propose an improved image classification model to reduce the cost of model construction. Aiming at the problem that network training usually requires the support of a large number of labeled samples, an image classification model based on semi-supervised deep learning is proposed, which uses labeled samples to guide the network to learn unlabeled samples. A convolutional neural network model for simultaneous processing of labeled and unlabeled data is constructed. The tagged data is used to train the Softmax classifier and provide the initial K-means clustering center for the untagged data. The nonsubsampling contourlet layer is used to replace the first convolutional layer of the full convolutional neural network to extract multi-scale depth features, and the nonsubsampling contourlet full convolutional neural network is constructed. The network can extract multi-scale information of the images to be classified, and extract more discriminative deep image features. In addition, the parameters of the nonsubsampled contourlet layers are pre-set and do not require network training. The proposed method has higher classification accuracy than the contrast method on polarimetric SAR images using the nonsubsampled contourlet full convolutional neural network. [ABSTRACT FROM AUTHOR]

Considering the problem of residual noise and spurious modes in the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), a rolling element bearing malfunction diagnostic method based on improved CEEMDAN (ICEEMDAN) is proposed. First, different from the CEEMDAN, which directly adds Gaussian white noise with a mean of zero, the proposed method adds the k th component obtained from white noise decomposed by empirical mode decomposition (EMD) to the vibration signal, and then the ICEEMDAN is employed to decompose the signal into several intrinsic mode functions (IMFs). Second, aiming at the uncertainty problem of entropy estimation in multi-scale fuzzy entropy (MFE), a refined composite multi-scale fuzzy entropy (RCMFE) is proposed to obtain the characteristic from the selected IMFs. Finally, smoothing factor of PNN is determined by fruit fly optimization algorithm (FOA), and the extracted features are input into the FOA-PNN model to achieve condition identification. Experimental results illustrate that the identification accuracy is more than 99%, which indicates its high effectiveness and superiority. [ABSTRACT FROM AUTHOR]

In the resource-constrained embedded systems, the designing of efficient deep neural networks is a challenging process, due to diversity in the artificial intelligence applications. The quantization in deep neural networks superiorly diminishes the storage and computational time by reducing the bit-width of networks encoding. In order to highlight the problem of accuracy loss, the quantization levels are automatically discovered using Policy Feedback-based Reinforcement Learning Method (PF-RELEQ). In this paper, the Proximal Policy Optimization with Policy Feedback (PPO-PF) technique is proposed to determine the best design decisions by choosing the optimum hyper-parameters. In order to enhance the sensitivity of the value function to the change of policy and to improve the accuracy of value estimation at the early learning stage, a policy update method is devised based on the clipped discount factor. In addition, specifically the loss functions of policy satisfy the unbiased estimation of the trust region. The proposed PF-RELEQ effectively balances quality and speed compared to other deep learning methods like ResNet-1202, ResNet-32, ResNet-110, GoogLeNet and AlexNet. The experimental analysis showed that PF-RELEQ achieved 20% computational work-load reduction compared to the existing deep learning methods on ImageNet, CIFAR-10, CIFAR-100 and tomato leaf disease datasets and achieved approximately 2% of improvisation in the validation accuracy. Additionally, the PF-RELEQ needs only 0.55 Graphics Processing Unit on an NVIDIA GTX-1080Ti to develop DNNs that delivers better accuracy improvement with fewer cycle counts for image classification. [ABSTRACT FROM AUTHOR]

As a third generation artificial neural network, spiking neuron network is expected to expand the artificial intelligence world. However, as a more detailed simulation of brain, a single run of spiking neural networks (SNNs) simulation can take hours to days. To get a better prediction of SNN simulation performance, existing work requires gathering result of actual runs to conduct accurate modeling. In this paper, we propose a nonempirical SNN simulation performance prediction method, prototyped in a hybrid CPU-FPGA cluster. Experiments show that our method, without actual simulation run, can get comparable accuracy with orders of magnitude less runtime cost. [ABSTRACT FROM AUTHOR]

In this paper, we propose a novel multi-scale correlation network (MSCNet) for homography estimation from coarse to fine. First, we extract multi-scale features through a siamese network to generate global and local correlations from feature maps of different scales. Second, we use a group dilated deconvolution block to capture global mapping by increasing the receptive fields in terms of different levels. Third, we employ the channel and spatial attention mechanism to achieve local refinement for small displacements. Finally, we adopt a knowledge distillation strategy to lightweight our model while maintaining relatively high estimation performance. Experimental results on Microsoft Common Objects in Context (MSCOCO) dataset show that our proposed MSCNet outperforms the state-of-the-art approaches in terms of accuracy and parameter count. [ABSTRACT FROM AUTHOR]

A Battery Management System (BMS) can prolong the life of the battery but it depends on the accuracy of the adopted scheme. Different techniques have been developed to enhance the BMS by monitoring the State of Health (SOH) of the battery. In this paper, the detection of battery voltage is analyzed by using the cycle counting method, which is a conventional technique and compared with Artificial Neural Network (ANN), a heuristic method. The advantage of the proposed ANN method is that SOH can be monitored without disconnecting the battery from the load. Also, the sampling data to the ANN are derived from various techniques including Open Circuit Voltage (OCV) method, Ambient temperature measurement, and valley point detection. A feed-forward backpropagation algorithm is used to achieve the purpose of real-time monitoring of the LAB. The results show that the precise estimation of SOH can be obtained by Feed-Forward Neural Network (FFNN) when trained with more sampling data. [ABSTRACT FROM AUTHOR]

In this paper, nature inspired search algorithms, namely particle swarm optimization (PSO) and genetic algorithm (GA) are used to design fractional order proportional and integral (FOPI) and artificial neural network (ANN) controller based distribution static compensator (DSTATCOM) and electronic load controller (ELC) for power quality improvement. Improvement in power quality is achieved using DSTATCOM and an ELC. DSTATCOM is designed using FOPI and ANN based controllers, as opposed to conventional PI controllers which are comparatively less efficient. PSO and GA techniques are employed to determine the optimal parameters for the controllers. The improvement in the performance of the ANN and FOPI as compared to PI controller for the DSTATCOM and ELC is validated using MATLAB based modeling and simulations. Linear consumer loads were used to perform a comparative study in terms of maximum percentage error. Further, we analyzed the system for a nonlinear load and demonstrated decrease in the harmonic distortion. [ABSTRACT FROM AUTHOR]

Person Re-ID is challenged by background clutter, body misalignment and part missing. In this paper, we propose a reliable part-based multiple levels attention deep network to learn multiple scales salience representation. In particular, person alignment and key point detection are sequentially carried out to locate three relative stable body components, then fused attention (FA) mode is designed to capture the fine-grained salient features from effective spatial of valuable channels of each part, regional attention mode is succeeded to weight the importance of different parts for highlighting the representative parts while suppressing the valueless ones. A late fusion-based multiple-task loss is finally adopted to further optimize the valuable feature representation. Experimental results demonstrate that the proposed method achieves state-of-the-art performances on three challenging benchmarks: Market-1501, DukeMTMC-reID and CUHK03. [ABSTRACT FROM AUTHOR]

Accurate river level prediction is vital in water resource management and flood mitigation. Recent advances in data-driven modeling, especially in deep learning, provide profound insights into predicting river levels when mechanistic hydrological knowledge is absent. However, they often do not capture predictive uncertainty well, making them less robust in the hydrological prediction as they overconfidently extrapolate. Moreover, they are not flexible in handling hydrological variables with heterogeneous characteristics. In this work, we present the Temporal Fusion Transformer-Gaussian Process (TFT-GP), a novel model for multi-horizon probabilistic river level prediction. We show how TFT-GP inherits the nice properties of the Gaussian process and deep neural networks, giving it excellent representative power and uncertainty quantification ability. The performance of TFT-GP is thoroughly compared with existing well-known deep learning models in three real-world hydrological datasets, and the results showed that TFT-GP is not only more accurate in point prediction but also more reasonable in uncertainty quantification. [ABSTRACT FROM AUTHOR]

In deep learning, convolutional neural networks (CNNs) are a class of artificial neural networks (ANNs), most commonly applied to analyze visual imagery. They are also known as Shift-Invariant or Space-Invariant Artificial Neural Networks (SIANNs), based on the shared-weight architecture of the convolution kernels or filters that slide along input features and provide translation-equivariant responses known as feature maps. Recently, various architectures for CNN based on FPGA platform have been proposed because it has the advantages of high performance and fast development cycle. However, some key issues including how to optimize the performance of CNN layers with different structures, high-performance heterogeneous accelerator design, and how to reduce the neural network framework integration overhead need to be improved. To overcome and improve these problems, we propose dynamic cycle pipeline tiling, data layout optimization, and a pipelined software and hardware (SW–HW)-integrated architecture with flexibility and integration. Some benchmarks have been tested and implemented on the FPGA board for the proposed architecture. The proposed dynamic tiling and data layout transformation improved by 2.3 times in the performance. Moreover, with two-level pipelining, we achieve up to five times speedup and the proposed system is 3.8 times more energy-efficient than the GPU. [ABSTRACT FROM AUTHOR]

In recent years, photovoltaic power generation technology has become the key planning direction of the country. It is important to effectively predict photovoltaic (PV) electricity generation capacity, so that the administrators can well schedule resource allocation. Currently, most of the photovoltaic electricity generation forecasting models took meteorological data as the input parameters of neural network. However, the input parameters and redundant data cause neural network to converge difficultly. Besides, single types of neural network models cannot well capture the comprehensive characteristics, which may influence forecasting effect in evolving process. As a result, we propose a hybrid neural network-based intelligent forecasting approach for PV electricity generation capacity. First, convolution neural network (CNN) is adopted to extract the connection between features and data from the perspective of convolution operations. And then, the extracted feature vector of time series is sent into the long short-term memory (LSTM) model. Finally, the forecasting values are predicated by training the outlined LSTM network. The experimental results indicate that such a hybrid CNN-LSTM model can significantly improve the precision of PV electricity generation prediction and provide an effective way to forecast generation power of PV system. [ABSTRACT FROM AUTHOR]

With the rapid development of intelligent techniques, smart finance has become a hot topic in daily life. Currently, financial credit is facing increasing business volume, and it is expected that investigating the intelligent algorithms can help reduce human labors. In this area, the prediction of latent credit default behaviors can help deal with loan approval affairs, and it is the most important research topic. Machine learning-based methods have received much attention in this area, and they can achieve proper performance in some scenarios. However, machine learning-based models cannot have resilient objective function, which can cause failure in having stable performance in different problem scenarios. This work introduces deep learning that has the objective function with high freedom degree, and proposes an artificial neural network-based intelligent prediction model for financial credit default behaviors. The whole technical framework is composed of two stages: information encoding and backbone network. The former makes encoding toward initial features, and the latter builds a multi-layer perceptron to output prediction results. Finally, the experiments are conducted on a real-world dataset to evaluate the efficiency of the proposed approach. [ABSTRACT FROM AUTHOR]

Recently, deep neural networks have achieved remarkable progress in class balancing instance segmentation. However, most applications in the real world have a long-tailed distribution, i.e., limited training examples in the majority of classes. The long-tailed challenge leads to a catastrophic drop in instance segmentation because the gradient of the head classes suppresses the gradient of the tail classes, leading to a bias towards the major classes. We propose LiCAM, a novel framework for long-tailed segmentation. It features an adaptive loss function named Moac Loss, which is adjustable during the training according to the monitored classification accuracy. LiCAM also cooperates with an oversampling technique named RFS, which alleviates the severe imbalance between head and tail classes. We conducted extensive experiments on the LVIS v1 dataset to evaluate LiCAM. With a coherent end-to-end training pipeline, LiCAM significantly outperforms other baselines. [ABSTRACT FROM AUTHOR]

The Internet of Things (IoT) is bringing a new dimension to the smart farming market. This helps the user to collect the data from the agricultural fields in real time and move it to remote areas for processing. With the available sensor data and the image taken from the fields, automated disease prediction is possible. Deep neural network is used for classification of disease using the leaf images. Agriculture is the backbone of our country, but our output is poor when compared to the global standards due to lack of using technologies in the fields. In this work, various sensors like humidity sensor, pH level monitoring sensor, Temperature sensor, and Soil moisture sensor are used in the agricultural fields for collecting the real-time data. Multiple Sensors are installed in various locations of farms with one common controller Raspberry PI 3 module (RPI3), which was used to control all these sensors. Camera interfacing with RPI can be observed on leaf disease. Convolutional neural network architecture is used for leaf disease detection and classification. The accuracy of the disease classification system using convolutional neural network is 96% when the system is iterated for 50 epochs. [ABSTRACT FROM AUTHOR]

With the rapid development of sensor technology, multi-modal data fusion methods based on deep neural networks provide a reliable guarantee for object recognition and detection in complex scenarios. Most of the existing RGB-D image salient object detection methods improve the salient object detection methods in 2D scenes, which have many problems, such as ineffective fusion of feature information, incomplete image feature extraction and so on. Aiming at these problems, we proposed a salient object detection method based on bifurcated fusion network. First, we use the high- level features of the global context to locate the salient object, then we use the low-level features of the local details to extract the edge information. Second, we model the RGB information and depth information through the gate channel transformation control mechanism, and construct the bifurcation backbone network model and generate the initial salient map. Finally, based on the initial salient map and low-level features, the final salient map is generated. Experimental results show that the proposed method can fully utilize the multi-layer feature information of the image and effectively detect the salient object. [ABSTRACT FROM AUTHOR]

Traffic flow forecasting is the basic challenge in intelligent transportation system (ITS). The key problem is to improve the accuracy of model and capture the dynamic temporal and nonlinear spatial dependence. Using real data is one of the ways to improve the spatial–temporal correlation modeling accuracy. However, real traffic flow data are not strictly periodic because of some random factors, which may lead to some deviations. This study focuses on capturing and modeling the temporal perturbation in real periodic data and we propose a spatial–temporal similar graph attention network (STSGAN) to address this problem. In STSGAN, the spatial–temporal graph convolution module is to capture local spatial–temporal relationship in traffic data, and the periodic similar attention module is to treat the nonlinear traffic flow information. Experiments on three datasets demonstrate that our model is best among all methods. [ABSTRACT FROM AUTHOR]