Intelligent Computing and Block Chain

Retinal vascular morphometry is an important biomarker of eye-related cardiovascular diseases such as diabetes and hypertension. And retinal vessel segmentation is a fundamental step in fundus image analyses and diagnoses. In recent years, deep learning based networks have achieved superior performance in medical image segmentation. However, for fine vessels or terminal branches, most existing methods tend to miss or under-segment those structures, inducing isolated breakpoints. In this paper, we proposed Bi-Directional ConvLSTM U-Net with Generative Adversarial Training (BLU-GAN), a novel deep learning model based on U-Net that generates precise predictions of retinal vessels combined with generative adversarial training. Bi-directional ConvLSTM, which can better integrate features from different scales through a coarse-to-fine memory mechanism, is employed to non-linearly combine feature maps extracted from encoding path layers and the previous decoding up-convolutional layers and to replace the simple skip-connection used in the original U-Net. Moreover, we use densely connected convolutions in certain layers to strengthen feature propagation, encourage feature reuse, and substantially reduce the number of parameters. Through extensive experiments, BLU-GAN has shown leading performance among the state-of-the-art methods on the DRIVE, STARE, CHASE_DB1 datasets for retinal vessel segmentation.

Choroidal neovascularization (CNV) is a retinal vascular disease that new vessels sprout from the choroid and then grow into retina, which usually appears in the late stage of Age-related macular degeneratoin (AMD). Because of the complex characteristics of CNV, it is time-consuming and laborious to manually segment CNV from spectral-domain optical coherence tomography (SD-OCT) images. In this paper, we propose an improved 3D U-Net model based on anisotropic convolution to segment CNV automatically. First, to adapt the special morphology and differences in the scale of CNV, a cross convolution module (CCM) is designed based on anisotropic convolution instead of general convolution with kernel size 3 $$\times $$ × 3 $$\times $$ × 3 in standard 3D U-Net architecture. Then, the first layer is adjusted to reduce the usage of GPU. In addition, a dilated connection layer (DCL) is proposed to improve the ability to capture multi-scale information while using low-level features. After verifying on a dataset that consists of 376 cubes from 21 patients, we obtain the mean dice of 84.73%. The experimental evaluation shows that our approach is able to achieve effective segmentation of CNV from SD-OCT images.

Diagnosis, treatment, and prevention of mental illness requires brain-based biomarkers that can serve as effective targets of evaluation. Here we target the neuroanatomically and neurophysiologically well-defined neuromoduatory systems that serve the computational role of generating globally-synchronized neural activity for the purpose of functional integration. By using the second-order blind identification (SOBI) algorithm, which works with temporal information, we show that (1) neuroelectrical signals associated with synchronized global network activity can be extracted from resting state EEG; (2) the SOBI extracted global resting state network (gRSN) can be quantitatively characterized by its spatial configuration, operationally defined as a hits vector; (3) individual differences in the gRSN’s spatial configuration can be analyzed and visualized in hits vector defined high-dimensional space. Our streamed-lined process offers a novel enabling technology to support rapid and low-cost assessment of much larger and diverse populations of individuals, addressing several methodological limitations in current investigation of brain function.

Segmentation of white matter hyperintensities (WMHs) from MR images is an essential step in computer-aided diagnosis of brain diseases, especially when considering their effect on cognition or stroke. At present, most of the research for WMH segmentation is based on deep learning methods. Although many deep learning segmentation methods have been proposed, their accuracy of these methods still needs to be improved, especially for discrete and small-sized deep WMHs. To cope with these challenges, and to improve the accuracy of WMH segmentation, an improved 3D U-net model, named PRU-net, was proposed in this paper. PRU-net integrates pyramid pooling and residual convolutional block in bottleneck layer of the U-net architecture. The pyramid pooling block was used to aggregate more context information, and the residual convolutional block was used to deepen the depth of bottleneck layers. Both the two blocks were employed to enhance the feature extraction of U-net. The experiments were based on the MICCAI 2017’s WMH Challenge datasets, and the results showed that the Dice similarity coefficient (DSC) of our method was 0.83 and the F1 score was 0.84, which were higher than those of compared methods. Through visual observation of the segmentation results, our method cans not only accurately segment large lesion areas, but also distinguish small lesions which are difficult to segment for conventional U-net models.

The image segmentation of port wine stains (PWS) lesions is of great significance to assess PDT treatment outcomes. However, it mainly depends on the manual division of doctors at present, which is time-consuming and laborious. Therefore, it is urgent and necessary to explore an efficient and accurate automatic extraction method for PWS lesion images. A two-way perceived color difference saliency algorithm (TPCS) for PWS lesion extraction is proposed to improve the efficiency and accuracy, and is compared with other image segmentation algorithms. The proposed algorithm shows the best performance with 88.91% accuracy and 96.36% sensitivity over 34 test images of PWS lesions.

Identifying neuroimaging-based biomarkers is greatly needed to boost the progress of mental disorder diagnosis. However, it has been well acknowledged that inaccurate diagnosis on mental disorders may in turn raise unreliable biomarkers. In this paper, we propose a new method that can detect typical patients with specific mental disorders, which is beneficial to further biomarker identification. In our method, we extend an advanced sample noise detection technology based on random forest to identify typical patients, and apply it to identify typical subjects from schizophrenia (SZ) and bipolar disorder (BP) patients with neuroimaging features estimated from resting fMRI data. To evaluate the capacity of our method, we investigate the typical subjects and whole subjects with respect to group differences, classification accuracy, clustering, and projection performance based on the identified typical subjects. Our results supported that the typical subjects showed greater group differences between SZ and BP, higher classification accuracy, more compact clusters in both clustering and projection. In short, our work presents a novel method to explore discriminative and typical subjects for different mental disorders, which is promising for identifying reliable biomarkers.

Machine learning is a powerful tool for simulating the quantitative structure activity relationship in drug discovery (QSAR). However, descriptor selection and model optimization remain two of most challenging tasks for domain experts to construct high-quality QSAR model. Therefore, we propose a QSAR-special automated machine learning method incorporating Automated Descriptor Selection with Automated Model Building (ADSMB) to efficiently and automatically build high-quality QSAR model. Automated Descriptor Selection provides a QSAR-special molecular descriptor selection mechanism to automatically obtain the descriptors without unique value, redundancy and low importance in QSAR dataset. Based on these QSAR-special descriptors, Automated Model Building constructs high-quality ensemble model of molecular descriptors and target activities under Bayesian optimization through Auto-Sklearn. Finally, we conduct experimental evaluation for our proposed method on Mutagenicity dataset. The results show ADSMB can obtain better and stable performance than the competing methods.

Early diagnosis of brain tumor is crucial for treatment planning. Quantitative analyses of segmentation can provide information for tumor survival prediction. The effectiveness of convolutional neural network (CNN) has been validated in medical image segmentation. In this study, we apply a widely-employed CNN namely UNet to automatically segment out glioma sub-regions, and then extract their volumes and surface areas. A sophisticated machine learning scheme, consisting of mutual information feature selection and multivariate linear regression, is then used to predict individual survival time. The proposed method achieves an accuracy of 0.475 on 369 training data based on leave-one-out cross-validation. Compared with using all features, using features obtained from the employed feature selection technology can enhance the survival prediction performance.

Time series data with multidimensional attributes (such as network flow, page view and advertising revenue) are common and important performance monitoring metrics in large-scale Internet services. When the performance monitoring metrics data deliver abnormal patterns, it is of critical importance to timely locate and diagnose the root cause. However, this task remains as a challenge due to tens of thousands of attribute combinations in search space. Moreover, faults will propagate from one attribute combination to another, resulting in a subtle mathematical relationship between the different attribute combinations. Only after spotting the root cause from the huge search space, one can take appropriate actions to mitigate the problem and keep the system running uninterrupted. In this paper, we propose a novel root cause localization algorithm to identify the attribute combinations most likely to blame. Tests on the real-world data from an Internet company show that this algorithm achieves an averaged F-score over 0.95 with a localization time less than 30 s.

Modern business intelligence relies on efficient processing on very large amount of stream data, such as various event logging and data collected by sensors. To meet the great demand for stream processing, many stream data storage systems have been implemented and widely deployed, such as Kafka, Pulsar and DistributedLog. These systems differ in many aspects including design objectives, target application scenarios, access semantics, user API, and implementation technologies. Each system use a dedicated tool to evaluate its performance. And different systems measure different performance metrics using different loads. For infrastructure architects, it is important to compare the performances of different systems under diverse loads using the same benchmark. Moreover, for system designers and developers, it is critical to study how different implementation technologies affect their performance. However, there is no such a benchmark tool yet which can evaluate the performances of different systems. Due to the wide diversities of different systems, it is challenging to design such a benchmark tool. In this paper, we present SSBench, a benchmark tool designed for stream data storage systems. SSBench abstracts the data and operations in different systems as “data streams” and “reads/writes” to data streams. By translating stream read/write operations into the specific operations of each system using its own APIs, SSBench can evaluate different systems using the same loads. In addition to measure simple read/write performance, SSBench also provides several specific performance measurements for stream data, including end-to-end read latency, performance under imbalanced loads and performance of transactional loads. This paper also presents the performance evaluation of four typical systems, Kafka, Pulsar, DistributedLog and ZStream, using SSBench, and discussion of the causes for their performance differences from the perspective of their implementation techniques.

As the scales of cluster systems increase, failures become normal and make reliability be a major concern. System logs contain rich semantic information of all components, and they are always used to detect abnormal behaviors and failures by mining methods.In this paper, we perform a failure characterization study of the Bluegene/L system logs. First, based on the event sequences parsed by LogMaster, we take advantage of a prediction method based on Long Short-Term Memory Network (LSTM), and train the N-ary event sequence patterns in Many-to-One scenario; Second, we extract the failure rules from the minimal event sequence patterns, which identify the key events (failure signs) that correlate to system failures in the large-scale cluster; At last, we evaluate our experiments on the publicly available production Bluegene/L dataset, and obtain some interesting rules and correlations of failure events.

With the development of the city, the traffic crowd congested scene is increasing frequency. And the traffic crowd congested may bring disaster. It is important for city traffic management to recognize traffic crowd congested scene. However, the traffic crowd scene is dynamically and the visual scales are varied. Due to the multi-scale problem, it is hard to distinguish the congested traffic crowd scene. To solve the multiple scales problem in traffic crowd congested scene recognition, in this paper, a traffic crowd congested scene recognition method based on dilated convolution network is proposed, which combines the dilated convolution and VGG16 network for traffic crowd congested scene recognition. To verify the proposed method, the experiments are implemented on two crowd datasets including the CUHK Crowd dataset and Normal-abnormal Crowd dataset. And the experimental results are compared with three states of the art methods. The experimental results demonstrate that the performance of the proposed method is more effective in congested traffic crowd scene recognition. Compared with the three state of the art methods, the average accuracy value, and the average AUC values of the proposed method are improved by 15.87 $$\%$$ % and 11.58 $$\%$$ % respectively.

As the scales of cluster systems increase, failures become normal and have made reliability management be a major concern for system administrators. Failure prediction is a proactive measure through mining failure patterns and predicting when the systems will fail. In general, it is helpful to improve the accuracy of failure prediction by mining true failure patterns. And currently, the statistical and data mining driven methods are often used for mining failure patterns. However, since the overwhelming volumes and complicated interleaving of logs, the efficient and accurate failure pattern detection and automated failure prediction are still challenging.In this paper, we utilize the FP-Growth algorithm based on Spark (Spark-FPGrowth) to mine the correlations among different events, which can obtain the long-tail frequent event sequences effectively. Since the preprocessed event transactions are not suitable for using frequent pattern mining algorithms, we propose an adaptive sliding window division method based on event density with/without overlapping to construct event sequence transactions. At last, we analyze the log characteristics and predict failures for three large-scale production systems, and the evaluation results show that the average accuracy rates have higher accuracy and efficiency in CMRI-Hadoop, LANL-HPC and Bluegene/L logs respectively.

Federated learning is a new machine learning paradigm. The goal is to build a machine learning model from the data sets distributed on multiple devices–so-called an isolated data island–while keeping their data secure and private. Most existing federated learning benchmarks work manually splits commonly-used public datasets into partitions to simulate real-world isolated data island scenarios. Still, this simulation fails to capture real-world isolated data island’s intrinsic characteristics. This paper presents a federated learning (FL) benchmark suite named FLBench. FLBench contains three domains: medical, financial, and AIoT. By configuring various domains, FLBench is qualified to evaluate federated learning systems and algorithms’ essential aspects, like communication, scenario transformation, privacy-preserving, data distribution heterogeneity, and cooperation strategy. Hence, it becomes a promising platform for developing novel federated learning algorithms. Currently, FLBench is open-sourced and in fast-evolution. We package it as an automated deployment tool. The benchmark suite is available from https://www.benchcouncil.org/flbench.html .

In recent years, social media takes the advantages of fast spreading speed, wide range and low cost to become the main channel for people to obtain news, which also makes it to be a hotbed for the proliferation of fake news, exposing users and society to huge risks. Due to the fact that there is some true information in fake news, traditional text feature detection algorithms are more difficult to detect the fake news. Therefore, it is necessary to use knowledge as auxiliary information to help detection. We propose a fake news detection framework using knowledge vectors, which can adopt existing and reliable news as knowledge sources and reduce the dependence on expert verification. The framework consists of three parts: event triple extraction based on reliable content, fusion knowledge vector and fake news detector. The experimental results on the data set show that the framework can fuse part of the knowledge information and optimize the detection performance.

Extracting information as much and precise as possible from nondestructive measurements remains a challenge, especially when advanced test applications are emerging in electromagnetic encephalography and high throughput physical property characterization of materials genome chips. To solve the inversion problem, various soft algorithms have been developed such as finite element method, machine learning and artificial neural network, whose performance is limited by indirectly processing of intermediate layers in digital computers. This paper proposes a novel direct method of analog network entity with reconfigurable electrical circuit auto-processing (RECAP), which mainly consists of voltage controlled elements, measurement unit, and automatic feedback unit. During each inversion process, after the test results are input, the circuit network performs initialization, choosing topology, automatically tuning the property for each network element, and finally approaching a convergent solution to user request after some cycles of self-adjustment. Principles and advantages are introduced with several instances, showing high accuracy and stable convergence ability, as well as helping judge whether the topology is suitable for optimization. This method can not only invert purely loss components, but also invert circuits containing reactive components. Based on the verification from cases, it is also found that the inversion efficiency of RECAP is linearly dependent on the number of elements N, which is better than the usual mature inversion algorithms. Therefore, it is then concluded as a promising tool for high performance inversion.

We exhibit a detailed implementation on the natural language processing for natural language programming in python. We break the natural language text into active verbs and plural nouns to realize the sentence breaker and loop finder. Various examples of the implementation are presented. The realization of the natural language text into a computer programming does benefit in understanding the structure of the natural language processing and also the construction of the natural language programming.

This paper proposes a dynamic and scalable blockchain system framework for structuring a blockchain systems (BC). Traditionally a BC maintain multiple nodes with a smart-contract engine (SC) running on nodes, possibly with one or more Oracles Machines (OMs). However, many sophisticated applications require a much flexible yet still secure and scalable system architecture. The new framework LSO (Ledgers, Smart contracts, Oracles) with an inter-unit collaboration protocol that can be used for system registration, identification, communication, privacy protocols, and scalability management. The LSO framework is a dynamic and scalable framework because BCs, SCs, and OMs can be added without affecting the overall structure and without performance degradation. This framework support this by running a collection of cooperative Collaboration Layers (CLs) that acts like a DNS (Domain Name System) in Internet but this time to interconnect various BCs, SCs, and OMs. This LSO framework is a part of ChainNet initiative where BCs are used as building blocks in information and communication systems.

Although the advanced technology stack of blockchain leads to the prosperity of its applications, the underlying mechanism of blockchain interoperability remains stagnant. Targeting at the efficiency and verification problems of existing cross-blockchain mechanisms, we design a Cross-Blockchain Information Synchronization and Verification (CISV) mode. In CISV, Cross-chain Information Synchronization (CIS) uses the design of touch block to provide high synchronization rate within low network delay, while Cross-chain Information Verification (CIV) applies ECC cryptography algorithm and Secure Multi-party Computation (SMC) to guarantee the accuracy and privacy of cross-chain information. Experiments executed on three mainstream blockchains (Ethereum, Hyperledger Fabric, and EOS) show that CISV has a reliable performance in blockchain interoperability with strong expansibility.

With the rapid development of blockchain technology, the reliability and security of blockchain smart contracts is one of the most emerging issues of greatest interest for researchers. In this paper, the framework of formal symbolic process virtual machine, called FSPVM-EOS, is presented to certify the reliability and security of EOS-based smart contracts in Coq proof assistant. The fundamental theoretical concepts of FSPVM-EOS are abstract formal symbolic process virtual machine and execution-verification isomorphism. The current version of FSPVM-EOS is constructed on a formal virtual memory model with multiple-level table structure, a formal intermediate specification language, which is a large subset of the C ++ programming language based on generalized algebraic datatypes, and the corresponding formal definitional interpreter. This framework can automatically execute the smart contract programs of EOS blockchain and simultaneously symbolically verify their reliability and security properties using Hoare logic in Coq.

Scientific data sharing faces some dilemmas in process of practice. Blockchain technology has the characteristics of decentralization, openness, independence, security and anonymity, it provided a new solution to solve the dilemmas of scientific data sharing. So, exploring the application of blockchain technology in scientific data sharing is of great significance to expand the applications of blockchain and improve the effectiveness of scientific data sharing. This paper introduced the basic concepts of blockchain, summarized the practical dilemmas faced by scientific data sharing, analyzed the coupling between blockchain and scientific data sharing. Finally, we proposed a scientific data sharing framework which called AVEI based on blockchain, and analyzed the performance of AVEI from three perspective. AVEI covers three processes, they are user identity role authenticate process, data verify process and data exchange process, and we also introduced incentive system to enhance the enthusiasm of participating nodes. In theory, AVEI can solve practical dilemmas such as inconsistent standards, violation of privacy and security, insufficient motivation, etc., and it has a good performance in data quality, data security, and sharing effects.

In the traditional supply chain traceability model, a centralized management method is prevalent in supply chain traceability system. Problems such as falsification of production data, difficulty in information sharing, and opaque data have become more and more prominent. Blockchain technology has the characteristics of decentralization, non-tampering, openness, transparency and traceability, and can overcome many shortcomings of centralized traceability systems. Building supply chain traceability system based on blockchain technology provides an effective way solving the problems arousing in the supply chain safety domain. We propose SCT-CC, a supply chain traceability system based on cross-chain technology of blockchain. First, we use cross-chain technology to design a multi-chain architecture. Then, we design smart contracts for different agencies in the supply chain. Finally, we use Hyperledger fabric as a development framework and built an experimental environment using Docker technology. We also test the query interface and the write interface. Through the analysis of the test results, the system can meet the needs of actual applications.

As an important blockchain application, CBDC (Central Bank Digital Currency) has received significant worldwide attention as it can restructure financial market, affect national currency policies, and introduce new regulation policies and mechanisms. It is widely predicted that CBDC will introduce numerous digital currency competitions in various aspects of the global financial market, and winners will lead the next wave of digital currency market. This paper applies the game theory to study the competitions between different countries, in particular to analyze whether they should adopt the CBDC program. We propose two game-theoretic models for CBDC adoption, both analyzing whether to adopt the CBDC program via the Nash equilibrium. Both game-theoretic models draw the same conclusion that each country should adopt the CBDC program regardless of the choices of other counties. In other words, current currency leaders should adopt CBDC because it may lose the premier status, and other countries should adopt CBDC otherwise they risk of getting even further behind in the digital economy. According to our game-theoretic models, the current market leader who has 90% of market shares may lose about 19.2% shares if it is not the first mover.

Stability control system (SCS) plays a very important role in the operation of power systems, so it is correspondingly essential to ensure the reliability of the SCS during its experimental and verification process. However, there are numerous problems in the management of the paper form based experiment mode of SCS today, such as long circulation cycle caused by low work efficiency, insecure and diseconomy on storage. More importantly, it is difficult to query and review historical records quickly, and it may even lead to the loss of records for various reasons. In this paper, taking advantage of blockchain technology, all the information of each experimental stage of SCS, including the approval information, experimental environment information, experimental equipment information of each manufacturer, experimental test reports and conclusions, are packaged and distributed managed. And then, a blockchain based experimental management system framework with the unified data form and module interface is proposed, in which the whole process of process record can be traced and cannot be tampered with. The experimental management system can be applied in the whole process of the SCS, including the experiment scheme making, experiment environment constructing, experiment processing, result evaluating, and results storage, etc. The blockchain based mechanism and system framework is expect to take many advantage to experiment management of SCS.

With the development of economic globalization, a product usually involves different organizations from the origin to retail. These organizations form a supply chain to collaborate to provide products to end-users. The arrival of the digital age offers excellent convenience for the supply chain management. However, it is difficult for existing supply chain management systems to share credit, which leads to end-users’ concerns about the quality of products. The blockchain technology, characterized by tamper-proof, traceability, and distributed sharing, is expected to solve the it. Also, supply chain management also increasingly depends on the use of the Internet of things. However, the throughput limitation of current blockchain systems significantly constrains their applications to IoT data storage and management. In this paper, we propose a cross-chain solution based on IOTA and Fabric for supply chain data management. We design and implement a supply chain data management system based on a cross-chain gateway of blockchain. The system has three layers, including blockchain layer, business layer, and application layer, which provide essential logistics functions in supply chain management. Finally, we conduct experiments to validate the effectiveness of the implemented system.

In essay marking, manual grading will waste a lot of manpower and material resources, and the subjective judgment of marking teachers is easy to cause unfair phenomenon. Therefore, this paper proposes an automatic essay grading model combining multi-channel convolution and LSTM. The model adds a dense layer after the embedding layer, obtains the weight assignment of text through softmax function, then uses the multi-channel convolutional neural network to extract the text feature information of different granularities, and the extracted feature information is fused into the LSTM to model the text. The model is experimented on the ASAP composition data set. The experimental results show that the model proposed in this paper is 6% higher than the strong baseline model, and the automatic scoring effect is improved to a certain extent.

Rapid advent of the era of big data, deep integration of education and technology, and multidisciplinary nature of educational research are driving the application of knowledge graphs in educational research. We use CSSCI and CSCD journal articles as data sources and analyze the Knowledge Graph of education using the graph database Neo4j in this article. Result shows that the number of literatures in the research area show a clear upward trend and go from the stages of brewing. Highly cited literatures’ research topics, subject categories, and authors are widely distributed. From the perspective of core institutions and authors, multiple normal universities have strong competitiveness. Citespace, SPSS and Bicomb are the most commonly used data processing tools for scholars. In further study of the area of knowledge graph, we should fully combine with the cutting-edge computer technology, based on big data and artificial intelligence to improve the level and quality of education research.

Python has become a hot spot for people to learn, but beginners who lack the relevant expertise can’t find the right course among the numerous learning resources of python. Therefore, based on python courses on MOOC of Chinese universities, the paper uses AHP and data mining to evaluate courses from the curriculum, student activity, students comment on several aspects and filters out python courses for beginners.

The burgeon of online education platforms and online training platforms represented by MOOC has brought new development opportunities for educational innovation. In so many educational and training scenarios, there is a large amount of distributed users' learning data, which leads to the non-standardization of learning data collection and the non-uniformity of storage. In this paper, we design a process-oriented model for evaluating learning indicators. Then, we take KFCoding, which is an online training platform, as a case study to gather data for the learning process. Finally, we model the collected data and analyze the relationship between learning process data and learning outcomes.

In the era of Big Data, all countries put innovation at the core of national development. The Higher Education Department of the Ministry of Education of the People’s Republic of China calls for the comprehensive promotion of the construction of new liberal arts, with the aim of introducing the latest information technology into the traditional liberal arts teaching. As a normal university, East China Normal University is an important cradle to promote the ‘‘CS for ALL’’ talents. Therefore, in the computer reform of Shanghai colleges and universities, it takes the lead in the reform of computer basic teaching of New Liberal Arts. The fundamental goal of this reform is to cultivate students’ Data Thinking and improve their Data Literacy. Good results are achieved in the aspects of target achievement degree and result satisfaction degree.

Automatic identification of coreference and the establishment of corresponding model is an essential part in course syllabus construction especially for the comprehensive Universities. In this type of tasks, the primary objective is to reveal as much information as possible about the course entities according to their names. However, it remains a difficulty to most of the latest algorithms since the references to courses are commonly in line with the specifications of each University. Thus, it is important to link the course entities with similar identities to the same entity name due to the contextual information. To resolve this issue, we put forward a graph neural network (GNN)-based pipeline which was designed for the characteristics of syllabus. It could provide both the similarity between each pair of course names and the structure of an entire syllabus. In order to measure the performance of presented approach, the comparative experiments were conducted between the most advanced techniques and the presented algorithm. Experimental results demonstrate that the suggested approach can achieve superior performance over other techniques and could be a potentially useful tool for the exact identification of the entities in the educational scenarios.

In the digital era, general education in programming faces the challenge of cultivating more graduates who have superior digital competence. To achieve this goal, the educational environment should carry out digital transformation. In this paper, we explore an ecosystem of general education in programming, in which teachers provide guidance and undergraduates are stimulated to learn. With the support of various information technologies, it integrates various teaching elements and provides an interactive learning environment with effective feedback. The deep teaching based on the ecosystem can spur the deep learning of students and cultivate students’ digital thinking, programming thinking, data thinking and design thinking. In the teaching experiments, the ecosystem has achieved a satisfying effect, which is based on Shuishan online system.

Constructivism and Connectivism are two important schools of learning theory. The development of these two theories reflects the characteristics of the educational information era. This paper reviews the origin, development background and main viewpoints of constructivism and connectivism, and summarizes the characteristics of the two theories. In response to the changes in learning in the education information 2.0 era, on the basis of the original two learning theories, a new theory that is more suitable for learning in the Internet era is proposed: connected constructivism, and discusses the main points of the new theory.

This paper establishes stock index trading strategy by building stock index predicting model. First of all, this paper mainly reviews the application of machine learning in stock prediction, and constructs the L-GSRF stock index prediction model based on Lasso regression, grid search and random forest algorithm. After the input of textual and numerical data, the L-GSRF stock index prediction model greatly reduces the prediction error and improves the prediction accuracy in the photovoltaic (PV) stock index prediction, comparing with the traditional random forest and support vector machine (SVM) algorithm. In this paper, the trading strategy based on the prediction model has achieved a high annualized return. Finally, this study further clarifies the shortcomings of machine learning methods and future research directions.

The ongoing COVID-19 pandemic has led to not only the loss of enormous lives, but the dramatic impact on global financial markets. By considering 29 global systemically important banks from four regions (North America, Europe, China, Japan), we employ the proposed truncated D-vine dynamic mixed copulas model to investigate the evolution of the systemic risk of global banking system during the COVID-19 pandemic period. From empirical results, as a worldwide shock, the COVID-19 pandemic does have increased the systemic risk of the global banking system. Specifically, the systemic risk level of the global banking sector was moderate during the period when the COVID-19 pandemic burst only in China, and increased rapidly when the virus spread over the world, then cooling down when emergency actions were taken by countries. In addition, the systemic risk contribution of banks in most regions (like North America, Europe, and Japan) under the similar epidemic situation during the COVID-19 period, seem to be not impacted by the evolution of the panic (as well as the systemic risk level), while the systemic risk contribution of Chinese banks kept falling due to its opposite situation to others in this period.

In recent years, consumer credit has flourished in China. A credit factory is an important mode to speed up the loan application process. Order scheduling in credit factories belongs to the np-hard problem and it has great significance for credit factory efficiency. In this work, we formulate order scheduling in credit factories as a multi-agent reinforcement learning (MARL) task. In the proposed MARL algorithm, we explore a new reward mechanism, including reward calculation and reward assignment, which is suitable for this task. Moreover, we use a convolutional auto-encoder to generate multi-agent state. To avoid physical costs during MARL training, we establish a simulator, named Virtual Credit Factory, to pre-train the MARL algorithm. Through experiments in Virtual Credit Factory and an A/B test in a real application, we compare the performance of the proposed MARL approach and some classic heuristic approaches. In both cases, the results demonstrate that the MARL approach has better performance and strong robustness.

With the development of the digital economy, the price of Bitcoin, which is the most representative digital currency, has fluctuated dramatically. Recent works have explored the volatility of the Bitcoin price and made predictions using financial time series models such as ARIMA. However, for high-frequency Bitcoin data, the financial time series models have poor prediction performance as they often do not accommodate the inherent characteristics of digital currency, such as blockchain information. Some other works in this topic use deep learning models, e.g., artificial neural networks. However, the complex structure and time-consuming training process of these deep learning models often incur low prediction efficiency for rapidly changing Bitcoin price. In this regard, Broad Learning System (BLS) is a new neural network that avoids the complex structure of hidden nodes in deep learning models by adding enhancement nodes in the input layer to improve training efficiency while delivering relatively high accuracy. Therefore, this work applies the broad learning system to predict the Bitcoin price and optimizes the prediction model with a genetic algorithm. Due to the lack of fundamental factors for digital currency, the proposed prediction model considers the macroeconomic variables and the information of Bitcoin blockchain, which is the underlying technology of bitcoin, as inputs. According to the experimental results, the BLS-based prediction model optimized with the genetic algorithm achieved a better performance than other machine learning models.

As financial enterprises have moved their services to the internet, financial fraud detection has become an ever-growing problem causing severe economic losses for the financial industry. Recently, machine learning has gained significant attention to handle the financial fraud detection problem as a binary classification problem. While significant progress has been made, fraud detection is still a notable challenge due to two major reasons. First, fraudsters today are adaptive, inventive, and intelligent, making their fraud characteristics are too deep stealth to be detected by simple detection models. Second, labeled samples for training the detection models are usually very few as collecting large-scale training data needs a certain performance-time and is costly. To address the two problems, we propose a novel multi-source transfer learning approach with self-supervised domain distance learning for financial fraud detection problems. The core idea is to transfer relevant knowledge from multiple data-rich sources to the data-poor target task, e.g., learning fraud patterns from several other related mature loan products to improve the fraud detection in a cold-start loan product. Specifically, since the feature distribution discrepancy across domains may cause useless or even negative knowledge transfer, we propose self-supervised domain distance learning under the Wasserstein metric to measure the domain relevance/relationships between target and source tasks. The learned Wasserstein distance helps in selectively transferring most relevant knowledge from source domains to target domains. Thus it reduces the risk of negative transfer as well as maximizes the multi-source positive transfer. We conduct extensive experiments under multi-source few-shot learning settings on real financial fraud detection dataset. Experimental analysis shows that the inter-domain relationships learned by our domain distance learning model align well with the facts and the results demonstrate that our multi-source transfer learning approach achieves significant improvements over the state-of-the-art transfer learning approaches.