Unsupervised learning is one of the fundamental machine learning methods. Clustering is a vital unsupervised learning task and can significantly contribute to the detection of hidden structures in unknown datasets. Clusterability is an important concept due to the fact that it can theoretically portray the extent to which a clustering algorithm can recover a benchmark clustering, with the absence of excessive experimental validations. Moreover, conventional batch-mode-clustering-oriented clusterability analysis should be extended to the incremental setting when the clustering algorithm is required to handle stream data. However, such clusterability analysis is facing two barriers. First, the incremental clustering algorithm proceeds in a step-wise manner and can merely access the newly arrived data of the current step. This extremely fragmentary view of the entire input data stream inevitably results in a biased perception of the underlying benchmark clustering. Second, incremental clustering is conventionally applied to real-time or massive-data scenarios. Such application scenarios typically require the computational power of mainstream SIMD (Single Instruction Multiple Data) hardware accelerators. However, strong data dependency inherently exists between two successive steps of an incremental clustering algorithm, which dramatically impairs data parallelism. In view of these constraints, we propose our roadmap to theoretically analyze and ensure the clusterability under an incremental setting in terms of a general clusterability metric: niceness (higher intra-cluster similarity than inter-cluster similarity). In our work, a nice-k clustering (a clustering that has k clusters and satisfies the niceness metric) is supposed to exist in the input data stream. Meanwhile, the input data stream is supposed to be divided into a series of micro-clusters, and the micro-clusters are incrementally clustered into clusters. In addition, we rely on an assumption (homogeneity assumption) that every micro-cluster merely contains homogenous data. First, we point out that a vital reason for the induction of heterogeneous clusters is the lack of representative micro-clusters. We propose Theorem 1 to iteratively identify a set of 2[Formula: see text] representative micro-clusters that can cover all k benchmark clusters. Therefore, we can trade the number of clusters for homogeneity and thus assure clusterability. Second, we demonstrate that evolution in the granularity of a micro-cluster can prompt SIMD-parallelism more than in the granularity of a single data point. Consequently, the clusterability-assured method of Theorem 1 is furthermore parallel-friendly. In all, we depict a roadmap to assure clusterability under both incremental and SIMD-friendly constraints.

With low storage cost and high retrieval efficiency, hashing techniques are widely used for multi-media retrieval, which has already become the present research focus. Currently, cross-modal hashing commonly employs graph-based loss to construct pair-wise semantic relations between training samples for model optimization. However, limited by the graph-based strategy, each edge in the graph only connects two samples, which only represent a bundle of pair-wise relationships. Besides, the edges in the graph are calculated by self-attention or feature distance, only considering pair-wise relations of heterogeneous samples and ignoring the class relations. In this paper, by hypergraph modeling the semantic tuples, a novel Deep Semantic Tuplet-based Hashing by Hypergraph Modeling (DSTH) approach is proposed to leverage the multilateral semantic relations, which could guide the model to learn class-discriminative semantic binary embedding. In more detail, based on the characteristic distribution, semantic tuples are constructed for each class in one mini-batch, which represents the multilateral semantic relationships between multiple samples and multiple classes. By considering semantic tuples as hyperedges to represent multilateral semantic relations, hypergraph modeling is designed, in which HyperGraph Neural Hetwork (HGNH) is introduced to formulate hypergraph node classification goals to fully learn the multilateral semantic information contained in the semantic tuples. Moreover, to utilize the heterogeneity of local structures in embedding, the adaptive neighborhood structure is explored by learning the structure embedding, which provides fine-grained ranking lists. Through extensive experiments on three benchmark datasets, the comprehensive results validate the advancement of our proposed DSTH framework over mainstream cross-modal hashing. The source code for the framework DSTH is freely available athttps://github.com/QinLab-WFU/DSTH.

Abstract The accurate detection of moving objects is essential in various applications of artificial intelligence, particularly in the field of intelligent surveillance systems. However, the moving cast shadow detection significantly decreases the precision of moving object detection because they share similar motion characteristics. To address the issue, the authors propose an innovative approach to detect moving cast shadows by combining the hybrid feature with a broad learning system (BLS). The approach involves extracting low‐level features from the input and background images based on colour constancy and texture consistency principles that are shown to be highly effective in moving cast shadow detection. The authors then utilise the BLS to create a hybrid feature and BLS uses the extracted low‐level features as input instead of the original data. BLS is an innovative form of deep learning that can map input to feature nodes and further enhance them by enhancement nodes, resulting in more compact features for classification. Finally, the authors develop an efficient and straightforward post‐processing technique to improve the accuracy of moving object detection. To evaluate the effectiveness and generalisation ability, the authors conduct extensive experiments on public ATON‐CVRR and CDnet datasets to verify the superior performance of our method by comparing with representative approaches.

Deep learning exhibits powerful capability in image inpainting task, particularly in generating pixel-level details closely with the human visual perception. However, the complex background or larger missing regions make it still encounters the artifacts. Many researchers have investigated that prior information is crucial for guiding the image inpainting. In this article, we introduce the dual-attention mechanism, including lightweight spatial attention and linearized attention, to construct an end-to-end texture and structure-guided image inpainting method. In the first stage, we build the detail inpainting network with the lightweight spatial attention. In this model, the extracted texture and structural features are fused with multi-layers and then the fused detail image is considered as the prior to guide the detail repair of corrupted images. In the second stage, we construct the content completing network by the repaired detail and the linearized Transformer module. This module not only overcomes the limitation of the receptive field size of convolutional kernels that can improve the long-range modeling of features but also can significantly reduce the computational complexity of the original Transformer. To demonstrate the superior effectiveness of the proposed method, we perform extensive experiments with advanced models on three datasets: CelebA-HQ, Places2, and Paris Street Views. Comparative results manifest that our method achieves excellent image inpainting results that are conform to the human visual system. The code is available at https://github.com/QinLab-WFU/TSGDAM

Driven by the high nonlinearity of deep neural networks, deep hashing has achieved the pictured great potential in cross-modal retrieval applications, significantly bridging the modality gap. Current deep cross-modal hashing usually utilizes affinity matching or local ranking to capture the local semantic relationships in the learned common space, leading to high neighborhood ambiguity. Simultaneously, most of these frameworks utilize additional regularization terms or margin thresholds to enhance the overall performance, in which searching the model's hyper-parameters under mass training data would have a substantial overhead. In this paper, with a novel extension of information-theoretic measures, a novel deep cross-modal hashing method, named Deep Neighborhood-preserving Hashing (DNpH), is designed to learn a highly separable discrete space, effectively mitigating the semantic gap across different modalities. Specifically, to minimize neighborhood ambiguity, the Quadratic Spherical Mutual Information (QSMI) is first introduced into deep cross-modal hashing to separate neighbors and non-neighbors well, while it is free of tuning parameters during model training compared with other similarity measures. To optimize quadratic mutual information loss smoothly, a square clamping method is developed to improve the stability of model optimization, avoiding converging on bad local optimum. Besides, two transformer encoders are exploited as feature extractors for multi-modal samples to learn the informative semantic representations. Finally, we compare our proposed DNpH framework with various state-of-the-art cross-modal hashing on four public datasets, and large amounts of experiment results demonstrate our contributions and show that DNpH outperforms the compared baselines on different evaluation metrics. The corresponding code is available athttps://github.com/QinLab-WFU/DNpH.

Autonomous vehicles must operate safely in their dynamic and continuously-changing environment. However, the operating environment of an autonomous vehicle is complicated and full of various types of uncertainties. Additionally, the operating environment has many configurations, including static and dynamic obstacles with which an autonomous vehicle must avoid collisions. Though various approaches targeting environment configuration for autonomous vehicles have shown promising results, their effectiveness in dealing with a continuous-changing environment is limited. Thus, it is essential to learn realistic environment configurations of continuously-changing environment, under which an autonomous vehicle should be tested regarding its ability to avoid collisions. Featured with agents dynamically interacting with the environment, Reinforcement Learning (RL) has shown great potential in dealing with complicated problems requiring adapting to the environment. To this end, we present an RL-based environment configuration learning approach, i.e.,DeepCollision, which intelligently learns environment configurations that lead an autonomous vehicle to crash. DeepCollision employs Deep Q-Learning as the RL solution, and selectscollision probabilityas the safety measure, to construct the reward function. We trained four DeepCollision models and conducted an experiment to compare them with two baselines, i.e., random and greedy. Results show that DeepCollision demonstrated significantly better effectiveness in generating collisions compared with the baselines. We also provide recommendations on configuring DeepCollision with the most suitable time interval based on different road structures.

With the aim to reduce car accidents, autonomous driving attracted a lot of attentions these years. However, recently reported crashes indicate that this goal is far from being achieved. Hence, cost-effective testing of autonomous driving systems (ADSs) has become a prominent research topic. The classical model-based testing (MBT), i.e., generating test cases from test models followed by executing the test cases, is ineffective for testing ADSs, mainly because of the constant exposure to ever-changing operating environments, and uncertain internal behaviors due to employed AI techniques. Thus, MBT must be adaptive to guide test case generation based on test execution results in a step-wise manner. To this end, we propose a natural language and model-based approach, named LiveTCM, to automatically execute and generate test case specifications (TCSs) by interacting with an ADS under test and its environment. LiveTCM is evaluated with an open-source ADS and two test generation strategies: Deep Q-Network (DQN)-based and Random. Results show that LiveTCM with DQN can generate TCSs with 56 steps on average in 60 seconds, leading to 6.4 test oracle violations and covering 14 APIs per TCS on average.

Requirements review is an effective technique to ensure the quality of requirements in practice, especially in safety-critical domains (e.g., avionics systems, automotive systems). In such contexts, a typical requirements review process often prioritizes requirements, due to limited time and monetary budget, by, for instance, prioritizing requirements with higher implementation cost earlier in the review process. However, such a requirement implementation cost is typically estimated by stakeholders who often lack knowledge about (future) requirements implementation scenarios, which leads to uncertainty in cost overrun. In this article, we explicitly consider such uncertainty (quantified as cost overrun probability) when prioritizing requirements based on the assumption that a requirement with higher importance, a higher number of dependencies to other requirements, and higher implementation cost will be reviewed with the higher priority. Motivated by this, we formulate four objectives for uncertainty-wise requirements prioritization: maximizing the importance of requirements, requirements dependencies, the implementation cost of requirements, and cost overrun probability. These four objectives are integrated as part of our search-based uncertainty-wise requirements prioritization approach with tool support, named as URP. We evaluated six Multi-Objective Search Algorithms (MOSAs) (i.e., NSGA-II, NSGA-III, MOCell, SPEA2, IBEA, and PAES ) together with Random Search ( RS ) using three real-world datasets (i.e., the RALIC, Word, and ReleasePlanner datasets) and 19 synthetic optimization problems. Results show that all the selected MOSAs can solve the requirements prioritization problem with significantly better performance than RS . Among them, IBEA was over 40% better than RS in terms of permutation effectiveness for the first 10% of prioritized requirements in the prioritization sequence of all three datasets. In addition, IBEA achieved the best performance in terms of the convergence of solutions, and NSGA-III performed the best when considering both the convergence and diversity of nondominated solutions.

Evidence has shown that the use of restricted natural languages can reduce ambiguities in textual use case specifications (UCSs). Restricted natural languages often come with specific editors that support particular use case templates and provide enforcement of the language's restrictions. However, whether restriction enforcement facilitates the definition of UCSs as compared to an editor without such support is a fundamental question to answer. To this end, we report results of a controlled experiment in which we compared two approaches for defining restricted UCSs: (i) a specific Restricted Use Case Modeling (RUCM) tool that supports restriction enforcement; and (ii) a general Office Word UCS template without such enforcement. We compared both approaches from multiple perspectives including restriction misuse, understandability, and restrictiveness. Results show that the restriction misuse rates are generally low, which indicates the usefulness of the RUCM, independent of the use of the editors. The results also indicate that the RUCM tool eases the application of more complex restrictions. We also found that the participants profited from extensive training prior to the experiment. The experiment participants further showed their strong willingness to recommend the RUCM tool to others and to use it in the future, which was not the case for the Office Word template.

Time-related properties are a critical type of extrafunctional requirements for designing real-time systems. Modeling and validating time-related properties at the requirements specification and analysis phases is important for the successful development of real-time systems in terms of cost, quality and productivity. In the literature and practice, timing analyses (e.g., Worst Case Execution Time) are often performed to ensure that the design of a real-time system fully conforms to its time-related constraints. However, such analyses are mostly performed at the design and implementation stages, but not at the requirements level. This paper presents a restricted, natural language based, use case modeling methodology (named as RUCM4RT) to specify functional requirements of real-time systems as use case models, along with associated time-related constraints. RUCM4RT was proposed based on the UML profile for Modeling and Analysis of Real-Time and Embedded Systems (MARTE). In addition, in this paper, we also propose a metamodel-based formalization mechanism named as UCMeta4RT to automatically formalize use case models. We have conducted two real-world case studies to evaluate our solution and 40 use cases were modeled, among which 27 realtime use cases, 118 time-related constraints and 47 other extrafunctional (also commonly called non-functional) constraints were specified. Results show that RUCM4RT was able to handle all the real-time related elements (e.g., time-related constraints) of the use case models.

Summary Ensuring that avionics software meets safety requirements at each development stage is very important to warrant the safe operation of an avionics system. Many safety requirements are imposed by various standards and industrial regulations that must be met by avionics software. One of such standards is DO‐178B/C, which provides guidelines (e.g., development process and objectives to satisfy in development activities) for meeting safety requirements. This paper presents a modeling methodology including a UML profile for specifying safety requirements on a component‐based architecture model and a set of design guidelines on avionics software. These safety requirements were identified from both standards (mainly DO‐178B/C) and current engineering practices in the domain of avionics systems. The methodology automatically enforces these safety requirements. We have applied the methodology on an industrial autopilot system, and several previously uncaught faults were revealed. Copyright © 2014 John Wiley & Sons, Ltd.