Abstract:While keeping frequent application updates, Android application developers need to detect Android runtime permission (ARP) bugs as quickly as possible. Android applications cannot effectively be tested for permission-related behaviors with automated testing tools since they are rarely designed for ARP bugs. This study proposes a state transition graph guided testing approach for detecting ARP bugs in Android applications. First, it analyzes the APK file of the application under test for permission misuse, instruments the APIs that may cause ARP bugs in the APK file, and re-signs the APK file. Then, it installs the APK file and dynamically explores the application to generate its state transition graph (STG). Finally, it detects ARP bugs quickly by automated testing with the guidance of STG. To evaluate the effectiveness of the approach, the study implements a prototype tool RPBDroid and conducts comparative experiments with the ARP bug detection tools SetDroid, PermDroid, and the automated testing tool APE. The experimental results show that RPBDroid successfully detects 15 ARP bugs out of 17 applications, which detects 14, 12, and 14 more ARP bugs than APE, SetDroid, and PermDroid respectively. In addition, RPBDroid reduces the average time required to detect ARP bugs by 86.42%, 86.72%, and 86.70% in comparison with SetDroid, PermDroid, and APE.

Abstract:A function-as-a-service (FaaS) workflow, composed of multiple function services, can realize a complex business application by orchestrating and controlling the function services. The current FaaS workflow execution systems achieve data transfer among function services mainly based on centralized data storages, resulting in heavy data transmission overhead and affecting application performance significantly. In the cases of high concurrency, frequent data transmission will also cause serious contention for network bandwidth resources, resulting in application performance degradation. To address the above problems, this study analyzes the fine-grained data dependency between function services and proposes a critical path-based FaaS workflow deployment optimization method. In addition, the study designs a dependency-sensitive data access and management mechanism to effectively reduce the data transmission between function services, thereby reducing the data transmission latency and end-to-end execution latency of FaaS workflow applications. The study implements a FaaS workflow system, FineFlow, and conducts experiments based on five real-world FaaS workflow applications. The experimental results show that FineFlow can effectively reduce the data transmission latency (the highest reduction and the average reduction are 74.6% and 63.8%, respectively) compared with the FaaS workflow platform with the centralized data storing-based function interaction mechanism. On average, FineFlow reduces the latency of the end-to-end FaaS workflow executions by 19.6%. In particular, for the FaaS workflow application with fine-grained data dependencies, FineFlow can further reduce its data transmission latency and the end-to-end execution latency by 28.4% and 13.8% respectively compared with the state-of-the-art work. In addition, FineFlow can effectively alleviate the impact of network bandwidth fluctuations on application performance by reducing cross-node data transmission, improving the robustness of application performance influenced by the network bandwidth changes.

Abstract:The prosperity of open-source software has spurred robust growth in the software industry and has also facilitated the formation of a supply chain development model based on open-source software. Essentially, the open-source software supply chain is a complex topology network, composed of key elements of the open-source ecosystem and their interrelations. Its globalized product advantages contribute to enhancing the development efficiency of the software industry. However, the open-source software supply chain also has characteristics such as intricate dependencies, widespread propagation, and an expanded attack surface, introducing new security risks. Although existing security management based on vulnerabilities and threat intelligence can achieve early warnings and proactive defense, the efficiency of vulnerability handling is severely affected due to delays in obtaining vulnerability threat information, and the lack of attack techniques and mitigation measures. Addressing these issues, a vulnerability threat intelligence sensing method for the open-source software supply chain is designed and implemented, which includes two parts: 1) Construction of the cyber threat intelligence (CTI) knowledge graph. In the process of constructing it, relevant technologies are utilized to achieve real-time analysis and processing of security intelligence. Particularly, the SecERNIE model and the software package naming matrix are introduced to address the challenges of vulnerability threat correlation mining and open-source software alias issues, respectively. 2) Vulnerability risk information push,based on the software package naming matrix, software package filtering rules are established to enable real-time filtering and pushing of vulnerabilities in open-source systems. This study validates the effectiveness and applicability of the proposed method through experiments. Results show that, compared to traditional vulnerability platforms like NVD, the proposed method advances the sensing time by an average of 90.03 days. The coverage rate of operating system software increases by 74.37%, and using the SecERNIE model, the relationships between 63492 CVE vulnerabilities and attack technique entities are mapped. Specifically, for the openEuler operating system, the traceable system software coverage rate reaches 92.76%, with 6239 security vulnerabilities detected. This study also identifies 891 vulnerability-attack correlations in openEuler, obtaining corresponding solutions that serve as a reference for vulnerability handling. Two typical attack scenarios in a real attack environment are verified, demonstrating the efficacy of the proposed method in vulnerability threat perception.

Abstract:Multi-modal affective computing is a fundamental and important research task in the field of affective computing, using multi-modal signals to understand the sentiment of user-generated video. Although existing multi-modal affective computing approaches have achieved good performance on benchmark datasets, they generally ignore the problem of modal reliability bias in multi-modal affective computing tasks, whether in designing complex fusion strategies or learning modal representations. This study believes that compared to text, acoustic and visual modalities often express sentiment more realistically. Therefore, voice and vision have high reliability, while text has low reliability in affective computing tasks. However, existing learning abilities of different modality feature extraction tools are different, resulting in a stronger ability to represent textual modality than acoustic and visual modalities (e.g., GPT3 and ResNet). This further exacerbates the problem of modal reliability bias, which is unfavorable for high-precision sentiment judgment. To mitigate the bias caused by modal reliability, this study proposes a model-agnostic multi-modal reliability-aware affective computing approach (MRA) based on cumulative learning. MRA captures the modal reliability bias by designing a single textual-modality branch and gradually shifting the focus from sentiments expressed in low-reliability textual modality to high-reliability acoustic and visual modalities during the model learning process. Thus, MRA effectively alleviates inaccurate sentiment predictions caused by low-reliability textual modality. Multiple comparative experiments conducted on multiple benchmark datasets demonstrate that the proposed approach MRA can effectively highlight the importance of high-reliability acoustic and visual modalities and mitigate the bias of low-reliability textual modality. Additionally, the model-agnostic approach significantly improves the performance of multi-modal affective computing, indicating its effectiveness and generality in multi-modal affective computing tasks.

Abstract:Event detection (ED) aims to detect event triggers in unstructured text and classify them into pre-defined event types, which can be applied to knowledge graph construction, public opinion monitoring, and so on. However, the data sparsity and imbalance severely impair the system’s performance and usability. Most existing methods cannot well address these issues. This is due to that during detection, they regard events of different types as independent and identify or classify them through classifiers or space-distance similarity. Some work considers the correlation between event elements under a broader category and employs multi-task learning for mutual enhancement; they overlook the shared properties of triggers with different event types. Research related to modeling event connections requires designing lots of rules and data annotation, which leads to limited applicability and weak generalizability. Therefore, this study proposes an event-detection method based on meta-attributes. It aims to learn the shared intrinsic information contained in samples across different event types, including (1) extracting type-agnostic semantics of triggers through semantic mapping from the representations of special symbols; (2) concatenating the semantic representations of triggers and samples in each event type as well as the label embedding, inputting them into a trainable similarity measurement layer, thereby modeling a public similarity metric related to triggers and event categories. By combining these representations into a measuring layer, the proposed method mitigates the effects of data sparsity and imbalance. Additionally, the full fusion model is constructed by integrating the type-agnostic semantic into the classification method. Experiments on ACE2005 and MAVEN datasets under various degrees of sparsity and imbalance, verify the effectiveness of the proposed method and build the connection between conventional and few-shot settings.

Abstract:Recently, deep learning has received increasing attention from researchers due to its excellent performance in various scenarios, but these methods often rely on the independent and identically distribution assumption. Domain adaptation is a problem proposed to mitigate the impact of distribution offset, which uses labeled source domain data and unlabeled target domain data to achieve better performance on target data. Existing methods are devised for static data, while the methods for time series data need to capture the dependencies between variables. Although these methods use feature extractors for time series data, such as recurrent neural networks, to learn the dependencies between variables, they often extract redundant information. This information is easily entangled with semantic information, affecting the model performance. To solve these problems, this study proposes a path-signature-based time-series domain adaptation (PSDA). On the one hand, this method uses path signature transformation to capture sparse dependencies between variables and eliminate redundant correlations while preserving semantic dependencies, thereby facilitating the extraction of discriminative features from temporal data. On the other hand, the invariant dependency relationships are preserved by constraining the consistency of dependency relationships among different domains, and the changing dependency relationships between domains are excluded, which is conducive to extracting generalized features from temporal data. Based on the above methods, the study further proposes a distance metric and generalized boundary theory and obtains the best experimental results on multiple time series domain adaptation standard datasets.

Abstract:In recent years, multi-agent reinforcement learning methods have demonstrated excellent decision-making capabilities and broad application prospects in successful cases such as AlphaStar, AlphaDogFight, and AlphaMosaic. In the multi-agent decision-making system in a real-world environments, the decision-making space of its task is often a parameterized action space with both discrete and continuous action variables. The complex structure of this type of action space makes traditional multi-agent reinforcement learning algorithms no longer applicable. Therefore, researching for parameterized action spaces holds important significance in real-world application. This study proposes a factored multi-agent centralised policy gradients algorithm for parameterized action space in multi-agent settings. By utilizing the factored centralised policy gradient algorithm, effective coordination among multi-agent is ensured. After that, the output of the dual-headed policy in the parameterized deep deterministic policy gradient algorithm is employed to achieve effective coupling in the parameterized action space. Experimental results under different parameter settings in the hybrid predator-prey scenario show that the algorithm has good performance on classic multi-agent parameterized action space collaboration tasks. Additionally, the algorithm’s effectiveness and feasibility is validated in a multi-cruise-missile collaborative penetration tasks with complex and high dynamic properties.

Abstract:As merchant review websites develop rapidly, the efficiency improvement brought by recommender systems makes rating prediction one of the emerging research tasks in recent years. Existing rating prediction methods are usually limited to collaborative filtering algorithms and various types of neural network models, and do not take full advantage of the rich semantic knowledge learned in advance by the current pre-trained models. To address this problem, this study proposes a personalized rating prediction method based on pre-trained language models. The method analyzes the historical reviews of users and merchants to provide users with rating predictions as a reference before consumption. It first designs a pre-training task for the model to learn to capture key information in the text. Next, the review text is processed by a fine-grained sentiment analysis method to obtain aspect terms in the review text. Subsequently, the method designs an aspect term embedding layer to incorporate the aforementioned external domain knowledge into the model. Finally, it utilizes an information fusion strategy based on the attention mechanism to fuse the global and local semantic information of the input text. The experimental results show that the method achieves significant improvement in both automatic evaluation metrics compared to the benchmark models.

Abstract:The label distribution in the real world often shows the long-tail effect, where a small number of categories account for the vast majority of samples. The temporal action detection problem is no exception. The existing temporal action detection methods often focus on the head categories with a large number of samples, while neglecting the few-sample categories. This study systematically defines the long-tail temporal action detection problem and proposes a weighted class-rebalancing self-training method (WCReST) based on a semi-supervised learning framework. WCReST makes full use of the large-scale unlabeled data that exists in the real world to rebalance the label distribution in the training samples to improve the model’s fit for the tail categories. Additionally, a pseudo-label loss weighting method is proposed for the temporal action detection task to enhance the stability of model training. Experiments are conducted on the THUMOS14 and HACS Segments datasets, using video samples from the THUMOS15 and ActivityNet1.3 datasets to form corresponding unlabeled datasets. In addition, the Dance dataset is collected to meet the application requirements of video review, which includes 35 action categories, 6632 labeled videos, and 13264 unlabeled videos, preserving the significant long-tail effect in data distribution. A variety of baseline models are used to conduct experiments on the THUMOS14, HACS Segments, and Dance datasets. The results demonstrate that the proposed WCReST can improve the model’s detection performance on tail action categories and can be applied to different baseline temporal action detection models to enhance their performance.

Abstract:As a fine-grained sentiment analysis method, aspect-based sentiment analysis is playing an increasingly important role in many application scenarios. However, with the ubiquity of social media and online reviews, cross-domain aspect-based sentiment analysis faces two major challenges: insufficient labeled data in the target domain and textual distribution differences between the source and target domains. Currently, many data augmentation methods attempt to alleviate these issues, yet the target domain text generated by these methods often suffers from shortcomings such as lack of fluency, limited diversity of generated data, and convergent source domain. To address these issues, this study proposes a method for cross-domain aspect-based sentiment analysis based on data augmentation from a large language model (LLM). This method leverages the rich language knowledge of large language models to construct appropriate prompts for the cross-domain aspect-based sentiment analysis task. It mines similar texts between the target domain and the source domain and uses context learning to guide the LLM to generate labeled text data in the target domain with domain-associated keywords. This approach addresses the lack of data in the target domain and the domain-specificity problem, effectively improving the accuracy and robustness of cross-domain sentiment analysis. Experiments on multiple real datasets show that the proposed method can effectively enhance the performance of the baseline model in cross-domain aspect-based sentiment analysis.

Abstract:Elephant flow identification is a fundamental task in network measurements. Currently, the mainstream methods generally employ sketch data structure Sketch to quickly count network traffic and efficiently find elephant flows. However, the rapid influx of numerous packets will significantly decrease the identification accuracy of elephant flows under network traffic jitters. To this end, this study proposes an elastic identification method for elephant flows supporting network traffic jitters, which is named RobustSketch. This method first designs a stretchable mice flow filter based on the cyclic Sketch chain, and adaptively increases and reduces the number of Sketch in real-time packet arrival rates. As a result, it always completely records all arrived packets within the current period to ensure accurate mice flow filtering even under network traffic jitters. Subsequently, this study designs a scalable elephant flow record table based on dynamic segmented hashing, which adaptively increases and reduces segments according to the number of candidate elephant flows filtered out by the mice flow filter. Finally, this can fully record all candidate elephant flows and keep high storage space utilization. Furthermore, the error bounds of the proposed mice flow filter and elephant flow recording table are provided by theoretical analysis. Finally, experimental evaluation is conducted on the proposed elephant flow identification method RobustSketch with real network traffic samples. Experimental results indicate that the identification accuracy of elephant flows of the proposed method is significantly higher than that of the existing methods, and can stably keep high accuracy of over 99% even under network traffic jitters. Meanwhile, its average relative error is reduced by more than 86%, which enhances the accuracy and robustness of elephant flow identification.

Abstract:Internet service providers employ routing protection algorithms to meet real-time, low-latency, and high-availability application needs. However, existing routing protection algorithms have the following three problems. (1) The failure protection ratio is generally low under the premise of not changing the traditional routing protocol forwarding mechanism. (2) The traditional routing protocol forwarding mechanism should be changed to pursue a high failure protection ratio, which is difficult to deploy in practice. (3) The optimal next hop and backup next hop cannot be utilized simultaneously, which causes poor network load balancing capability. For the three problems, this study proposes a routing protection algorithm based on the shortest path serialization graph, which does not need to change the forwarding mechanism, supports incremental deployment and adopts both optimal next hop and backup next hop without routing loops, with a high failure protection ratio. The proposed algorithm mainly includes the following two steps. (1) A sequence number for each node is calculated, and the shortest path sequencing graph is generated. (2) The shortest path serialization graph is generated based on the node sequence number and reverse order search rules, and the next hop set between node pairs is calculated according to the backup next hop calculation rules. Tests on real and simulated network topologies show that the proposed scheme has significant advantages over other routing protection schemes in the average number of backup next hops, failure protection ratio, and path stretch.

Abstract:Identity-based matchmaking encryption is a new cryptographic primitive that allows both the receiver and the sender to specify each other’s identity and communicate with each other only when the identities match. Meanwhile, it provides a non-interactive secret handshake protocol to get rid of real-time interaction and further improve participant privacy. This study proposes an identity-based matchmaking encryption (IB-ME) scheme in prime-order groups under symmetric external Diffie-Hellman (SXDH) assumption under the standard model. Realizing short parameters and reducing the matchmaking times during decryption are the most efficient identity-based matchmaking encryption scheme. Additionally, this study also puts forward the first inner product with equality matchmaking encryption (IPE-ME) scheme under the SXDH assumption in the standard model. Technically, it first constructs two schemes in composite-order groups, then simulates them with dual pairing vector space (DPVS) into prime-order groups, and further reduces the parameter size by decreasing the required dimension of dual basis. Finally, for the proposed IPE-ME scheme, this study replaces the equality policy in the first layer of an IB-ME scheme with inner-product policy.

Abstract:As mobile data is growing everyday, how to predicate the wireless traffic accurately is crucial for the efficient and sensible allocation of communication and network resources. However, most existing prediction methods use a centralized training architecture, which involves large-scale traffic data transmission, leading to security issues such as user privacy leakage. Federated learning can train a global model with local data storage, which protects users’ privacy and effectively reduces the burden of frequent data transmission. However, in wireless traffic prediction, the amount of data from the single base station is limited, and the traffic patterns vary among different base stations, making it difficult to capture the traffic patterns and resulting in poor generalization of the global model. In addition, traditional federated learning methods employ averaging in model aggregation, ignoring the differences in guest contributions, which further leads to the degradation of the global model performance. To address the above issues, this study proposes an attention-based “intra-cluster average, inter-cluster attention” federated wireless traffic prediction model. The model first clusters base stations based on their traffic data to better capture the traffic variation characteristics of base stations with similar traffic patterns. At the same time, a warm-up model is designed to alleviate data heterogeneity by a small amount of base station data to improve the generalization ability of the global model. The study introduces the attention mechanism in the aggregation stage to quantify the contributions of different objects to the global model and incorporates the warm-up model in the model iteration process to improve the prediction accuracy of the model. Extensive experiments are conducted on two real-world datasets (Milano and Trento), and the results show that the DualICA outperforms all baseline methods. The mean absolute error performance gain over the state-of-the-art method is up to 10.1% and 9.6% on the two datasets, respectively.

Abstract:The network traffic measurement technology of programmable switches is capable of handling high-speed network traffic and offers significant advantages in terms of flexibility and real-time processing. However, due to the necessity of configuring the internal logic of switches using the complex P4 programming language, the deployment of measurement tasks becomes intricate and error-prone. Furthermore, measurement accuracy is often constrained by the available measurement resources within the switch of measurement tasks. This study proposes a detailed exploration of intent-based networking and network traffic measurement technology, introducing an intent-driven network traffic distributed measurement method. Firstly, an intent representation format based on measurement intent primitives is designed, and an intent compiler is developed to translate abstract intent representations into executable P4 code. Secondly, a network traffic distributed measurement approach is introduced, utilizing the resources of multiple switches to collaboratively complete a measurement task in a distributed manner. The dynamic allocation of measurement resources and counter-configuration algorithms are exemplified with heavy-hitter measurements. Finally, experimental results demonstrate the feasibility and certain advantages of the proposed method.

Abstract:Constructing post-quantum key encapsulation mechanisms based on Lattice (especially NTRU Lattice) is one of the popular research fields in Lattice-based cryptography. Commonly, most Lattice-based cryptographic schemes are constructed over cyclotomic rings, which, however, are vulnerable to some attacks due to their abundant algebraic structures. An optional and more secure underlying algebraic structure is the large-Galois-group prime-degree prime-ideal number field. NTRU-Prime is an excellent NTRU-based key encapsulation mechanism over the large-Galois-group prime-degree prime-ideal number field and has been widely adopted as the default in the OpenSSH standard. This study aims to construct a key encapsulation mechanism over the same algebraic structure but with better performance than NTRU-Prime. Firstly, this work studies the security risks of cyclotomic rings, especially the attacks on quadratic power cyclotomic rings, and demonstrates the security advantages of a large-Galois-group prime-degree prime-ideal number field in resisting these attacks. Next, an NTRU-based key encapsulation mechanism named CNTR-Prime over a large-Galois-group prime-degree prime-ideal number field is proposed, along with the corresponding detailed analysis and parameter sets. Then, a pseudo-Mersenne incomplete number theoretic transform (NTT) is provided, which can compute polynomial multiplication efficiently over a large-Galois-group prime-degree prime-ideal number field. In addition, an improved pseudo-Mersenne modular reduction algorithm is proposed, which is utilized in pseudo-Mersenne incomplete NTT. It is faster than Barrett reduction by 2.6% in software implementation and is 2 to 6 times faster than both Montgomery reduction and Barrett reduction in hardware implementation. Finally, a C-language implementation of CNTR-Prime is presented. When compared to SNTRU-Prime, CNTR-Prime has advantages in security, bandwidth, and implementation efficiency. For example, CNTR-Prime-761 has an 8.3% smaller ciphertext size, and its security is strengthened by 19 bits for both classical and quantum security. CNTR-Prime-761 is faster in key generation, encapsulation, and decapsulation algorithms by 25.3×, 10.8×, and 2.0×, respectively. The classical and quantum security of CNTR-Prime-653 is already comparable to that of SNTRU-Prime-761, with a 13.8% reduction in bandwidth, and it is faster in key generation, encapsulation, and decapsulation by 33.9×, 12.6×, and 2.3×, respectively. This study provides an important reference for subsequent research, analysis, and optimization of similar Lattice-based cryptographic schemes.

Abstract:With the rapid development of quantum computing, especially the optimization and progress of the Shor quantum algorithm and its variants, the current classical public-key cryptography based on factoring large integers and discrete logarithm problems is facing serious security threats. To cope with quantum attacks, post-quantum cryptography has been proposed, among which lattice-based cryptography is commonly viewed as the most promising one due to its outstanding performance in security, bandwidth, and efficiency. Most of the existing lattice-based post-quantum cryptographic schemes use cyclotomic rings, especially power-of-two cyclotomic rings, as their underlying algebraic structures. However, targeted attacks against cyclotomic rings have been proposed, exploiting subfields, small Galois groups, and ring homomorphisms in these rings. This study uses the large-Galois-group prime-degree prime-ideal field as the new underlying algebraic structure, which has characteristics of high security, prime order, large Galois group, and inert modulus.First, this study proposes a post-quantum digital signature scheme based on the large-Galois-group prime-degree prime-ideal field, which is named Dilithium-Prime, and the recommended parameter sets are provided. Next, considering that the traditional number theory transform (NTT) algorithm cannot be used to multiply polynomials efficiently in the large-Galois-group prime-degree prime-ideal field, this study designs efficient polynomial multiplication strategies for Dilithium-Prime, including NTT for the large-Galois-group prime-degree prime-ideal field and small polynomial multiplication. Finally, this study provides a portable C language implementation of Dilithium-Prime, along with the implementation details and constant-time implementation skills, and compares Dilithium-Prime with other lattice-based digital signature schemes. The experimental results show that the public key size, secret key size, and signature size of Dilithium-Prime are reduced by 1.8%, 10.2%, and 1.8%, respectively, compared to CRYSTALS-Dilithium. The efficiency of the signature algorithm is improved by 11.9%, and the key generation algorithm and the verification algorithm are 2.0× and 2.5× slower than those of CRYSTALS-Dilithium, respectively. However, Dilithium-Prime can withstand the cryptographic attack against cyclotomic rings, which is exactly what CRYSTALS Dilithium lacks. Compared to NCC-Sign, Dilithium-Prime’s key generation algorithm, signature algorithm, and verification algorithm are 4.2×, 35.3×, and 7.2× faster, respectively, than those of NCC-Sign under the same security level and bandwidth.

Abstract:Log-structured merge-tree (LSM-tree) based key-value storage is widely used in many applications due to its excellent read and write performance. Most existing LSM-trees utilize a multi-level structure to store data. Although the multi-level data structure can serve moderately write-intensive applications well, this structure is not well suited for highly write-intensive applications. This is because storing data in multi-levels introduces the write amplification problem, where new data insertion triggers the reorganization of a large portion of the data already stored in multiple levels. This huge (and sometimes frequent) data reorganization is expensive and degrades write performance in many highly write-intensive applications. In addition, the multi-level structure does not provide consistently excellent read performance for hot data. This is because the multi-level structure cannot optimize the read operation of hot data by merging overlapping ranges in a timely manner. To address the above two challenges, this study proposes LazyStore, a novel single-level LSM-tree based on a hybrid storage architecture. LazyStore solves the write amplification problem by storing data in a single logical level instead of multiple logical levels. As a result, expensive multi-level data reorganization is largely eliminated. To further improve write performance, LazyStore distributes data at the logical level to multiple storage devices, such as DRAM, NVM, and SSD, based on the capacity and read/write performance of each storage device. Furthermore, LazyStore introduces real-time merge operations to improve the read performance of hot data ranges. Experiments show that LazyStore improves write performance by 3 times and reduces write amplification by nearly 4 times compared to other multi-level LSM-trees. For hot range reads, LazyStore’s real-time data merge optimization can reduce the latency of range query processing by a factor of two.

Abstract:Star-JOIN queries based on local differential privacy (LDP) have attracted a lot of attention from researchers in recent years. Existing Star-JOIN queries based on the OLH mechanism and hierarchical tree structures face issues such as privacy leakage risks at the root node and the lack of guidance on selecting an appropriate τ value for the τ-truncation mechanism. To remedy the shortcomings of the existing algorithms, this study proposes an effective Star-JOIN query algorithm, longitudinal path random response for join (LPRR-JOIN), to satisfy the requirements of LDP. In the LPRR-JOIN algorithm, full advantage is taken of the longitudinal path structure of the hierarchical tree and the GRR mechanism to propose an algorithm called LPRR to perturb users’ tuples. This algorithm utilizes the combinations of nodes along the longitudinal paths of all attributes as the perturbation domain. In the LPRR-JOIN algorithm, tuples are mapped by each user to corresponding node combinations, followed by local perturbation of the mapped tuples using the GRR mechanism. To guard against frequency attacks on the fact table, the algorithm permits users to locally truncate the count of their tuples based on a threshold τ, where tuples are deleted if their count exceeds τ and supplemented if it falls below τ. Two solutions are proposed within LPRR-JOIN to compute the optimal τ value. The first is to solve the optimization equation over bias caused by τ-truncation and perturbation variance due to LPRR. The second is to obtain the distribution of τ under the constraints of LDP and compute the median value from the distribution. The LPRR-JOIN algorithm employs an overall error function constructed from the bias and perturbation variance resulting from τ-truncation to derive an optimal τ value through the optimization of the error objective function. Additionally, by integrating a user grouping strategy, the algorithm ascertains the overall distribution of τ values and identifies a suitable τ value using the median. When compared with current algorithms across three diverse multi-relational datasets, experimental outcomes demonstrate the superiority of the LPRR-JOIN algorithm in query response performance.

Abstract:The density-based spatial clustering of applications with noise (DBSCAN) algorithm is one of the clustering analysis methods in the field of data mining. It has a strong capability of discovering complex relationships between objects and is insensitive to noise data. However, existing DBSCAN methods only support the clustering of unimodal objects, struggling with applications involving multi-model data. With the rapid development of information technology, data has become increasingly diverse in real-life applications and contains a huge variety of models, such as text, images, geographical coordinates, and data features. Thus, existing clustering methods fail to effectively model complex multi-model data and cannot support efficient multi-model data clustering. To address these issues, in this study, a density-based clustering algorithm in multi-metric spaces is proposed. Firstly, to characterize the complex relationships within multi-model data, this study uses a multi-metric space to quantify the similarity between objects and employs aggregated multi-metric graph (AMG) to model multi-model data. Next, this study employs differential distances to balance the graph structure and leverages a best-first search strategy combined with pruning techniques to achieve efficient multi-model data clustering. The experimental evaluation on real and synthetic datasets, using various experimental settings, demonstrates that the proposed method achieves at least one order of magnitude improvement in efficiency with high clustering accuracy, and exhibits good scalability.

Abstract:As mobile devices are widely used, the performance of their graphics processors has increasingly improved. To meet users’ continuous pursuit of excellent experience, the screen resolution and refresh rate of mobile devices are constantly increasing every year. At the same time, the programmable shading pipeline in mobile games is becoming more complex, which leads to game applications becoming the main source of power consumption for mobile devices. This work studies the rendering pipeline in mobile games and proposes a motion-aware rendering frame rate adjustment method to ensure rendering quality in power-saving mode. Unlike previous prediction models that only consider rendering errors of historical frames, this method builds a nonlinear model between camera pose and inter-frame rendering error and predicts error based on the new frame’s camera pose, thus achieving more accurate frame rate adjustment strategies. In addition, the method also includes a lightweight scene recognition module that can adjust the error threshold according to the specific scene where the player is located, thereby adopting different degrees of frame rate adjustment strategies. Quantitatively compared with the prediction model that only considers historical frame errors, the proposed model improves the prediction accuracy on game frame sequences by more than 30%. At the same time, in the qualitative comparison of user experiments, under the same frame-skipping ratio, the proposed algorithm can achieve higher rendering quality and better user experience. The algorithm integrates historical frame errors and camera information to predict more accurate future frame errors. It also combines prediction and scene recognition results to achieve better dynamic frame rate adjustment performance.

Abstract:Multi-view data depicts objects from different perspectives, with features in different views exhibiting correlations, complementary, and diverse information. Therefore, it is crucial to make full use of this information for the processing of multi-view data. However, the processing and analysis of multi-view data will be difficult due to the inherent challenges of dealing with a vast number of features and the presence of noise features in multi-view data. Unsupervised multi-view feature selection, emerging as a critical component in multi-view data learning, efficiently learns more accurate and compact representations from the original high-dimensional multi-view data without relying on label information to remarkably improve the performance of data analysis. This study reviews and categorizes these models based on the similarities and differences in the working mechanisms of existing unsupervised multi-view feature selection models, while also detailing their limitations. Furthermore, this study points out promising future research directions in the field of unsupervised feature selection.

Abstract:With the application of artificial intelligence (AI) and end-to-end recognition methods in handwritten mathematical expression recognition, there has been a significant improvement in recognition accuracy. However, in contrast to tests on public datasets, real-world applications involving human input introduce more uncertain factors into recognition algorithms in practice. Factors such as personalized stroke information, ambiguous handwritten characters, and uncertain formula structures can significantly impact the performance of the recognition method. To address these challenges, HchMER, a hybrid human-machine intelligence method for handwritten mathematical expression recognition, is proposed. HchMER combines handwritten mathematical formula recognition algorithms, knowledge bases, and user feedback to enhance the machine's comprehension of user-input mathematical expressions, thereby improving the editing speed and accuracy of handwritten mathematical expressions. To assess the effectiveness of HchMER, it is compared with MyScript Math Recognition (MyScript) and a mature commercial product named “Microsoft Ink Equation” (InkEquation). Results show that HchMER outperformed MyScript and InkEquation in accuracy by 23.2% and 26.51%, respectively. In terms of average completion time, HchMER exceeded MyScript by 44.46% (9.6 s) but fell short of InkEquation by 11.48% (4.05 s). Furthermore, participants affirm HchMER in a questionnaire survey and semi-structured interviews.