Abstract ：

With the rapid development of mobile technology and smart devices, crowdsensing has shown its large potential to collect massive data. Considering the limitation of calculation power, edge computing is introduced to release unnecessary data transmission. In edge-computing-enabled crowdsensing, massive data is required to be preliminary processed by edge computing devices (ECDs). Compared with the traditional central platform, these ECDs are limited by their own capability so they may only obtain part of relative factors and they can't process data synthetically. ECDs involved in one task are required to cooperate to process the task data. The privacy of participants is important in crowdsensing, so blockchain is used due to its decentralization and tamper-resistance. In crowdsensing tasks, it is usually difficult to obtain the assessment criteria in advance so reinforcement learning is introduced. As mentioned before, ECDs can't process task data comprehensively and they are required to cooperate quality assessment. Therefore, a blockchain-based framework for data quality in edge-computing-enabled crowdsensing (BFEC) is proposed in this paper. DPoR (Delegated Proof of Reputation), which is proposed in our previous work, is improved to be suitable in BFEC. Iteratively, the final result is calculated without revealing the privacy of participants. Experiments on the open datasets Adult, Blog, and Wine Quality show that our new framework outperforms existing methods in executing sensing tasks.

Keyword ：

block-chain crowdsensing edge computing devices quality assessment reinforcement learning

Cite：

Copy from the list or Export to your reference management。

Abstract ：

Tax evasion is an illegal activity in which individuals or entities avoid paying their true tax liabilities. Efficient detection of tax evasion has always been a crucial issue for both governments and academic researchers. Recent research has proposed the use of machine learning technology to detect tax evasion and has shown good results in some specific areas. Regrettably, there are still two major obstacles to detect tax evasion. First, it is hard to extract powerful features because of the complexity of tax data. Second, due to the complicated process of tax auditing, labeled data are limited in practice. Such obstacles motivate the contributions of this work. In this paper, we propose a novel tax evasion detection framework named FBNE-PU (Fusion of the basic feature and network embedding with PU learning for tax evasion detection), a multistage method for detecting tax evasion in real-life scenarios. In this paper, we perform an in-depth analysis of the characteristics of the transaction network and propose a novel network embedding algorithm, the PnCGCN. It significantly improves detection performance by extracting powerful features from basic features and the tax-related transaction network. Moreover, we use nnPU (positive-unlabeled learning with non-negative risk estimator) to assign pseudo labels for unlabeled data. Finally, an MLP is trained as the decision function. Experiments on three real-world datasets demonstrate that our method significantly outperforms the comparison methods in the tax evasion detection task. Additionally, the source code and the experimental details have been made available at (https://github.com/PiggyGaGa/FBNE-PU).

Keyword ：

graph convolutional network network embedding PU learning Tax evasion detection transaction network

Cite：

Copy from the list or Export to your reference management。

Abstract ：

Visual question answering has been greatly advanced by deep learning technologies, but still remains an open problem subjected to two aspects of factors. First, previous works estimate the correctness of each candidate answer mainly by its semantic correlations with visual questions, overlooking the fact that some questions and their answers are semantically inconsistent. Second, previous works that require external knowledge mainly uses the knowledge facts retrieved by key words or visual objects. However, the retrieved knowledge facts may only be related to the semantics of the question, but are useless or even misleading for answer prediction. To address these issues, we investigate how to capture the purpose of visual questions and propose a Purpose Guided Visual Question Answering model, called PGVQA. It mainly has two appealing properties: (1) It can estimate the correctness of candidate answers based on the Question Purpose (QP) that reveals which aspects of the concept are examined by visual questions. This is helpful for avoiding the negative effect of the semantic inconsistency between answers and questions. (2) It can incorporate the knowledge facts accordant with the QP into answer prediction, which helps to improve the probability of answering visual questions correctly. Empirical studies on benchmark datasets show that PGVQA achieves state-of-the-art performance. © 2022 Elsevier Ltd

Keyword ：

Benchmarking Deep learning Knowledge based systems Semantics

Cite：

Copy from the list or Export to your reference management。

Abstract ：

In complementary-label learning (CLL), the complementary transition matrix, denoting the probabilities that true labels flip into complementary labels (CLs) which specify classes observations do not belong to, is crucial to building statistically consistent classifiers. Most existing works implicitly assume that the transition probabilities are identical, which is not true in practice and may lead to undesirable bias in solutions. Few recent works have extended the problem to a biased setting but limit their explorations to modeling the transition matrix by exploiting the complementary class posteriors of anchor points (i.e., instances that almost certainly belong to a specific class). However, due to the severe corruption and unevenness of biased CLs, both anchor points and complementary class posteriors are difficult to predict accurately in the absence of true labels. In this article, rather than directly predicting these two error-prone items, we instead propose a divided-T estimator as an alternative to effectively learn transition matrices from only biased CLs. Specifically, we exploit semantic clustering to mitigate the adverse effects arising from CLs. By introducing the learned semantic clusters as an intermediate class, we factorize the original transition matrix into the product of two easy-to-estimate matrices that are not reliant on the two error-prone items. Both theoretical analyses and empirical results justify the effectiveness of the divided-T estimator for estimating transition matrices under a mild assumption. Experimental results on benchmark datasets further demonstrate that the divided-T estimator outperforms state-of-the-art (SOTA) methods by a substantial margin.

Keyword ：

Biased complementary labels (CLs) Estimation error multiclass classification Research and development Robustness Self-supervised learning semantic clustering Semantics Training Training data transition matrix

Cite：

Copy from the list or Export to your reference management。

Abstract ：

The multiuser behavior recognition task based on environmental sensors can provide reliable health monitoring, suspicious person identification and behavior correction. Compared with camera equipment and wearable sensors, the task can achieve acquisition of binary data from the environmental sensors without requiring wearable sensors. Therefore, privacy protection of users and use burden can be improved. However, there are still challenges in this behavior recognition scenario: First, the data consistency shown by the different behaviors of a single user in the same scenario need to be guaranteed. Second, the interactive behavior of multiusers may cause a data association problem. Therefore, the multiuser behavior recognition task based on environmental sensors has, apart from application value, important research challenges. In response, we propose the divide and conquer dynamic memory network model (DC-DMN). Based on the periodicity of user behavior, personal habits, time and spatial characteristics, the multiuser behavior recognition ability of the model can be enhanced. First, the GRU model is used to solve the consistency problem of different behaviors at the data level. Then, we expand the model memory based on the idea of a dynamic memory network. In addition, two sections of memory are designed to integrate and store data more effectively. In this way, the data association and support problem can be solved. Finally, we use three standard datasets to conduct experiments and compare them with the existing benchmark methods in two dimensions of accuracy and recall. Experiments show that DC-DMN performs well in three different datasets. It can effectively solve the problems of data consistency and data association, thereby improving the recognition accuracy.

Keyword ：

attention mechanism data association Data models dynamic memory network framework Heuristic algorithms Hidden Markov models Intelligent sensors Multiuser behavior recognition Sensors Task analysis Wearable sensors

Cite：

Copy from the list or Export to your reference management。

Abstract ：

Background Bio-entity Coreference Resolution (CR) is a vital task in biomedical text mining. An important issue in CR is the differential representation of identical mentions as their similar representations may make the coreference more puzzling. However, when extracting features, existing neural network-based models may bring additional noise to the distinction of identical mentions since they tend to get similar or even identical feature representations. Methods We propose a context-aware feature attention model to distinguish similar or identical text units effectively for better resolving coreference. The new model can represent the identical mentions based on different contexts by adaptively exploiting features, which enables the model reduce the text noise and capture the semantic information effectively. Results The experimental results show that the proposed model brings significant improvements on most of the baseline for coreference resolution and mention detection on the BioNLP dataset and CRAFT-CR dataset. The empirical studies further demonstrate its superior performance on the differential representation and coreferential link of identical mentions. Conclusions Identical mentions impose difficulties on the current methods of Bio-entity coreference resolution. Thus, we propose the context-aware feature attention model to better distinguish identical mentions and achieve superior performance on both coreference resolution and mention detection, which will further improve the performance of the downstream tasks.

Keyword ：

Context-aware Coreference resolution Mention detection Neural network

Cite：

Copy from the list or Export to your reference management。

Abstract ：

Video summarization (VS) is a widely used technique for facilitating the effective reading, fast comprehension, and effective retrieval of video content. Certain properties of the new video data, such as a lack of prominent emphasis and a fuzzy theme development border, disturb the original thinking mode based on video feature information. Moreover, it introduces new challenges to the extraction of video depth and breadth features. In addition, the diversity of user requirements creates additional complications for more accurate keyframe screening issues. To overcome these challenges, this paper proposes a hierarchical spatial–temporal cross-attention scheme for video summarization based on comparative learning. Graph attention networks (GAT) and the multi-head convolutional attention cell are used to extract local and depth features, while the GAT-adjusted bidirection ConvLSTM (DB-ConvLSTM) is used to extract global and breadth features. Furthermore, a spatial–temporal cross-attention-based ConvLSTM is developed for merging hierarchical characteristics and achieving more accurate screening in similar keyframes clusters. Verification experiments and comparative analysis demonstrate that our method outperforms state-of-the-art methods. © 2022 by the authors.

Keyword ：

cross-attention; spatial–temporal features; video summarization

Cite：

Copy from the list or Export to your reference management。

Abstract ：

Currently, deep learning methods have been widely applied and thus promoted the development of different fields. In the financial accounting field, the rapid increase in the number of financial tickets dramatically increases labor costs; hence, using a deep learning method to relieve the pressure on accounting is necessary. At present, a few works have applied deep learning methods to financial ticket recognition. However, first, their approaches only cover a few types of tickets. In addition, the precision and speed of their recognition models cannot meet the requirements of practical financial accounting systems. Moreover, none of the methods provides a detailed analysis of both the types and content of tickets. Therefore, this paper first analyzes the different features of 482 kinds of financial tickets, divides all kinds of financial tickets into three categories, and proposes different recognition patterns for each category. These recognition patterns can meet almost all types of financial ticket recognition needs. Second, regarding the fixed format types of financial tickets (accounting for 68.27% of the total types of tickets), we propose a simple yet efficient network named the Financial Ticket Faster Detection network (FTFDNet) based on a Faster R-CNN. Furthermore, according to the characteristics of the financial ticket text, in order to obtain higher recognition accuracy, the loss function, Region Proposal Network (RPN), and Non-Maximum Suppression (NMS) are improved to make FTFDNet focus more on text. Finally, we perform a comparison with the best ticket recognition model from the ICDAR2019 invoice competition. The experimental results prove the effectiveness of these improvements. The accuracy of this method reaches 97.4% and the recognition speed increases by 50%.

Keyword ：

Deep learning Financial accounting Image text recognition Ticket detection

Cite：

Copy from the list or Export to your reference management。

Abstract ：

With the development of the economy, the number of financial tickets is increasing. The traditional invoice reimbursement and entry work bring more and more burden to financial accountants. However, standard OCR technology weakly supports financial tickets with various layouts and mixed Chinese and English characters. In view of this problem, this paper designs a method of financial ticket all-content text information detection and recognition based on deep learning. This method can effectively suppress the common noise of ticket image and extract financial information from ticket image in batch. At the same time, aiming at the problem of multi-character mixed character recognition, we propose a financial ticket character recognition framework (FTCRF), which can improve the accuracy of multi-character mixed character recognition and make the detection and recognition of financial ticket surface information more efficient. The experimental results show that the average recognition accuracy of the character sequence is 91.75%. The average recognition accuracy of the whole ticket is 87%, which significantly improves the efficiency of the financial accounting system.

Keyword ：

Deep learning Financial accounting Image text recognition Ticket detection

Cite：

Copy from the list or Export to your reference management。

Abstract ：

With the rapid development of wearable devices with various sensors, massive sensing data for health management have been generated. This causes a potential revolution in medical treatments, diagnosis, and prediction. However, due to the privacy risks of health data aggregation, data comparative analysis under privacy protection faces challenges. Order-preserving encryption is an effective scheme to achieve private data retrieval and comparison, but the existing order-preserving encryption algorithms are mainly aimed at either integer data or single characters. It is urgent to build a lightweight order-preserving encryption scheme that supports multiple types of data such as integer, floating number, and string. In view of the above problems, this paper proposes an order-preserving encryption scheme (WRID-OPES) based on weighted random interval division (WRID). WRID-OPES converts all kinds of data into hexadecimal number strings and calculates the frequency and weight of each hexadecimal number. The plaintext digital string is blocked and recombined, and each block is encrypted using WRID algorithm according to the weight of each hexadecimal digit. Our schemes can realize the order-preserving encryption of multiple types of data and achieve indistinguishability under ordered selection plaintext attack (IND-OCPA) security in static data sets. Security analysis and experiments show that our scheme can resist attacks using exhaustive methods and statistical methods and has linear encryption time and small ciphertext expansion rate.

Keyword ：

IoT; order-preserving encryption; privacy protection; random interval division; wearable devices

Cite：

Copy from the list or Export to your reference management。