Multivariate time series anomaly detection is a crucial factor in real-world applications but a challenging task due to the complex temporal dependencies and system dynamics. Reconstruction-based methods have made great improvements in recent years. However, we observe an issue these methods are suffering, that they primarily measure deviations in the time points themselves when performing anomaly detection but ignore changes in the dynamic properties of the system. In these cases, they are unable to produce sufficient reconstruction errors to detect anomalies, so some potential abnormal time points caused by the dynamic evolution of the system are missing. To address this problem, we propose a novel method, SDA2D, which models system dynamics by the derivative of the NCDE-derived state vector with respect to time, enabling the learning of reconstruction deviation and system evolution jointly. Our experimental results show that SDA2D achieves noticeable improvements in four benchmark datasets, and the visualization also provides further instructions for anomaly diagnosis, which helps locate the sources of these anomalies.

Active learning can significantly reduce the cost of labeling instances while improving model performance. However, similar to other traditional algorithms, active learning encounters the problem of class imbalance and delivers sub-optimal performance. Additionally, existing approaches suffer from poor performance and are time-consuming. To address these issues, we propose an Actively Incrementally Weighted Broad Learning System (AI-WBLS). Firstly, we introduce an active learning framework based on the weighted broad learning system, which employs a double uncertainty sample selection strategy to enhance the value and reasonableness of sample selection in each iteration of active learning. To further improve the model's adaptability during the iterative learning process, an adaptive weighting strategy is designed to adaptively modify the penalty weights according to the changes in the sample labels. Finally, an efficient incremental paradigm is developed to update the model with newly labelled samples instead of re-training, resulting in improved performance and efficiency. Extensive comparative experiments confirm that our approach outperforms other imbalanced active learning methods.

In last-mile delivery, efficient courier scheduling for heterogeneous tasks (i.e., delivery, pick-up, and customer expansion tasks) simultaneously benefits customers, couriers, and the platform. Existing scheduling methods for heterogeneous tasks neglect long-term optimization or are based on simple agent mobility settings. In this work, we focus on a scenario with varying task characteristics compared to existing studies and explore reinforcement learning for courier scheduling in last-mile delivery, considering fine-grained courier mobility. It is challenging due to: i) the large state and action space; ii) the courier cooperation under asynchronous decision-making caused by various task service times and courier travel times between location pairs; and iii) the multi-objective optimization under the dynamic environment. Therefore, we propose a heterogeneous task-aware cooperative courier scheduling system,HeroCS, to improve courier working efficiency and fairness in last-mile delivery. Specifically, we first design a distance-constrained courier modeling module to generate the single courier scheduling decisions, where a distance-aware top-$k$action pruning scheme is proposed to reduce state representation and action space. Then, a cooperative scheduling learning module is explored for optimizing courier cooperation. A status-aware action masking scheme is designed to solve asynchronous decision-making in model training. We also design a hybrid reward function to optimize various objectives adaptively. Extensive evaluation with real-world data from one of the largest logistics companies in China demonstrates that compared to state-of-the-art methods,HeroCSimproves the task completion rate by up to 40.7% and reduces courier unfairness by up to 83.3%.

Last-mile Delivery Parcel Volume (LDPV) quantifies the number of parcels destined for a specific region, particularly a manually divided Area-Of-Interest (AOI). Accurate prediction of AOI-level LDPV is crucial for the efficient management of logistics resources. However, the straightforward adaptation of existing prediction models often falls short, primarily due to (I) a lack of consideration for the intuition behind AOI divisions, and (II) a reliance solely on fully observed historical data, which may not inform future trends. To overcome the above pitfalls, leveraging rich AOI data and advanced parcel travel time estimation services in JD Logistics, this paper introduces a novel framework called Dual-view Prediction Networks (DualPNs). It combines a Vector-Quantified AutoEncoder (VQ-AE) and a Template-Augmented Zero-Inflated Poisson (TA-ZIP), enabling both point and probabilistic distribution predictions of AOI-level LDPV. Specifically, VQ-AE utilizes a vector quantization technique to distill a large number of AOIs into representative templates, thereby addressing the first pitfall. Subsequently, TA-ZIP dynamically integrates fully observed and lookahead features, aligning them with template-specific decoders to parameterize the probabilistic distributions, thus resolving the second pitfall. We conduct extensive experiments in two cities, comprising over 47,000 and 126,000 AOIs respectively, to demonstrate the superiority of our DualPNs over other baselines. Moreover, a real-world case study highlights the effectiveness of DualPNs for enhancing downstream courier allocation by yielding an average improvement of 1.51% in the on-time delivery rate.

Reconstruction-based methods are competitive choices for multivariate time series anomaly detection (MTS AD). However, one challenge these methods may suffer is over generalization, where abnormal inputs are also well reconstructed. In addition, balancing robustness and sensitivity is also important for final performance, as robustness ensures accurate detection in potentially noisy data, while sensitivity enables early detection of subtle anomalies. To address these problems, inspired by idempotent generative network, we take the view from the manifold and propose a novel module named **I**dempotent **G**eneration for **A**nomaly **D**etection (IGAD) which can be flexibly combined with a reconstruction-based method without introducing additional trainable parameters. We modify the manifold to make sure that normal time points can be mapped onto it while tightening it to drop out abnormal time points simultaneously. Regarding the latest findings of AD metrics, we evaluated IGAD on various methods with four real-world datasets, and they achieve visible improvements in VUS-PR than their predecessors, demonstrating the effective potential of IGAD for further improvements in MTS AD tasks. Our instructions on integrating IGAD into customized models and example codes are available at https://github.com/ProEcho1/Idempotent-Generation-for-Anomaly-Detection-IGAD.

Online advertising has been the mainstream monetization approach for internet-based companies, in which bid optimization plays a crucial role in enhancing advertising performance. Currently, the bid optimization problem has narrowed down to two specific forms: Budget-constrained bidding (BCB) and Multi-constraint bidding (MCB). Existing solutions try to solve BCB/MCB via linear programming solvers, learning methods, or feedback control. However, in large-scale complex e-commerce, they still suffer from inefficiency, poor convergence, or slow adaptation to the changing market. This research presents an online receding optimization method as a solution for practical bid optimization problems. We conduct a theoretical analysis of the optimal bidding strategy's structure. Further, an online receding optimization process is designed based on open-loop feedback control, which periodically updates a constructed optimal bid formulation that can be solved by linear programming. Then, considering large-scale linear programming problems, we propose an efficient down sampling scheme. Besides, a neural-network-based auction scale prediction is used to adapt to the changing market. Finally, a series of online A/B experiments onTaobao Sponsored Searchcompare our work to industrial methods and state-of-the-art from several aspects. The proposed method has been implemented onTaobao, a billion-scaled online advertising business, for over a year.

Urban computing leverages data analysis to improve urban areas' efficiency and sustainability, tackling tasks like traffic management, crime forecasting, and air quality predictions. Current models, while efficient, often struggle with tasks beyond their initial training due to limited flexibility. Typically, new tasks require developing specialized models, which may not perform optimally with limited data. To overcome these challenges, we propose the development of a universal pretrained model that understands a city's various aspects comprehensively. This model serves as a robust foundation, ready to be quickly adjusted for different urban tasks as they arise, even if they occur in different cities. Unlike language models, urban computing models must handle unique spatial-temporal dynamics, making standard pretraining techniques inadequate. Our approach includes a spatial-temporal module with multi-graph convolution and temporal attention mechanisms, capturing the necessary spatial-temporal patterns during pretraining. We also integrate a prompt-tuning module within this framework, which can be adapted for new predictive tasks. The results of extensive experiments on four urban predictive tasks across two cities demonstrate the effectiveness of our model

The COVID-19 pandemic has challenged countries worldwide to strike a balance between implementing epidemic control measures and maintaining economic activity. In response, many countries have adopted sustainable, precise, region-specific, and multilevel prevention and control measures. To apply these measures more effectively and purposefully, it is imperative to quantify their impact on the transmission of COVID-19 within urban areas. Here, we propose a dynamic metapopulation susceptible-exposed-infectious-removed (SEIR) model that incorporates the urban mobility network to simulate the spread of COVID-19 in Beijing and investigate the effects of precise intervention measures. Our proposed model accurately fits the real epidemic trajectory, even with the significant changes in human mobility patterns before and after the epidemic. Additionally, it can also serve as a useful policy evaluation tool by simulating the impact of perturbations in mobility networks on epidemic transmission dynamics. Based on this tool, our results demonstrate that point-of-interest capacity limitation measures can significantly reduce the number of infections with only a minor loss of urban mobility. Furthermore, we show that community dynamic management measures can effectively control and mitigate COVID-19 spread while enabling the normal operation of most economic and social activities. By quantifying the impact of precise intervention measures on new infections and mobility losses, our model enables a cost-benefit analysis of these measures, thus informing targeted and sustainable policy responses to COVID-19.

In response to the repeated outbreaks of the COVID-19, many countries implement the region-specific, multilevel epidemic prevention and control policies. To fully understand the impact of these interventions on urban mobility, it is urgent to analyze spatial–temporal mobility pattern at the neighborhood level and structural changes in urban mobility networks. Here, we construct urban mobility networks among points of interest (POIs), using large-scale anonymous mobility data from de-identified mobile phone users. We comprehensively investigate the changes of urban mobility networks during two waves of the COVID-19 pandemic in Beijing from both graph and subgraph perspectives. Beyond an overall mobility reduction in Beijing, we find that the mobility change is spatially and temporally heterogeneous among different urban regions. We uncover a disproportionately large reduction in long-distance, nighttime, and non-essential travel. This results in a more geographically fragmented, local, and regional network in the pandemic. We demonstrate that these structural changes slow down the spatial spread of the COVID-19 in the mobility network.

Effectively predicting the evolution of COVID-19 is of great significance to contain the pandemic. Extensive previous studies proposed a great number of SIR variants, which are efficient to capture the transmission characteristics of COVID-19. However, the parameter estimation methods in previous studies are based on data from epidemiological investigations, which inevitably have caused a large delay. The popularity of digital trajectory data world-wide makes it possible to understand epidemic spreading from human mobility perspective. The major advantage of digital trajectory data lies in that the co-location level of a population is reflected at every moment, making it possible to forecast the evolution in advance. We showed that the mobility data contributed by mobile phone users could be exploited to estimate the contact probability between individuals, thus revealing the dynamic transmission of COVID-19. Specifically, we developed an estimation method to obtain human co-location levels and quantified the variations of human mobility during the epidemic. Then, we extended the infection rate with a real-time co-location level to further forecast the transmission of an epidemic, predicting the epidemic size much more accurately than conventional methods. Finally, the proposed method was applied to evaluate the quantitative effect of different non-pharmacological interventions by predicting the epidemic situations with various mobility characteristics. The empirical results and simulations corroborated our theoretical analysis, providing effective guidance to contain the pandemic.

Despite the substantial progress in predicting human mobility, most existing methods fail to reveal the spatiotemporal patterns under significant interventions such as COVID-19, which disrupt the routine of human mobility. To fill this gap, this paper presents a unified framework for learning human mobility in both regular and intervened scenarios through explicit modeling of the intervention and the intervened system. To be concrete, we design a novel Deep State-Space Model (DSSM) called AGC-ODE: Adaptive Graph Controlled Neural Ordinary Differential Equation for human mobility prediction during COVID-19. The transition equation that describes continuous-time dynamics of human mobility is parameterized with a graph-controlled Neural ODE, and the latent control that guides the equation propagating is inferred through the multi-head gating filters. Additionally, an information capacity constraint is applied to foster the disentanglement of interventions. Lastly, AGC-ODE utilizes a data-driven initialization strategy to improve DSSM’s initial state estimation. We conduct extensive experiments and analysis on two real-world datasets of Beijing and the U.S. to demonstrate the superiority and interpretability of our model. Furthermore, we introduce a deployed system that is based on AGC-ODE and how it helps epidemic prevention during the COVID era and work resumption in the post-COVID era.

Multivariate time series(MTS) is a universal data type related to various real-world applications. Data imputation methods are widely used in MTS applications to deal with the frequent data missing problem. However, these methods inevitably introduce biased imputation and training-redundancy problems in downstream training. To address these challenges, we propose TriD-MAE, a generic pre-trained model for MTS data with missing values. Firstly, we introduce TriD-TCN, an end-to-end module based on TCN that effectively extracts temporal features by integrating dynamic kernel mechanisms and a time-flipping trick. Building upon that, we designed an MAE-based pre-trained model as the precursor of specialized downstream models. Our model cooperates with a dynamic positional embedding mechanism to represent the missing information and generate transferable representation through our proposed encoder units. The overall mixed data feed-in strategy and weighted loss function are established to ensure adequate training of the whole model. Comparative experiment results in time series prediction and classification manifest that our TriD-MAE model outperforms the other state-of-the-art methods within six real-world datasets. Moreover, ablation and interpretability experiments are delivered to verify the validity of TriD-MAE's

To optimize wind energy utilization, wind turbines are increasingly being deployed in remote and challenging locations, including mountains and offshore sites, posing risks on blade hitting tower and potentially resulting in blade breakage. To address these challenges, this paper proposes the Adaptive Weight-Gate Recurrent Unit-Attention Mechanism model for detecting wind turbine hitting towers. The proposed approach includes several key steps. Firstly, a method of training sample adaptive weight is proposed to address the issue of wind condition imbalance. Additionally, we employ discrete Fourier transform to extract multidimensional vibration characteristics from the supervisory control and data acquisition system and perform feature extraction using principal component analysis to determine the detection target. The vibration signature monitoring model combines the Gate Recurrent Unit with an attention mechanism. Lastly, we employ kernel density estimation to establish residual thresholds and design a dynamic alarm strategy based on the consecutive excess times in the training set. The effectiveness and superiority of our proposed approach are validated through experimental results on real datasets collected from various wind farms.

Modeling and predicting the radio spectrum is vital for spectrum management, such as spectrum sharing and anomaly detection. Nevertheless, the precise spectrum prediction is challenging due to the interference from both intra-spectrum and external factors. To tackle these complex internal and external correlations, we develop a model named TF$^2$AN, consisting of three components: 1) a robust signal detection algorithm based on image processing, 2) an attention-based Long Short-term Memory network to capture the temporal-frequency correlations, 3) a generalized fusion module to take the heterogeneous external factors into account. This structure shows prominent effectiveness for spectrum prediction on a single monitoring station with sufficient data. However, when the data derived from a single station is insufficient, the performance of the deep learning model will decline a lot. Considering that more than one monitoring station is deployed in practice, the new challenge becomes how to enhance our model by leveraging the data from multiple stations or frequency bands. Therefore, we further propose T-TF$^2$AN, a transfer learning-based framework for data augmentation and knowledge sharing in spectrum prediction. Compared to TF$^2$AN, better performance is achieved. Besides, the model interpretability and training efficiency are also discussed with two case studies, respectively.

As the IoT technology becomes well established and sharing economy expands worldwide, the bike-sharing system (BSS) has spread fast in the last decade. When introducing the BSS in a new city, the operator often faces many challenges: e.g., optimizing station siting (physical or electric), constructing bike lanes, and making strategies for the initial distribution and rebalancing of bikes. These challenges require an accurate interblock flow prediction before deploying the BSS. In this article, we derive blocks from the urban road network and extract blocks’ features based on the distribution and types of POI. Then, the interblock flow can be predicted based on the features of both start/end block and neighbor blocks. We propose a unified architecture with a generalized attention mechanism named the relation graph network (RGN) to extract block features and make predictions across cities. The evaluations on real-world data sets show that RGN outperforms other popular graph neural networks in multiple metrics. We further simulate the applications of station site selection and bike lane planning based on the prediction from RGN, and the result provides a noticeable increase in the BSS’s deployment/operation efficiency.

Recent years have witnessed the fast proliferation of camera sensors networks (CSNs) in numerous Internet of Things (IoT) applications. In order to a capture distinct image of targets from interesting directions, we leverage a special type of coverage called full-view coverage. Full-view coverage guarantees to obtain the images of a point from every direction, whereas it demands much more sensors than a conventional coverage. To this end, we investigate the problem of deploying the minimum number of rotatable camera sensors to achieve the full-view coverage of a set of target points, namely, optimal deployment for the full-view point coverage (OFP) problem. In this work, camera sensors are capable of rotating freely with infinite orientations, thus not only the deployment locations but also the orientations for each camera sensor are required to be optimized. To tackle this challenging problem, we first prove that the OFP problem is NP-hard. Then, we propose two approximation algorithms—iterative screening algorithm (ISA) and improved ISA (IISA) to solve the OFP. We further perform extensive simulations and conduct physical testings to demonstrate the superiority and effectiveness of our proposed solutions. Experimental results show that IISA can generally reduce the total number of required camera sensors by more than 20% compared with the state-of-the-art work.

As the sharing economy develops and bike-sharing apps emerge, the dockless bike-sharing system (DLBS) has become a competitive alternative to the docked bike-sharing system because of its convenience of finding and parking without physical docks. Meanwhile, new demands are rapidly increasing as DLBS expands, e.g., crowd-sourced re-balancing and pre-ordering during rush hours. A more fine-grained destination prediction is required to tackle these issues. In this paper, we propose a probabilistic-trip-based destination prediction method named P3M. To overcome the uncertainty due to docks’ absence, we introduce the virtual docks derived from POIs and convert a single trip recorded in GPS into several probabilistic trips among POIs using an innovative user behavior model Walking-Riding-Walking Probabilistic Trip. To deal with sparsity, P3M adapts a trip-wised parameter share strategy together with a statistical-based history-feature extractor for better performance without overfitting. Compared with the baseline method, P3M reduces the mean absolute errors measured with distance by 31.55% (from 1.1036 km to 0.7554 km) and is less sensitive to the sparsity of user’s records. Further, we analyze the application of P3M in different types of DLBS and use two simulations to prove its efficiency under insufficient bike supply circumstances.

Radio map, serving as an efficient indicator of wireless environments, has been widely used in smart-city applications, including network monitoring/planning, anomaly signal detection, and indoor/outdoor localization. It is hard to maintain an update-to-date fine-grained radio map within a large area, since the radio map changes rapidly due to the internal and external factors. Previous studies usually relied on time-consuming site surveys at densely predefined reference points, leading to either coarse-grained or out-of-date radio maps. In this paper, we propose a fine-grained radio map reconstruction framework, called Supreme, based on crowd-sourced data in an image super-resolution manner. Specifically, Supreme explores spatial-temporal relationships in historical coarse-grained radio maps and builds a real-time fine-grained radio map using deep spatial-temporal reconstruction networks. Furthermore, a heterogeneous data fusion module is devised to make full use of external information. To evaluate the performance of Supreme, we conduct extensive experiments and ablation studies on a large-scale dataset with a total of six-month data collected from two university campuses. Besides, we investigate the transferability of Supreme in different locations and service networks, showing that the fine-tuned model can largely reduce the training time and achieve better performance. Experimental results demonstrate that our model outperforms state-of-the-art baselines and a case study on the localization is enhanced with marginal improvements on accuracy.

Recently, a considerable research interest has grown up in the millimetre wave wireless system as the most promising technologies in the next generation communication. Since high‐frequency channels of the millimetre wave are easily attenuated in space, beamforming technology relying on the massive multi‐input‐multi‐output system is introduced to transmit the millimetre wave in a very narrow directional beam, so as to greatly improve the transmit efficiency. Then a challenging problem lies in that how to optimise the overall throughput by allocating the antenna resources to different mobile users in the massive MIMO antenna system. In this study, the authors handle such a difficult problem in two different cases. They first begin with the one‐direction case, i.e. all sub‐arrays are deployed in several parallel rows along the edge of a rectangle antenna array. They decompose the problem and solve it gradually. Then they generalise the authors' result to the two‐dimensional case, where the sub‐arrays can be deployed in orthogonal directions. They apply the similar scheme, decompose the problem and solve each sub‐problem progressively. Both NP‐hard problems are solved with time efficient approximation algorithms. Simulation results demonstrate the efficiency of the proposed algorithms in different cases.

Recently, millimeter wave wireless communications have emerged as one of the most promising technologies to significantly improve the throughput of massive multiple-input multiple-output (MIMO) system. Since high-frequency channels are quite easily attenuated in space, beamforming technology based on the massive MIMO is introduced to transmit the millimeter waves in a very narrow directional beam. One challenging problem in this is how to optimize the overall throughput by allocating the available antenna resources to different mobile users. In this paper, we tackle such a difficult problem and formulate it as an antenna selection combinatorial optimization, which is NP-hard. We first begin with the simplified one-dimension case, i.e., all antennas are deployed on a single line segment. We design a novel iterative greedy antenna selection algorithm (iGAS), that allocates antennas to different users in an iterative way, with each iteration maximizing the marginal increase of overall throughput. We then generalize our result to the two-dimension case. Simulation results are provided to demonstrate the efficiency of the proposed algorithms.

Retweeting is an important mechanism for information diffusion, popular event prediction, and so on. Due to the increasing requirements, in recent years, the task has attracted extensive attentions. In this paper, we propose a novel framework using probabilistic matrix factorization technique to predict retweeting behavior. Our study consists of three components. First, we convert retweeting behavior problem to a matrix factorization problem. Second, following the intuition that a user's social network will affect his retweeting behavior, we extensively study how to model social information to improve the prediction accuracy. Finally, message semantic embedding information is employed in designing a semantic regularization term to constrain the matrix factorization objective function. We also propose a set of metrics to construct the embeddings among messages based on messages' structural and textual features. The empirical results and analysis demonstrate that our methods perform better than the state-of-the-art approaches.

The approach of data deduplication has been widely used in backup systems and primary storage such as virtual machine platform. However, the reading speed in those systems suffers due to chunk fragmentation in deduplication. So it has become an important problem to improve reading performance in deduplication systems. In this paper, firstly we propose a new storage method using multiple disks to boost reading performance, which is called MMD. MMD takes advantage of the multiple parallelized disks, each of which is used as independent logical device. Then we present a deduplication model based on MMD, which focuses on optimization of data layout on disks to improve reading speed. Two I/O scheduling algorithms in that model are discussed, which aim at assigning the containers in deduplication systems to appropriate disks. Experiments show that MMD can achieve an obvious reading performance improvement than RAID in deduplication systems.