Search

Anomaly detection plays a crucial role in various real-world applications, including healthcare and finance systems. Owing to the limited number of anomaly labels in these complex systems, unsupervised anomaly detection methods have attracted great attention in recent years. Two major challenges faced by the existing unsupervised methods are as follows: 1) distinguishing between normal and abnormal data when they are highly mixed together and 2) defining an effective metric to maximize the gap between normal and abnormal data in a hypothesis space, which is built by a representation learner. To that end, this work proposes a novel scoring network with a score-guided regularization to learn and enlarge the anomaly score disparities between normal and abnormal data, enhancing the capability of anomaly detection. With such score-guided strategy, the representation learner can gradually learn more informative representation during the model training stage, especially for the samples in the transition field. Moreover, the scoring network can be incorporated into most of the deep unsupervised representation learning (URL)-based anomaly detection models and enhances them as a plug-in component. We next integrate the scoring network into an autoencoder (AE) and four state-of-the-art models to demonstrate the effectiveness and transferability of the design. These score-guided models are collectively called SG-Models. Extensive experiments on both synthetic and real-world datasets confirm the state-of-the-art performance of SG-Models.

Bus bunching is harmful to the efficiency and stability of bus transit systems, consequently delaying the arrival time of passengers and lowering the public transportation’s adoption rate. Traditional solutions adjust the additional holding time of buses at certain stations to mitigate this phenomenon. These methods sacrifice the system efficiency in exchange for even headway between neighboring buses. Little work focuses on optimizing multiple strategies when a single bus line not only has a bus bay to increase bus dwell time but also owns several dedicated bus lanes to accelerate. In this work, we develop a hierarchical multi-agent reinforcement learning (HMARL) framework to combine these two strategies. Speeding up certain buses via dedicated lanes can counteract the negative influence of additional holding time. Next, to support the two strategies, we devise a two-layer policy scheme, one for high-level policy deciding holding or accelerating and the other for low-level policy determining the specific dwell time or increase of speed. Besides, to handle the issue that the controlling actions of agents are asynchronous and temporally extended, we establish a duration-critic module based on the Recurrent Neural Networks (RNN) mechanism to model other agents’ impact during the period between two consecutive control. We evaluate the proposed framework on a simulated bus line with a quasi-real-world pattern to compare the performance of both traditional headway-based control methods and existing MARL methods. Results show that our method outperforms other baselines, not only stabilizing a strongly unstable bus line but also shortening the traveling times of passengers.

Predicting the next destination is a key in human mobility behavior modeling, which is significant in various fields, such as epidemic control, urban planning, traffic management and recommendation. To achieve this, one typical solution is designing modules based on RNN to capture their preferences to various locations. Although these RNN-based methods can effectively learn individual’s hidden personalized preferences to her visited places, the interactions among users can only be weakly learned through the representations of locations. Targeting this, we propose an end-to-end framework named personalized and group preference guided network (PG2Net), considering the users’ preferences to various places at both individual and collective levels. Specifically, PG2Net concatenates Bi-LSTM and attention mechanism to capture each user’s long-term mobility tendency. To learn population’s group preferences, we utilize spatial and temporal information of the visitations to construct a spatial-temporal dependency module. We adopt a graph embedding method to map users’ trajectory into a hidden space, capturing their sequential relation. In addition, we devise an auxiliary loss to learn the vectorial representation of her next location. Experimental results on two Foursquare check-in datasets and one mobile phone dataset indicate the advantages of our model compared to the state-of-the-art baselines. Source code is available at https://github.com/urbanmobility/PG2Net.

Graphical abstract:

Time-series and event sequences are widely collected data types in real-world applications. Modeling and forecasting of such temporal data play an important role in an informed decision-making process. A major limitation of previous methods is that they either focus on time-series or events, rather than the combination of the two worlds. In fact, the two types of data often provide complementary information, emphasizing the necessity of jointly modeling the both. In this paper, we propose the RNN-ODE collaborative model for joint modeling and forecasting of heterogeneous time-series and event sequence data, which combines several useful techniques from both Bayesian and deep learning for its interpretability. Specifically, we devise a tailored encoder to combine the advances in deep temporal point processes models and variational recurrent neural networks. To predict the probability of event occurrence over an arbitrary continuous-time horizon, we base our model on the mathematical foundation of Neural Ordinary Differential Equations (NODE). Extensive experimental results on simulations and real data sets show that compared with existing methods, our integrated approach can achieve more competitive forecasting performance of both time-series and event sequences.

Anomalies are ubiquitous in real-world time-series data which call for effective and timely detection, especially in an unsupervised setting for labeling cost saving. In this paper, we develop an unsupervised density reconstruction model for multi-dimensional time-series anomaly detection. In particular, it directly handles an important realistic setting that the detection is achieved towards raw time-series contaminated with noise for training, in contrast to most existing anomaly detection works that assume the training data is in general clean i.e., not contaminated with anomaly. It extends recent advancements in deep generative models and state space models to achieve robust anomaly detection. Our approach comprises of a novel state space based generative model, a filtering based inference model, together with a carefully-designated emission model based on robust statistics theory. Extensive experimental results are conducted to show that our approach can adapt to complex patterns even given severely contaminated training data. We also develop visualization techniques to help better understand the behavior of the anomaly detection models. Empirical results show that our method outperforms state-of-the-arts on both synthetic and real-world datasets.

For cloud network performance profiling, network tomography based on end-to-end measurement is often used in deducing the network performance for its efficiency. However, most tomography problems are under-constrained, which require additional assumptions or probing monitors planted among network switches, which are often unavailable in software-defined networking (SDN) environment. On the other hand, SDN-based flow mirroring could provide accurate flow information, but the cost of both gathering and analysing the packet traces is tremendous that it is impossible to cover the whole network. We propose ScoutFlow, a method combining SDN flow measurement and end-to-end performance tomography, to achieve accurate performance profiling for cloud network while keeping low monitoring overhead. We evaluate ScoutFlow in our campus data center cloud, and the experiment shows good scalability and accuracy.

This paper reviews the recent literature on solving the Boolean satisfiability problem (SAT), an archetypal NP-complete problem, with the aid of machine learning (ML) techniques. Over the last decade, the machine learning society advances rapidly and surpasses human performance on several tasks. This trend also inspires a number of works that apply machine learning methods for SAT solving. In this survey, we examine the evolving ML SAT solvers from naive classifiers with handcrafted features to emerging end-to-end SAT solvers, as well as recent progress on combinations of existing conflict-driven clause learning (CDCL) and local search solvers with machine learning methods. Overall, solving SAT with machine learning is a promising yet challenging research topic. We conclude the limitations of current works and suggest possible future directions. The collected paper list is available at https://github.com/Thinklab-SJTU/awesome-ml4co.

Deep generative models have demonstrated their effectiveness in learning latent representation and modeling complex dependencies of time series. In this article, we present a smoothness-inducing sequential variational auto-encoder (VAE) (SISVAE) model for the robust estimation and anomaly detection of multidimensional time series. Our model is based on VAE, and its backbone is fulfilled by a recurrent neural network to capture latent temporal structures of time series for both the generative model and the inference model. Specifically, our model parameterizes mean and variance for each time-stamp with flexible neural networks, resulting in a nonstationary model that can work without the assumption of constant noise as commonly made by existing Markov models. However, such flexibility may cause the model fragile to anomalies. To achieve robust density estimation which can also benefit detection tasks, we propose a smoothness-inducing prior over possible estimations. The proposed prior works as a regularizer that places penalty at nonsmooth reconstructions. Our model is learned efficiently with a novel stochastic gradient variational Bayes estimator. In particular, we study two decision criteria for anomaly detection: reconstruction probability and reconstruction error. We show the effectiveness of our model on both synthetic data sets and public real-world benchmarks. [ABSTRACT FROM AUTHOR]

AbstractDue to their success in routine tasks such as voice recognition, image classification, and text processing, extensive attention has been aroused on how to use artificial intelligence (AI)-based automation tools in the judicial-trial process to improve efficiency. Meanwhile, judicial trial is a complex task that requires accurate insight and subtle analysis of the cases, law, and common knowledge. Applying the results provided by AI-based automation tools directly to the judicial-trial process is controversial due to their irregular logic and low accuracy. Based on this observation, this article investigates the logic underlined in judicial trials and the technical characteristics of AI, and proposes an AI-based semantic assist approach for judicial trials that is logical and transparent to the judges.

•We proposed a data analysis framework leveraging a Bayesian model to identify human activities at a citywide scale from the mobile phone data.•Recognizing that attractiveness of different activities changes over time, we specifically designed our framework to combine the geolocation information in POI data with the temporal visitation pattern extracted from LBSN data when inferring activity types.•Compared with some official surveys and a real-world check-in dataset, our method exhibits a high level of ability to capture and analyze human activity patterns at an urban scale with big data fusion. Based on the inferred user activity sequences, we further explored the various lifestyles of Shanghai inhabitants with an LDA topic model. Our findings offer new insights into urban life that were previously inaccessible.

We design a highly efficient channel drop filter (CDF) with only one channel drop micro-cavity based on photonic crystal (PC) one-way waveguide. By means of the new nature of waveguide-cavity interaction, over 95% channel drop efficiency can be realized in the structure. Some multichannel drop filters with high drop efficiencies are also engineered based on such the structure. These numerical results are all calculated by using the finite element method (FEM), which agrees well with the theoretical analysis result.

An enhanced dynamic wavelength and bandwidth allocation (DWBA) algorithm in hybrid WDM/TDM PON is proposed and experimentally demonstrated. In addition to the fairness of bandwidth allocation, this algorithm also considers the varying propagation delays between ONUs and OLT. The simulation based on MATLAB indicates that the improved algorithm has a better performance compared with some other algorithms.

Optical circuit-switched networks can offer end-to-end dedicated connection with high-capacity bandwidth for largescale applications, but end-systems' performance in these applications often becomes the bottleneck of bulk data transfer in ultra-high-speed network. We propose a bandwidth scheduler based on the end-systems' effective transfer rate and study its performance through experiment.

A very important issue in executing a scientific workflow in optical grid is how to map the workflow tasks onto multiple distributed resources. In this paper, we propose a new cost-based workflow scheduling algorithm (MCBS) which tries to minimize the cost of the workflow execution while meeting the deadline defined by users. The simulation results prove the efficiency of the MCBS algorithm.

Due to the growing demand of flexible resource management for cloud computing services, researches on live virtual machine migration have attained more and more attention. Live migration of virtual machine across different hosts has been a powerful tool to facilitate system maintenance, load balancing, fault tolerance and so on. In this paper, we use a measurement-based approach to compare the performance of two major live migration technologies under certain network conditions, i.e., VMotion and XenMotion. The results show that VMotion generates much less data transferred than XenMotion when migrating identical VMs. However, in network with moderate packet loss and delay, which are typical in a VPN (virtual private network) scenario used to connect the data centers, XenMotion outperforms VMotion in total migration time. We hope that this study can be helpful in choosing suitable virtualization environments for data center administrators and optimizing existing live migration mechanisms.

In the field of cloud computing over optical networks, virtualization is an important issue. Typically this can be implemented on hardware or software. In this paper, we propose an infrastructure-as-a-service (IaaS) framework-OptoVisor, with optical network virtualization implemented on management plane. This framework provides flexible resource scheduling and monitoring. Then we show the viability of our proposal with experimental demonstration and performance test.

With rapid growth of Internet applications, supporting differentiated service and enlarging system capacity have been new tasks for next generation access system. In recent years, research in OFDMA Passive Optical Networks (PON) has experienced extraordinary development as for its large capacity and flexibility in scheduling. Although much work has been done to solve hardware layer obstacles for OFDMA PON, scheduling algorithm on OFDMA PON system is still under primary discussion. In order to support QoS service on OFDMA PON system, a novel dynamic wavelength bandwidth allocation (DWBA) algorithm is proposed in this paper. Per-stream QoS service is supported in this algorithm. Through simulation, we proved our bandwidth allocation algorithm performs better in bandwidth utilization and differentiate service support.

Reliable service delivery becomes more significant due to increased dependency on electronic services all over society and the growing importance of reliable service delivery. As the capability of PON increasing, both residential and business customers may be included in a PON. Meanwhile, OPEX have been proven to be a very important factor of the total cost for a telecommunication operator. Thus, in this paper, we present the partial protection PON architecture and compare the operational expenditures (OPEX) of fully duplicated protection and partly duplicated protection for ONUs with different distributed fiber length, reliability requirement and penalty cost per hour. At last, we propose a differentiated protection method to minimize OPEX.

We propose a simple configuration of wavelength-tunable self-seeding Fabry-Perot fiber laser at ONUs for upstream transmission in hybrid WDM/TDM PON. The performances of the side-mode suppression ratio (SMSR), tuning range, wavelength and power stability for the proposed laser module are experimentally investigated. The performance benefits from the upstream wavelengths sharing are showed via simulations.

A novel polarization channel drop filter (PCDF) based on two-dimensional (2D) photonic crystals (PCs) is presented. It consists of two line defect waveguides and two point defect micro-cavities. In the line-defect waveguides, the transverse-electric (TE) and transverse-magnetic (TM) polarization lights are guided using photonic band-gap and total internal reflection effect, respectively. The light at the resonant frequency for TE polarization can be transferred from one waveguide to the other using the proposed system. Compared with the existing four-port PCDF based on PCs, the three-port structure can realize a multi-channel wavelength system of PCDF more easily and can be an essential device in future polarization wavelength division multiplexing (PWDM) systems.

With the rapid increase of real-time applications, jitter, delay, and throughput have become the three important QoS criteria in the scheduling of input-queued (IQ) switches with virtual output queues (VOQ). In this paper, we propose a novel frame-based scheduling algorithm, referred to as dichotomy slot allocation (DSA), to achieve high throughput, bounded delay, and bounded jitter. DSA possesses three major characteristics. First, DSA schedules the switch on a per-VOQ basis, and strives to provision QoS guarantees for each traffic stream corresponding to each VOQ. Second, DSA allocates the exact amount of slots to schedule the actual traffic of each VOQ, and decides the time slot for each packet. Third, DSA equally allocates the slots in a frame to each port pair, and then adjusts the scheduling to meet the actual traffic demand. The adjustment process employs a specially designed order named Dichotomy Order to guarantee QoS. Both analysis and simulation results demonstrate that DSA achieves good performance in jitter and throughput.

The stable multicast flow aggregation (MFA) problem in internet protocol (IP) over optical network under the dynamical scenario is studied. Given an optical network topology, there is a set of head ends and access routers attached to the optical network, in which each head end can provide a set of programs (IP multicasting flows) and each access router requests a set of programs, we find a set of stable light-trees to accommodate the optimally aggregated multicast IP flows if the requests of access routers changed dynamically. We introduce a program correlation matrix to describe the preference of end users' requests. As the original MFA problem is NP-complete, a heuristic approach, named most correlated program first (MCPF), is presented and compared with the extended least tree first (ELTF) algorithm which is topology-aware. Simulation results show that MCPF can achieve better performance than ELTF in terms of stability with negligible increment of network resource usage.

Abstract We study the connection establishment of label switched paths (LSPs), and compare the LSP blocking performance of the overlay and peer models in IP/MPLS over optical networks. We consider two dynamic routing algorithms for the establishment of LSP connections, of which one is for the overlay model and the other is for the peer model. Our investigations on two typical network topologies, namely NSFNET and ARPA2 networks, show that the number of add/drop ports (or transceivers) on optical cross-connects (OXCs) has a significant impact on the LSP blocking performance for both models. We show by computer simulation that in each case, there is a threshold value for the add/drop ratio, which can achieve almost the best blocking performance. This threshold value remains virtually unchanged as the traffic load varies, but it does depend on the network topology and the number of wavelengths per fiber. This will then indicate the number of add/drop ports to be used so that one can achieve a near optimal blocking performance without incurring unnecessarily excessive network costs. Our investigations reveal that the peer model achieves a much better blocking performance than the overlay model when the number of add/drop ports is relatively high, but that this is not always true when the number of add/drop ports is small.

We demonstrate a 7-bit photonic true-time-delay (TTD) system which uses an 8\times8 micro-optical-electro-mechanical system (MOEMS) optical switch for phased array antennas (PAAs) beamforming applications. The switch controls the optical signal to pass by the fiber delay lines (FDLs) of different lengths. Different time delays between adjacent channels are obtained due to the chromatic dispersion of FDLs. Therefore, the system cannot be disturbed by the environment. The measured time delay responses are nearly linear with the wavelength spacing between optical carriers as well as the lengths of FDLs, which agrees well with the theoretical analysis.

We studied the transmission performance of metro networks employing cascaded thin-film filters (TFFs) for three modulation formats including non-return-to-zero (NRZ), duobinary, and carrier-suppressed RZ (CSRZ). The effects caused by TFFs' dispersion and passband characteristics are investigated for 10- and 40-Gb/s signals, respectively. Simulation results show that the penalty resulted from dispersion can be partially compensated by the passband effects of the cascaded filters. CSRZ is shown to be a preferred format compared with NRZ and duobinary primarily originating from its good back-to-back performance.

Intraband crosstalk penalties for externally and directly modulated lasers are evaluated theoretically and experimentally. The large linewidth of a directly modulated laser reduces beat noise. Theory and experiment agree closely for a single crosstalk source.

We demonstrate an all-optical reconfigurable logic gate based on dominant nonlinear polarization rotation accompanied with cross-gain modulation effect in a single semiconductor optical amplifier (SOA). Five logic functions, including NOT, OR, NOR, AND, and NAND, are realized using 10-Gb/s on-off keying signals with flexible wavelength tunability. The operation principle is explained in detail. By adjusting polarization controllers, multiple logic functions corresponding to different input polarization states are separately achieved using a single SOA with high flexibility.

A cost-effective and service-differentiated provisioning strategy is very desirable to service providers so that they can offer users satisfactory services, while optimizing network resource allocation. Providing differentiated protection services to connections for surviving link failure has been extensively studied in recent years. However, the differentiated protection services for workflow-based applications, which consist of many interdependent tasks, have scarcely been studied. This paper investigates the problem of providing differentiated services for workflow-based applications in optical grid. In this paper, we develop three differentiated protection services provisioning strategies which can provide security level guarantee and network-resource optimization for workflow-based applications. The simulation demonstrates that these heuristic algorithms provide protection cost-effectively while satisfying the applications' failure probability requirements.

We analyze the multicast capability of a multicast-capable optical add/drop multiplexer (MC-OADM) and investigate the dynamic network performance of the MC-OADM ring in comparison with that of the normal OADM ring without multicast capability. The simulation results show that the multicast scheme using MC-OADMs provides more multicast services and occupies fewer resources.

In survivable wavelength-division multiplexed networks, the backup resource can be shared by the primary light paths, which is called primary-backup sharing. We find that primary-backup sharing can have an effect on the link availability. We focus on routing under dynamic availability of the wavelength resource while the requested reliability of traffic has to be met, using the modified shortest path algorithm. We propose a new algorithm called the dynamic link availability algorithm and evaluate the performance of the proposed algorithm on the NSFNET.