Search

During the COVID-19 pandemic, online classes became the only option for many students. The main challenge for these classes was conducting risky and complex chemical or biological experiments in a domestic environment. To address this challenge, a smart experiment system called MRLab was developed. MRLab used wearables such as a smart glove and head-mounted device to record sensory data and a multimodal hybrid fusion model GVVS to interpret the user's experimental intent, which essentially transforms the user's abstract behavioral actions into a probabilistic set of experimental intent that can be computed. Different experiments in MRLab used different libraries of experimental intents. The SrNet model in GVVS was used to estimate the probability of the user's gesture behavior generated from the smart glove, while the SIPA algorithm compared speech information entered during the experiment with the experimental intent library to estimate the probability of the user's intent. At the same time, the scene visual channel monitored the information about the object the user intended to operate, with the SVF algorithm computing the probability of the intended object in real-time. The results from ANOVA and post-hoc comparative testing conducted on 21 volunteers revealed that MRLab outperformed other experiment modes, including WEB, AR, and VR, with a higher intention understanding rate, efficiency, and user satisfaction. Therefore, MRLab proved to be a useful alternative to traditional physics laboratory experiments during the pandemic, along with being an additional teaching tool for remote learning purposes. [ABSTRACT FROM AUTHOR]

Abstract The current teaching mode in geometry mainly focuses on two-dimensional levels, and the teaching tools utilized are static and not interactive. This paper proposes the use of a smart pen for three-dimensional geometry experimental teaching and a multimodal intention understanding and human-computer collaboration algorithm for the smart pen. The primary innovations of this paper lie in the development of a smart pen and a virtual platform for geometry education tailored for geometry experimental instruction. This system can promptly perceive and comprehend user behavior in real-time. Furthermore, a standardized topological equivalence model is proposed as the basis for a point selection strategy. By establishing correspondence between the modeled point selection model and the actual operation scene, the behavioral intent imposed on the model is applied to the operation object of the actual scene. Additionally, CNN-based and information entropy-based multimodal fusion intention understanding models are proposed for different input modalities to capture the operational intention of users by fusing their multimodal input data. The algorithm further improves the accuracy rate through an error correction mechanism based on implicit interaction to achieve better human-computer collaboration. The algorithm proposed in this paper has resulted in a 0.47-second improvement in point selection average time and achieved an intent understanding accuracy of 98.21%. This improvement leads to better fault tolerance and fluency during human-computer interaction, reduces the cognitive load on the user, and improves the overall user experience. [ABSTRACT FROM AUTHOR]

Abstract The naturalness and safety of human-computer interaction have always been primary research focuses in the field of human-computer interaction. This paper proposes a multimodal intention understanding algorithm (MIUIC), which incorporates comfort analysis, as a solution to address the issues of low intention understanding rate, weak interaction, and weak collaboration that are often observed in most massage systems. The algorithm efficiently fuses multimodal data based on objective implicit information to address the challenge of low intention understanding rates caused by non-standard expression of natural behavior. Moreover, this algorithm incorporates comfort reasoning to detect and address intentions related to security threats while providing the ability for robots to make behavioral decisions through inverse active interaction, leading to more equitable human-robot interactions. To test the validity and safety of the MIUIC algorithm, we embedded the algorithm into a mechanical arm massage system. Subsequently, 45 elderly volunteers were invited to participate in experimental tests. Finally, to verify the validity and safety of the MIUIC algorithm, we assessed the algorithm in terms of four aspects, including multimodal intention recognition rate, the ability to reduce data dispersion, the intention enhancement rate under reverse human-machine interaction, and the rate of avoiding dangerous intentions. In conclusion, the MIUIC algorithm enhances the intention understanding rate and promotes. [ABSTRACT FROM AUTHOR]

Virtual reality (VR) experimental platforms are frequently utilized in the teaching of chemical experiments because of its high level of interest, lifelike effect, and safe and dependable advantages. However, users are unable to operate genuine experimental equipment on the present VR experimental platform, and these systems are unable to precisely assess and manage users' operation intentions. In light of the aforementioned concerns, this paper develops a smart glove with cognitive capabilities, which is composed of a variety of simple commercial sensors and binocular cameras. The smart glove can be applied to the virtual‐reality fusion chemical experiment platform established in this paper. Additionally, a navigation interaction algorithm based on multimodal intention understanding is also proposed in this study (hereinafter referred to as NIAMIU algorithm). The smart glove can accurately and efficiently obtain the user's operation intention through the algorithm. The experiment showcased that the smart glove detailed in this paper is capable of precisely determining the position of users' hands in space and guiding and rectifying the user's interactive behavior. Meanwhile, users may converse with the smart glove in an unsupervised setting and conclude the experiment by following the directions supplied by the smart glove. In comparison to a conventional data glove, the smart glove designed in this paper considerably augments the accuracy and efficiency of human‐computer interaction. [ABSTRACT FROM AUTHOR]

Despite the emergence of various human-robot collaboration frameworks, most are not sufficiently flexible to adapt to users with different habits. In this article, a Multimodal Reinforcement Learning Human-Robot Collaboration (MRLC) framework is proposed. It integrates reinforcement learning into human-robot collaboration and continuously adapts to the user's habits in the process of collaboration with the user to achieve the effect of human-robot cointegration. With the user's multimodal features as states, the MRLC framework collects the user's speech through natural language processing and employs it to determine the reward of the actions made by the robot. Our experiments demonstrate that the MRLC framework can adapt to the user's habits after repeated learning and better understand the user's intention compared to traditional solutions. [ABSTRACT FROM AUTHOR]

Lacking human-computer collaborative capability is one of the key problems commonly faced by the existing escort robots. To this end, a helping elderly escort interactive system based on reverse active integration of multimodal intentions (MES) is proposed in this paper in response to the elderly's decline in language expression ability, memory, and other abilities. This system can understand the intention of the elderly based on scene perception and three-modal information: speech, gesture, and posture. In detail, the system can extract the interactive intention from the nondeterministic multimodal data input by the elderly and evaluate the trust degree of the extracted intention. The evaluation of intention trust degree is such a process that the system autonomously judges the feasibility of the elderly's intention and corrects the wrong intention expressed by the elderly due to memory decline by reverse thinking of "find reasons based on the results"; when the intention cannot be extracted, the system will take the initiative to ask the elderly for enhanced information conducive to the intention extraction, so as to quickly and correctly extract the interactive intention of the elderly. This design aims at improving the quality of elderly care, making the interaction between the elderly and the robot more natural, improving the accuracy in intention extraction from fuzzy expression, as well as breaking the traditional "master-slave" human-computer interaction and improving the harmony of human-computer interaction. Further, the implementation principle of the system is detailed, and the system is evaluated in this paper. The evaluation experiment was conducted by a robot Pepper embedded with the system. Through experiments, it is verified that Pepper can quickly and accurately get the real intention of the elderly in the interaction with the elderly. In a challenging environment (such as the unclear expression of the elderly), it can still correctly extract the real intention of the elderly with an accuracy of 97% and can effectively avoid the wrong intention expressed by the elderly. This puts forward a valuable research path for the challenge in human-computer collaborative interaction. [ABSTRACT FROM AUTHOR]

While facing complex surroundings, robots need to identify the same intention which is expressed in different ways. In order to solve the problem of assisting robots to get a better intention understanding, a self-tuning multimodal fusion algorithm is put forward in this paper, which is not restricted by the expressions of interacting participants and environment. The multimodal fusion algorithm can be transferred to different application platforms. Robots can own the understanding competence and adapt new tasks by changing the content of the robot knowledge base. Compared with other multimodal fusion algorithms, this paper attempts to transfer the basic structure of feed-forward neural networks on discrete sets, which has strengthened the consistency and perfect the complementary relations between multiple mode, and has realized the simultaneous operation of fusion operator's self-tuning and intention search. There are three kinds of modes selected in the paper: speech, gesture, and scene objects, where the single modal classifiers are trained separately. This method conducted a human-computer interaction experiment on the bionic robot Pepper platform, which proved that the method can effectively improve the accuracy and robustness of robots in aspects of understanding human intentions, and reduce the uncertainty about intention judgment in a single modal interaction. [ABSTRACT FROM AUTHOR]

A growing number of studies have been conducted over the past few years on the positioning of daily massage robots. However, most methods used for research have low interactivity, and a systematic method should be designed for accurate and intelligent positioning, thus compromising usability and user experience. In this study, a massage positioning algorithm with online learning capabilities is presented. The algorithm has the following main innovations: (1) autonomous massage localization can be achieved by gaining insights into natural human-machine interaction behavior and (2) online learning of user massage habits can be achieved by integrating recursive Bayesian ideas. As revealed by the experimental results, combining natural human-computer interaction and online learning with massage positioning is capable of helping people get rid of positioning aids, reducing their psychological and cognitive load, and achieving a more desirable positioning effect. Furthermore, the results of the analysis of user evaluations further verify the effectiveness of the algorithm. [ABSTRACT FROM AUTHOR]

At present, virtual-reality fusion smart experiments mostly employ visual perception devices to collect user behavior data, but this method faces the obstacles of distance, angle, occlusion, light, and a variety of other factors of indoor interactive input devices. Moreover, the essence of the traditional multimodal fusion algorithm (TMFA) is to analyze the user's experimental intent serially using single-mode information, which cannot fully utilize the intent information of each mode. Therefore, this paper designs a multimodal fusion algorithm (hereinafter referred to as MFA, Algorithm 4) which focuses on the parallel fusion of user's experimental intent. The essence of the MFA is the fusion of multimodal intent probability. At the same time, this paper designs a smart glove based on the virtual-reality fusion experiments, which can integrate multichannel data such as voice, visual, and sensor. This smart glove can not only capture user's experimental intent but also navigate, guide, or warn user's operation behaviors, and it has stronger perception capabilities compared to any other data glove or smart experimental device. The experimental results demonstrate that the smart glove presented in this paper can be widely employed in the chemical experiment teaching based on virtual-reality fusion. [ABSTRACT FROM AUTHOR]

This paper proposes an intention understanding algorithm (KDI) based on an elderly service robot, which combines Neural Network with a seminaive Bayesian classifier to infer user's intention. KDI algorithm uses CNN to analyze gesture and action information, and YOLOV3 is used for object detection to provide scene information. Then, we enter them into a seminaive Bayesian classifier and set key properties as super parent to enhance its contribution to an intent, realizing intention understanding based on prior knowledge. In addition, we introduce the actual distance between the users and objects and give each object a different purpose to implement intent understanding based on object-user distance. The two methods are combined to enhance the intention understanding. The main contributions of this paper are as follows: (1) an intention reasoning model (KDI) is proposed based on prior knowledge and distance, which combines Neural Network with seminaive Bayesian classifier. (2) A set of robot accompanying systems based on the robot is formed, which is applied in the elderly service scene. [ABSTRACT FROM AUTHOR]

With the development of science and technology and the intensification of the aging of the world's population, elderly care robots have begun to enter people's lives. However, the current elderly care system lacks intelligence and is just a simple patchwork of some traditional elderly products, which is not conducive to the needs of the elderly for easy understanding and easy operation. Therefore, this paper proposes a flexible mapping algorithm (FMFD), that is, a gesture can correspond to a flexible mapping of multiple semantics in the same interactive context. First, combine the input gesture with the surrounding environment to establish the current interactive context. Secondly, when the user uses the same gesture to express different semantics, the feature differences formed due to different cognitions are used as the basis to realize the mapping from one gesture to multiple semantics. Finally, four commonly used gestures are designed to demonstrate the results of flexible mapping. Experiments show that compared with the traditional gesture-based human-computer interaction, the proposed flexible mapping scheme greatly reduces the number of gestures that users need to remember, improves the fault tolerance of gestures in the human-computer interaction process, and meets the concept of elderly caregiver robots. [ABSTRACT FROM AUTHOR]

• The focus of this research is to develop a human-computer interaction system suitable for the elderly care robot. It can extract the intention of fuzzy expression, and the accuracy rate is 97.7%. It can avoid the wrong intention of the elderly due to forgetfulness. A human-computer collaborative division of labor decision is proposed to reduce the user interaction compliance. This paper puts forward a valuable research path for a series of challenging problems in the environment of helping the elderly. This paper proposes a fusion model of an improved Hidden Markov Model (HMM) and an improved CF trust evaluation algorithm, and a human-computer collaboration mechanism embedded in the X-arm manipulator. It can correctly predict user intentions of interaction, correct erroneous intentions, and carry out human-computer collaboration under the user's multimodal fuzzy expression or wrong expression, and finally complete the intention task. It can correctly predict the user interactive intentions, correct the wrong intention, and carry out human-computer collaboration in the case of user multimodal fuzzy expression or wrong expression. The main contributions of this paper are: when humans input intentions to robots, users no longer interact strictly through discrete and explicit interactive operations, but can use non-deterministic and unclear multimodal data to express interactive intentions; When the model predicts user's intentions, it comprehensively considers modal information such as user's voice, gesture, posture, and extracts the user's intention from the user's daily habits and the context of the interactive environment. The process of trust evaluation, reverse analysis, and active acquisition of necessary information on the user's intention determines whether the intention is feasible, and then analyzes the intention through the human-computer collaboration mechanism, reasonably assigns the tasks that the human and the computer need to complete. Finally, human and computer cooperate together to achieve the user's intention. The algorithm proposed in this paper can be considered as a new method of intention understanding in human-computer interaction and a research of exploring the computational principle of natural interaction. The algorithm proposed in this paper has been verified in many daily life scenarios such as helping the elderly to drink water, take medicine, read, watch TV and so on, and it has also proposed a valuable research path for the challenging problem of human-computer interaction. [ABSTRACT FROM AUTHOR]

Initialization for a 3D hand model is significant in determining a 3D hand model that corresponds to a user׳s 3D hand pose in the initial frame. We propose a motion-towards-each-other-based human–computer cooperation algorithm with the aim to reduce the operational burden on the user and cause the computer׳s response to the user׳s operation to fall within acceptable scopes. The computer first recognizes the user׳s interactive intentions through the hand gesture sensing algorithm and at last cooperates with the user until they achieve the meeting point. The main contributions of this study are sensing the operator׳s interactive intentions and focusing on the motion-towards-each-other-based interactive cooperation mechanisms between a human and a computer to achieve low cognitive and operational burdens. The proposed approach can effectively reduce users׳ cognitive and operational burdens. The proposed initialization system is successfully applied to several application systems. [ABSTRACT FROM AUTHOR]

The development of three dimensional (3D) freehand tracking is motivated by the need for 3D freehand-based human computer interface in the Selection-Move-Release (SMR) system. The high dimensionality of articulated hand models is one of the major obstacles for tracking 3D hand in real time. A practical Task-Phase-Trajectory-Mentality (TPTM) model for the SMR system is built and a novel freehand tracking algorithm is introduced by using a camera that is calibrated by a charge-coupled device. The TPTM model is proposed and explained based on the cognitive theory in the proposed method. The manner by which the TPTM model guides the realtime design of freehand tracking is also shown. The proposed method reduces the high dimensionality down to eight degrees of freedom (DOF), optimizes the dynamic number of particles for each frame, and achieves realtime hand tracking and high accuracy of 3D full DOF hand gestures. The algorithm presented in this paper is successfully applied to a 3D virtual teaching system. [ABSTRACT FROM AUTHOR]

Abstract: Articulated hand tracking systems have been commonly used in virtual reality applications, including selection–move–release systems that use freehand as input device. However, building an effective real-time hand tracker remains a challenge. Motivated by the need for three-dimensional (3D) freehand-based human–computer interface in the selection–move–release systems, the current study aims to develop 3D freehand tracking, in which the high dimensionality of articulated hand models is one of the dominating obstacles for tracking 3D hand in real-time. This study focuses on forming the behavioral model for users in selection–move–release systems to build a natural, direct, and effective human–computer interface. First, the statistical models for users were learned. Second, the behavioral models were derived from the statistical models to form a general behavioral model database. Third, a freehand tracking algorithm for interaction between human and computer was presented based on the behavioral models. Fourth, the proposed approach was tested in several real selection–move–release systems, and the experimental results were provided. Unlike previous studies on this subject, the proposed behavioral model was composed of Correlation among Local Motion Models and cognitive information. Experimental results show that the proposed algorithm can achieve satisfactory results in relation to time cost and accuracy compared with some previous studies. [Copyright &y& Elsevier]

Abstract: Human hand shape features extraction from image frame sequences is one of the key steps in human hand 2D/3D tracking system and human hand shape recognition system. In order to satisfy the need of human hand tracking in real time, a fast and accurate method for acquirement of edge features from human hand images without consideration of hand over face is put forward in this paper. The proposed approach is composed of two steps, the coarse location phase (CLP) and the refined location phase (RLP) from coarseness to refinement. In the phase of CLP, the hand contour is approximately described by a polygon with concave and convex, an approach to obtaining hand shape polygon using locating points and locating lines is meticulously discussed. Then, a coarse location (CL) algorithm for extraction of interested hand shape features, such as contour, fingertips, roots of fingers, joints and the intersection of knuckle on different fingers, is proposed. In the phase of RLP, a multi-scale approach is introduced into our study to refine the features obtained by the CL algorithm. By means of defining the response strength of different types of features, a refined location (RL) algorithm is proposed. The major contribution of this paper is that the novel detection operators for features of hand images are presented in the above two steps, which have been successfully applied to our 3D hand shape tracking system and 2D hand shape recognition system. A number of comparative studies with real images and online videos demonstrate that the proposed method can extract the three defined human hand image features with high accuracy and high speed. [ABSTRACT FROM AUTHOR]

Driving distraction detection can effectively prevent the occurrence of traffic accidents. Thus, monitoring a driver's state is very important for road safety. At present, most driving distraction detection methods focus on singular aspects, such as gaze distraction or hand distraction, and rarely focus on cognitive distraction as it is difficult to detect. This study proposed a driving distraction detection method based on gaze activity. A multi‐channel convolutional neural network was used to classify the driver's gaze area and calculate the gaze activity. In cognitive distraction, the driver's gaze activity is significantly lower than that in the normal driving state. The activity thresholds via experiments have been obtained and used to determine whether drivers were in a cognitively distracted state. Through experiments, the accuracy of this method reached 92.36%. To identify driving distractions more comprehensively, a gaze distraction algorithm based on the two‐second rule has been added to the method. The experiment demonstrated that our method in combination with this algorithm improved all of the indicators (when compared to only the gaze activity algorithm), and the accuracy rate increased to 95%. [ABSTRACT FROM AUTHOR]

Various deep learning‐based video frame interpolation methods have been proposed in the past few years, but how to generate high quality interpolated frames in videos with large motions, complex backgrounds and rich textures is still a challenging issue. To deal with this limitation, a frame interpolation method based on residual blocks and feature pyramids is proposed. U‐Net is the main architecture of our method, which can capture multi‐layer information, segment objects from the background and obtain parameters with motion information to guide frame interpolation. However, the upsampling and subsampled of U‐Net will lose important information. In order to acquire more detailed contextual information, shortcut connection is used in the encoder basic module. At the same time, feature pyramid network is employed to capture features at different scales of the decoder to improve the representation of inter‐frame spatial‐temporal features. The experimental results show that the proposed method outperform the baseline methods in both of objective and subjective evaluations on different datasets. In particular, the method has obvious advantages on datasets which contain complex background. [ABSTRACT FROM AUTHOR]

The ubiquitous use and advancement in built-in smartphone sensors and the development in big data processing have been beneficial in several fields including healthcare. Among the basic vitals monitoring, pulse rate monitoring is the most important healthcare necessity. A multimedia video stream data acquired by built-in smartphone camera can be used to estimate it. In this paper, an algorithm that uses only smartphone camera as a sensor to estimate pulse rate using PhotoPlethysmograph (PPG) signals is proposed. The results obtained by the proposed algorithm are compared with the actual pulse rate and the maximum error found is 3 beats per minute. The standard deviation in percentage error and percentage accuracy is found to be 0.68 % whereas the average percentage error and percentage accuracy is found to be 1.98 % and 98.02 % respectively. [ABSTRACT FROM AUTHOR]

Highlights: [•] Ubiquitous sensors are used to collect human activity data. [•] Human activity recognition is modeled as a classification problem. [•] Extract features combining pattern mining with discriminative ability descriptor. [•] Features with low or high frequency have limited discriminative power. [Copyright &y& Elsevier]

Video frame interpolation is used to generate intermediate frames by estimating the movement of pixels between the input frames. However, problems of blurring, object occlusion, and sudden brightness changes occur in naturally obtained video frames. We propose a context-based video frame interpolation method via depthwise over-parameterized convolution. First, the proposed network obtains the context graphs of the input frames. Subsequently, an adaptive collaboration of flows is adopted to warp the input frames and the context graphs. Then, the frame synthesis network is used to fuse the warped input frames and context graphs to obtain a preliminary estimate of the interpolated frame. Finally, a post-processing module is employed to refine the result. Experimental results on several datasets demonstrate that the proposed method performs qualitatively and quantitatively better than state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Pruning methods to compress and accelerate deep convolutional neural networks (CNNs) have recently attracted growing attention, with the view of deploying pruned networks on resource-constrained hardware devices. However, most existing methods focus on small granularities, such as weight, kernel and filter, for the exploration of pruning. Thus, it will be bound to iteratively prune the whole neural networks based on those small granularities for high compression ratio with little performance loss. To address these issues, we theoretically analyze the relationship between the activation and gradient sparsity, and the channel saliency. Based on our findings, we propose a novel and effective method of weak sub-network pruning (WSP). Specifically, for a well-trained network model, we divide the whole compression process into two non-iterative stages. The first stage is to directly obtain a strong sub-network by pruning the weakest one. We first identify the less important channels from all the layers and determine the weakest sub-network, whereby each selected channel makes a minimal contribution to both the feed-forward and feed-backward processes. Then, a one-shot pruning strategy is executed to form a strong sub-network enabling fine tuning, while significantly reducing the impact of the network depth and width on the compression efficiency, especially for deep and wide network architectures. The second stage is to globally fine-tune the strong sub-network using several epochs to restore its original recognition accuracy. Furthermore, our proposed method impacts on the fully-connected layers as well as the convolutional layers for simultaneous compression and acceleration. Comprehensive experiments on VGG16 and ResNet-50 involving a variety of popular benchmarks, such as ImageNet-1K, CIFAR-10, CUB-200 and PASCAL VOC, demonstrate that our WSP method achieves superior performance on classification, domain adaption and object detection tasks with small model size. Our source code is available at https://github.com/QingbeiGuo/WSP.git. [ABSTRACT FROM AUTHOR]

Although group convolution operators are increasingly used in deep convolutional neural networks to improve the computational efficiency and to reduce the number of parameters, most existing methods construct their group convolution architectures by a predefined partitioning of the filters of each convolutional layer into multiple regular filter groups with an equal spatial group size and data-independence, which prevents a full exploitation of their potential. To tackle this issue, we propose a novel method of designing self-grouping convolutional neural networks, called SG-CNN, in which the filters of each convolutional layer group themselves based on the similarity of their importance vectors. Concretely, for each filter, we first evaluate the importance value of their input channels to identify the importance vectors, and then group these vectors by clustering. Using the resulting data-dependent centroids, we prune the less important connections, which implicitly minimizes the accuracy loss of the pruning, thus yielding a set of diverse group convolution filters. Subsequently, we develop two fine-tuning schemes, i.e. (1) both local and global fine-tuning and (2) global only fine-tuning, which experimentally deliver comparable results, to recover the recognition capacity of the pruned network. Comprehensive experiments carried out on the CIFAR-10/100 and ImageNet datasets demonstrate that our self-grouping convolution method adapts to various state-of-the-art CNN architectures, such as ResNet and DenseNet, and delivers superior performance in terms of compression ratio, speedup and recognition accuracy. We demonstrate the ability of SG-CNN to generalize by transfer learning, including domain adaption and object detection, showing competitive results. Our source code is available at https://github.com/QingbeiGuo/SG-CNN.git. [ABSTRACT FROM AUTHOR]

In this paper, the QS-1 sediments (3.64 m long; at 1500 m water depth) collected from the Qiongdongnan Basin were used to analyze the relationship between sedimentary geochemical anomalies and methane seepage. Mo-U covariation, Ce anomaly, and Mg/Ca-Sr/Ca end-member model were analyzed to identify the variations of redox environment and authigenic carbonate precipitation, finally constructing the methane release events in this study area. These observations suggest that: 1) Significant Mo/U enrichments and Ba front indicate the presence of high-flux active methane seepage with SMI of 260 cmbsf in this study area. The depositional environment evolved as oxic-anoxic-reducing-sulfide environment from bottom to top, as evidenced by Mo-U covariation and Ce anomaly; 2) Based on Sr/Ca and Mg/Ca end-member model proposed by Bayon et al. (2007), the high-Mg calcite was the dominant authigenic carbonate currently which usually precipitated behind the methane seep events; and 3) The active cold seep can be divided into three methane release events (MREs): MRE I, the most intense methane seepage occurred at 22.1 ka BP; MRE II, three low-intensity methane seepage occurred at 19.3, 16.5, and 11.3 ka BP, respectively; and MRE III, the continuous methane seep began at 9.3 ka BP until now. The synthesis of geochemical indicators has significant implications for tracing methane seepage of varying intensity and duration. • High quality samples drilled from high-flux active cold seep area were investigated. • AOM caused Mo/U enrichments and authigenic carbonate precipitation. • Sedimentary Mg/Ca and Sr/Ca ratios revealed the dominant presence of high-Mg calcite. • Three methane release events were identified using combined geochemical proxies. [ABSTRACT FROM AUTHOR]

With the development of deep learning, fatigue detection technology for drivers has achieved remarkable achievements. Although the image-based approach achieves good accuracy, it inevitably leads to greater model complexity, which is unsuitable for mobile terminal devices. Luckily, human skeletal data significantly reduces the impact of noise and input data volume while retaining valid information, and it can better deal with real-world driving scenarios with the benefit of robustness in complex driving situations. This paper proposes a lightweight multi-scale spatio-temporal attention graph convolutional network (MS-STAGCN) to efficiently utilize skeleton data to identify driver states by aggregating locally and globally valid face information, which achieves good performance even for lightweight design. The experimental results show that the method achieves 92.4% accuracy on the NTHU-DDD dataset, which can be applied to fatigue detection tasks of the driver in real-world driving scenarios in the future. • A non-threshold method based on the facial skeleton for driver fatigue detection in video. • Multi-scale spatio-temporal features are effectively captured. • Performance outperforms the compared methods on the NTHU-DDD and NTU-RGBD datasets. [ABSTRACT FROM AUTHOR]

Coronal Mass Ejection (CME) is a major solar activity that affects the earth, thus CMEs detection is of great importance for space weather forecast, disaster prevention and reduction. We model the detection of Coronal Mass Ejections (CMEs) as the classification of the brightest block in the current running difference image. Because CMEs usually correspond to the areas with high gray values or complex texture features, multiple features including gray features and texture features are extracted to represent the brightest block. And classifier is designed based on these features. Our method includes four steps: first, because the CMEs spread along the radial direction of the sun, in order to facilitate the analysis, the original coordinate is transformed into the polar coordinate; Secondly, because the typical appearance of the CMEs is bright or complex texture enhancement, we use the brightest block to represent the whole image; Thirdly, we extract the gray, texture and HOG features of the brightest gray blocks. Finally, we use the extracted features to design decision trees as the base classifiers, and AdaBoost is used to obtain the final ensemble classifier. As far as we know, this is the first time that the learning based classification framework is presented in the CMEs detection. Moreover, multiple feature fusion is first used to model the various CMEs. Experimental results show that the integration of multi-feature based detection algorithm proposed can achieve better detection results. [ABSTRACT FROM AUTHOR]

Stock comments published by experts are important references for accurate stock trends prediction. How to comprehensively and accurately capture the topic of expert stock comments is an important issue which belongs to text classification. The Bidirectional Encoder Representations from Transformers (BERT) pretrained language model is widely used for text classification, due to its high identification accuracy. However, BERT has some limitations. First , it only utilizes fixed length text, leading to suboptimal performance in long text information exploration. Second , it only relies on the features extracted from the last layer, resulting in incomprehensive classification features. To tackle these issues, we propose a multi-layer features ablation study of BERT model for accurate identification of stock comments' themes. Specifically, we firstly divide the original text to meet the length requirement of the BERT model based on sliding window technology. In this way, we can enlarge the sample size which is beneficial for reducing the over-fitting problem. At the same time, by dividing the long text into multiple short texts, all the information of the long text can be comprehensively captured through the synthesis of the subject information of multiple short texts. In addition , we extract the output features of each layer in the BERT model and apply the ablation strategy to extract more effective information in these features. Experimental results demonstrate that compared with non-intercepted comments, the topic recognition accuracy is improved by intercepting stock comments based on sliding window technology. It proves that intercepting text can improve the performance of text classification. Compared with the BERT, the multi-layer features ablation study we present in the paper further improves the performance in the topic recognition of stock comments, and can provide reference for the majority of investors. Our study has better performance and practicability on stock trend prediction by stock comments topic recognition. [ABSTRACT FROM AUTHOR]

• We build a new standalone control module based on the scaled sigmoid function to enrich the neural network module family to enable the neural architecture optimization. • Our DNAL method produces no candidate neural architectures but one, thus drastically improving the learning efficiency, i.e., costing 20 epochs for CIFAR-10 and 10 epochs for ImageNet. • It is applicable to conventional CNNs, lightweight CNNs, and stochastic supernets. • Extensive experiments confirm that our DNAL method achieves excellent performance on various CNN architectures, including VGG16, ResNet50, MobileNetV2, and ProxylessNAS, over the task of CIFAR-10 and ImageNet-1K classification. Efficient neural networks has received ever-increasing attention with the evolution of convolutional neural networks (CNNs), especially involving their deployment on embedded and mobile platforms. One of the biggest problems to obtaining such efficient neural networks is efficiency, even recent differentiable neural architecture search (DNAS) requires to sample a small number of candidate neural architectures for the selection of the optimal neural architecture. To address this computational efficiency issue, we introduce a novel architecture parameterization based on scaled sigmoid function , and propose a general Differentiable Neural Architecture Learning (DNAL) method to obtain efficient neural networks without the need to evaluate candidate neural networks. Specifically, for stochastic supernets as well as conventional CNNs, we build a new channel-wise module layer with the architecture components controlled by a scaled sigmoid function. We train these neural network models from scratch. The network optimization is decoupled into the weight optimization and the architecture optimization, which avoids the interaction between the two types of parameters and alleviates the vanishing gradient problem. We address the non-convex optimization problem of efficient neural networks by the continuous scaled sigmoid method instead of the common softmax method. Extensive experiments demonstrate our DNAL method delivers superior performance in terms of efficiency, and adapts to conventional CNNs (e.g., VGG16 and ResNet50), lightweight CNNs (e.g., MobileNetV2) and stochastic supernets (e.g., ProxylessNAS). The optimal neural networks learned by DNAL surpass those produced by the state-of-the-art methods on the benchmark CIFAR-10 and ImageNet-1K dataset in accuracy, model size and computational complexity. Our source code is available at https://github.com/QingbeiGuo/DNAL.git. [ABSTRACT FROM AUTHOR]