Time series forecasting has historically been a key area of academic research and industrial applications. In multi-horizon and multi-series forecasting tasks, accurately capturing the local information in a sequence and effectively sharing global information across different sequences are very challenging, due to the complex dependencies over time in a long sequence and the heterogeneous nature across multiple time series. In this paper, from the perspective of causal inference, we give a theoretical analysis of these difficulties and establish a causal graph to identify the confounding relationship that generates harmful bias and misleads the time series model to capture the spurious correlations. We propose a causal triple attention time series forecasting model with three interpretable attention modules, which leverages the front-door adjustment to remove the confounding effect and help the model effectively utilize the local and global temporal information. We evaluate the performance of our model on four benchmark datasets and the results demonstrate the superiority over the state-of-the-art methods.

Marketing campaigns are a set of strategic activities that can promote a business's goal. The effect prediction for marketing campaigns in a real industrial scenario is very complex and challenging due to the fact that prior knowledge is often learned from observation data, without any intervention for the marketing campaign. Furthermore, each subject is always under the interference of several marketing campaigns simultaneously. Therefore, we cannot easily parse and evaluate the effect of a single marketing campaign. To the best of our knowledge, there are currently no effective methodologies to solve such a problem, i.e., modeling an individual-level prediction task based on a hierarchical structure with multiple intertwined events. In this paper, we provide an in-depth analysis of the underlying parse tree-like structure involved in the effect prediction task and we further establish a Hierarchical Capsule Prediction Network (HapNet) for predicting the effects of marketing campaigns. Extensive results based on both the synthetic data and real data demonstrate the superiority of our model over the state-of-the-art methods and show remarkable practicability in real industrial applications.

Although the Conditional Variational Auto-Encoder (CVAE) model can generate more diversified responses than the traditional Seq2Seq model, the responses often have low relevance with the input words or are illogical with the question. A causal analysis is carried out to study the reasons behind, and a methodology of searching for the mediators and mitigating the confounding bias in dialogues is provided. Specifically, we propose to predict the mediators to preserve relevant information and auto-regressively incorporate the mediators into generating process. Besides, a dynamic topic graph guided conditional variational auto-encoder (TGG-CVAE) model is utilized to complement the semantic space and reduce the confounding bias in responses. Extensive experiments demonstrate that the proposed model is able to generate both relevant and informative responses, and outperforms the state-of-the-art in terms of automatic metrics and human evaluations.

Multi-task learning(MTL) is an open and challenging problem in various real-world applications. The typical way of conducting multi-task learning is establishing some global parameter sharing mechanism across all tasks or assigning each task an individual set of parameters with cross-connections between tasks. However, for most existing approaches, all tasks just thoroughly or proportionally share all the features without distinguishing the helpfulness of them. By that, some tasks would be intervened by the unhelpful features that are useful for other tasks, leading to undesired negative transfer between tasks. In this paper, we design a novel architecture named the Multiple-level Sparse Sharing Model (MSSM), which can learn features selectively and share knowledge across all tasks efficiently. MSSM first employs a field-level sparse connection module (FSCM) to enable much more expressive combinations of feature fields to be learned for generalization across tasks while still allowing for task-specific features to be customized for each task. Furthermore, a cell-level sparse sharing module (CSSM) can recognize the sharing pattern through a set of coding variables that selectively choose which cells to route for a given task. Extensive experimental results on several real-world datasets show that MSSM outperforms SOTA models significantly in terms of AUC and LogLoss metrics.

Multi-dimensional Hawkes processes (MHP) has been widely used for modeling temporal events. However, when MHP was used for modeling events with spatio-temporal characteristics, the spatial information was often ignored despite its importance. In this paper, we introduce a framework to exploit MHP for modeling spatio-temporal events by considering both temporal and spatial information. Specifically, we design a graph regularization method to effectively integrate the prior spatial structure into MHP for learning influence matrix between different locations. Indeed, the prior spatial structure can be first represented as a connection graph. Then, a multi-view method is utilized for the alignment of the prior connection graph and influence matrix while preserving the sparsity and low-rank properties of the kernel matrix. Moreover, we develop an optimization scheme using an alternating direction method of multipliers to solve the resulting optimization problem. Finally, the experimental results show that we are able to learn the interaction patterns between different geographical areas more effectively with prior connection graph introduced for regularization.

Improving safety and convenience is always the top priority in designing today's intelligent transportation system. In this paper, we study the problem of how to manage vehicle traffic at intersections by jointly considering safety, driver's comfort, and efficiency, in vehicular ad hoc networks. We propose a distributed intersection management protocol (DIMP), which distributedly coordinates vehicle traffic from different directions by making vehicles exchange critical driving information and adaptively react based on the information. DIMP dynamically guides vehicles to adjust their speed in a way such that both safety and driver's comfort are satisfied. By following DIMP, vehicles can pass the intersections safely and efficiently at a comfortable speed during acceleration/deceleration. We extensively evaluate DIMP, and the evaluation results show that DIMP is both effective and efficient in managing vehicle traffic at intersections.

Accurately identifying time-invariant operational relationships among different components is critical to autonomic management of complex manufactural systems. In this paper, we collect time series of sensor readings from manufacturing systems, and propose a solution leveraging Sparse Group LASSO to discover structured pairwise nonlinear relationships and quantify them by mathematical formulas. We consider both real-life operational patterns and underlying physical reactions inside the manufactural systems, which leads to a learning formulation for combined periodic and aperiodic system behaviors. An accelerated gradient descent algorithm is developed to efficiently solve the related optimization problem. We estimate sample correlations between proximal time points to improve the accuracy of the discovered relationships and the nonlinear quantitative formulas. The method is evaluated using both synthetic and real- world datasets, which shows superior performance over the state of the art in discovering nonlinear relationships in manufactural systems.

Complex systems are prevalent in many fields such as finance, security and industry. A fundamental problem in system management is to perform diagnosis in case of system failure such that the causal anomalies, i.e., root causes, can be identified for system debugging and repair. Recently, invariant network has proven a powerful tool in characterizing complex system behaviors. In an invariant network, a node represents a system component, and an edge indicates a stable interaction between two components. Recent approaches have shown that by modeling fault propagation in the invariant network, causal anomalies can be effectively discovered. Despite their success, the existing methods have a major limitation: they typically assume there is only a single and global fault propagation in the entire network. However, in real-world large-scale complex systems, it's more common for multiple fault propagations to grow simultaneously and locally within different node clusters and jointly define the system failure status. Inspired by this key observation, we propose a two-phase framework to identify and rank causal anomalies. In the first phase, a probabilistic clustering is performed to uncover impaired node clusters in the invariant network. Then, in the second phase, a low-rank network diffusion model is designed to backtrack causal anomalies in different impaired clusters. Extensive experimental results on real-life datasets demonstrate the effectiveness of our method.

Vehicle trajectory information can enable many promising applications, but obtaining the information is very challenging. Thanks to the Vehicular Adhoc Network (VANET), in which vehicles and access points can communicate, we can collect trajectory data from vehicles in an area through wireless communication. In this paper, we select intersections to disseminate requests for collecting moving trajectories of all the vehicles in the area, while at the same time minimizing the number of selected intersections. We prove the underlying intersection selection problem is NP-Complete and propose a greedy heuristic to address it approximately. The performance, efficiency and practical issues such as broadcast intervals of the scheme are analyzed in this paper. To deal with the situation that access points are unavailable, we propose a distributed protocol to make vehicles conduct message broadcasting at the selected intersections. The simulation shows our distributed protocol is both efficient and effective.

The rapid growth of mobile app traffic brings huge pressure to today's cellular networks. While this fact is commonly concerned by all the mobile carriers, little work has been done to analyze app's network resource usage. In this paper, we, for the first time, profile network resource usages for mobile apps by establishing a quantitative mapping between them. We design AppWiR, a crowdsourcing-based mining system that collects app behavior information from phones and mines hundreds of indicators in different network layers. It builds a two-layer causal relationship among app behaviors, network traffics, and network resources. With such relationship knowledge, we model, quantify, and predict the network resource usage for each mobile app. We fully implement the AppWiR crowdsourcing app in Android smartphones to collect data from users. To evaluate its real-world performance, we deploy the AppWiR system and conduct a trial in a leading LTE carrier's network in different geographic areas and network coverages. The trial shows that the AppWiR can accurately estimate and predict the resource usages for mobile apps.

High network connectivity and low energy consumption are two major challenges in wireless sensor networks (WSNs). It is even more challenging to achieve both at the same time. To tackle the problem, this paper proposes a novel disjoint Set Division (SEDO) algorithm for joint scheduling and routing in WSNs. We finely divide sensors into different disjoint sets with guaranteed connectivity based on their geographical locations to monitor the interested area. We propose a class of scheduling and routing algorithms, which sequentially schedule each disjoint set to be on and off and balance the energy consumption during packet transmission. Simulation results show that SEDO outperforms existing schemes with lower packet delivery latency and longer network lifetime.

Sensor nodes deployed outdoors for field surveillance are subject to environmental detriments. In this article, we propose a heterogeneous sensor network composed of sensor nodes with different environmental survivability to make it robust to environmental damage and keep it at a reasonable cost. We, for the first time, study the scheduling problem in such heterogeneous sensor networks for critical location surveillance applications. Our goal is to monitor all the critical points for as long as possible under different environmental conditions. We identify the underlying problem, theoretically prove its NP-complete nature, and propose a novel adaptive greedy scheduling algorithm to solve the problem. The algorithm incorporates several heuristics to schedule the activity of both regular and robust sensors to monitor all the critical points, while at the same time minimizing and balancing the network energy consumption. Simulation results show that our algorithm efficiently solves the problem and outperforms other alternatives.

The long term operation of physical systems inevitably leads to their wearing out, and may cause degradations in performance or the unexpected failure of the entire system. To reduce the possibility of such unanticipated failures, the system must be monitored for tell-tale symptoms of degradation that are suggestive of imminent failure. In this work, we introduce a novel time series analysis technique that allows the decomposition of the time series into trend and fluctuation components, providing the monitoring software with actionable information about the changes of the system's behavior over time. We analyze the underlying problem and formulate it to a Quadratic Programming (QP) problem that can be solved with existing QP-solvers. However, when the profiling resolution is high, as generally required by real-world applications, such a decomposition becomes intractable to general QP-solvers. To speed up the problem solving, we further transform the problem and present a novel QP formulation, Non-negative QP, for the problem and demonstrate a tractable solution that bypasses the use of slow general QP-solvers. We demonstrate our ideas on both synthetic and real datasets, showing that our method allows us to accurately extract the degradation phenomenon of time series. We further demonstrate the generality of our ideas by applying them beyond classic machine prognostics to problems in identifying the influence of news events on currency exchange rates and stock prices. We fully implement our profiling system and deploy it into several physical systems, such as chemical plants and nuclear power plants, and it greatly helps detect the degradation phenomenon, and diagnose the corresponding components.

An accurate and automated identification of operational behavior switching is critical to the autonomic management of complex systems. In this paper, we collect sensor readings from those systems, which are treated as time series, and propose a solution to discover switching behaviors by inferring the relationship changes among massive time series. The method first learns a sequence of local relationship models that can best fit the time series data, and then combines the changes of local relationships to identify the system level behavior switching. In the local relationship modeling, we formulate the underlying switching identification as a segmentation problem, and propose a sophisticated optimization algorithm to accurately discover different segments in time series. In addition, we develop a hierarchical optimization strategy to further improve the efficiency of segmentation. To unveil the system level behavior switching, we present a density estimation and mode search algorithm to effectively aggregate the segmented local relationships so that the global switch points can be captured. Our method has been evaluated on both synthetic data and datasets from real systems. Experimental results demonstrate that it can successfully discover behavior switching in different systems.

This paper proposes a novel framework to automatically pinpoint suspicious sensors that lead to the quality change in physical systems such as manufacture plants. Our framework treats sensor readings as time series, and contains three main stages: time series transformation to feature series, feature ranking, and ranking score fusion. In the first step, we transform time series into a number of different feature series to describe the underlying dynamics of each sensor data. After that, the importance scores of all feature series are computed by utilizing several feature selection and ranking techniques, each of which discovers specific aspects of feature importance and their dependencies in the feature space. Finally we combine importance scores from all the rankers and all the features to obtain the final ranking of each sensor with respect to the system quality change. Our experiments based on synthetic time series as well as sensor data from a real system demonstrate the effectiveness of proposed method. In addition, we have implemented our framework as a production engine, and successfully applied it to several real physical systems.

Scalable and diagnosable are the two most crucial needs for voice call quality assessment in mobile networks. However, while these two requirements are widely accepted by mobile carriers, they do not receive enough attention during the development. Current related research mainly focuses on audio feature analysis, which is costly, sensitive to language and tones, and infeasible to be applied to large-scale mobile networks. In this paper, we revisit this problem, and for the first time explore wireless network, the causal factor that directly impacts the mobile voice quality but yet lacks attention for decades. We design CrowdMi, a wireless analytical tool that model the mobile voice quality by crowdsourcing and mining the network indicators of cellphones. CrowdMi mines hundreds of network indicators to build a causal relationship between voice quality and network conditions, and carefully calibrates the model according to the widely accepted perceptual objective listening quality assessment (POLQA) voice assessment standard. We implement a light-load CrowdMi Client App in Android smartphones, which automatically collects data through user crowdsourcing and outputs to the CrowdMi Server in our data center that runs the mining algorithm. We conduct a pilot trial in VoLTE network in different geographical areas and network coverages. The trial shows that the CrowdMi does not require any additional hardware or human effort, and has very high model accuracy and strong diagnosability.

Existing multipath routing protocols for MANET ignore the topology-exposure problem. This paper analyzes the threat of topology-exposure and proposes a Topology-Hiding Multipath Routing protocol (THMR). THMR doesn't allow packets to carry routing information, so malicious nodes cannot deduce topology information and launch various attacks based on that. The protocol can also establish multiple node-disjoint routes in a route discovery attempt and exclude unreliable routes before transmitting packets. We formally prove that THMR is loop-free and topology-hiding. Simulation results show that our protocol has better capability of finding routes and can greatly increase the capability of delivering packets in the scenario where there are attackers at the cost of low routing overhead.

Time series represent sequences of data points where usually their order is defined by the time when they were recorded. Thus, virtually any sequential recordings can be stored as time series: Stock prices, weather conditions, physical system parameters change, product quality monitoring, etc. This leads to ubiquity of time series in all scientific and practical fields (Esling and Agon, 2012); hence, they have attracted significant research efforts over the past decades. The time series can be univariate, i.e., only one variable recorded, or multivariate, i.e., a set of observations from different sources recorded at some time points. The tasks of time series analysis essentially defined to extract meaningful information from the collections of data points or to organize fast and easy access to the necessary data. In this article, we will briefly discuss the major tasks such as classification, clustering, prediction, segmentation and indexing of time series.

Sensor nodes deployed outdoors are subject to environmental detriments and often need to cache data for an extended period of time. This paper introduces sensor nodes which are robust to environmental damages, and proposes to utilize Network Coding to back up data in the robust sensors for future data retrieval in an energy efficient way. Our goal is to help regular sensors select robust sensors to back up their data with low energy consumption, such that when needed, all the data can be retrieved by querying only a subset of robust sensors. We formally formulate this backup problem, theoretically prove its NP-Completeness, discover two novel theoretical guidelines for problem solving, and propose two algorithms accordingly to tackle this NP-C problem. The guidelines are based on random linear network coding and provide lower bounds of the number of robust sensors that each regular sensor should choose for data backup, such that the required fault tolerance is provided. A centralized algorithm and a distributed algorithm are developed based on the guidelines such that regular sensors can back up their data efficiently. Both analysis and simulation show our algorithms are effective in achieving fault tolerance, low energy consumption, and high retrieval efficiency.

GPS navigators have been widely adopted by drivers. However, due to the sensibility of GPS signals to terrain, vehicles cannot get their locations when they are inside a tunnel or on a road surrounded by high-rises where satellite signal is blocked. This incurs safety and convenience problems. To address the issue, we propose a novel Grid-based On-road localizaTion system (GOT), where vehicles with and without accurate GPS signals self-organize into a Vehicular Ad Hoc Network (VANET), exchange location and distance information and help each other to calculate an accurate position for all the vehicles inside the network. The location information can be exchanged among vehicles one or multiple hops away in this paper. We explore fuzzy geometric relationship among vehicles, and apply a novel grid-based mechanism to evaluate the geometric relationships and calculate vehicle locations. Simulation shows our GOT system is effective and efficient in calculating vehicular positions.