We propose two solutions to outlier detection in time series based on recurrent autoencoder ensembles. The exploit autoencoders built using sparsely-connected neural networks (S-RNNs). Such make it possible generate multiple with different network connection structures. are ensemble frameworks, specifically an independent framework and a shared framework, both of which combine S-RNN enable detection. This ensemble-based approach aims reduce the effects some being overfitted outliers, this...

When planning routes, drivers usually consider a multitude of different travel costs, e.g., distances, times, and fuel consumption. Different may choose routes between the same source destination because they have driving preferences (e.g., time-efficient v.s. fuel-efficient driving). However, existing routing services support little in modeling multiple costs personalization-they deliver that minimize single cost shortest or fastest routes) to all drivers. We study problem how recommend...

Time series are generated in diverse domains such as economic, traffic, health, and energy, where forecasting of future values has numerous important applications. Not surprisingly, many methods being proposed. To ensure progress, it is essential to be able study compare empirically a comprehensive reliable manner. achieve this, we propose TFB, an automated benchmark for Series Forecasting (TSF) methods. TFB advances the state-of-the-art by addressing shortcomings related datasets,...

The monitoring of a system can yield set measurements that be modeled as collection time series. These series are often sparse, due to missing measurements, and spatiotemporally correlated, meaning spatially close exhibit temporal correlation. analysis such offers insight into the underlying enables prediction behavior. While techniques presented in paper apply more generally, we consider case transportation systems aim predict travel cost from GPS tracking data probe vehicles. Specifically,...

Different uses of a road network call for the consideration different travel costs: in route planning, time and distance are typically considered, green house gas (GHG) emissions increasingly being considered. Further, costs such as GHG time-dependent uncertain. To support uses, we propose techniques that enable construction multi-cost, time-dependent, uncertain graph (MTUG) model based on GPS data from vehicles traversed network. Based MTUG, define stochastic skyline routes consider...

Cyber-physical systems often consist of entities that interact with each other over time. Meanwhile, as part the continued digitization industrial processes, various sensor technologies are deployed enable us to record time-varying attributes (a.k.a., time series) such entities, thus producing correlated series. To accurate forecasting on series, this paper proposes two models combine convolutional neural networks (CNNs) and recurrent (RNNs). The first model employs a CNN individual combines...

Innovations in transportation, such as mobility-on-demand services and autonomous driving, call for high-resolution routing that relies on an accurate representation of travel time throughout the underlying road network. Specifically, a road-network edge is modeled time-varying distribution captures variability traffic over fact different drivers may traverse same at speeds. Such stochastic weights be extracted from data sources GPS loop detector data. However, even very large are incapable...

Origin-destination (OD) matrices are used widely in transportation and logistics to record the travel cost (e.g., speed or greenhouse gas emission) between pairs of OD regions during different intervals within a day. We model as distribution because when traveling pair regions, vehicles may at speeds even same interval, e.g., due driving styles waiting times intersections. This yields stochastic matrices. consider an increasingly pertinent setting where set vehicle trips is for instantiating...

A variety of real-world applications rely on far future information to make decisions, thus calling for efficient and accurate long sequence multivariate time series forecasting. While recent attention-based forecasting models show strong abilities in capturing long-term dependencies, they still suffer from two key limitations. First, canonical self attention has a quadratic complexity w.r.t. the input length, falling short efficiency. Second, different variables’ often have distinct...

Traffic time series forecasting is challenging due to complex spatio-temporal dynamics-time from different locations often have distinct patterns; and for the same series, patterns may vary across time, where, example, there exist certain periods a day showing stronger temporal correlations. Although recent models, in particular deep learning based show promising results, they suf-fer being agnostic. Such agnostic models employ shared parameter space irrespective of assume that are similar...

Time series data occurs widely, and outlier detection is a fundamental problem in mining, which has numerous applications. Existing autoencoder-based approaches deliver state-of-the-art performance on challenging real-world but are vulnerable to outliers exhibit low explainability. To address these two limitations, we propose robust explainable unsupervised auto encoder frameworks that decompose an input time into clean using autoencoders. Improved explainability achieved because better...

Transformer-based models have achieved some success in time series forecasting. Existing methods mainly model from limited or fixed scales, making it challenging to capture different characteristics spanning various scales. In this paper, we propose multi-scale transformers with adaptive pathways (Pathformer). The proposed Transformer integrates both temporal resolution and distance for modeling. Multi-scale division divides the into resolutions using patches of sizes. Based on each scale,...

With the growing volumes of vehicle trajectory data, it becomes increasingly possible to capture time-varying and uncertain travel costs in a road network, including time fuel consumption. The current paradigm represents network as weighted graph; blasts trajectories into small fragments that fit under-lying edges assign weights edges; then applies routing algorithm resulting graph. We propose new paradigm, hybrid graph , targets more accurate efficient path cost distribution estimation....

Motivated by the increasing availability of vehicle trajectory data, we propose learn-to-route, a comprehensive trajectory-based routing solution. Specifically, first construct graph-like structure from trajectories as infrastructure. Second, enable given an arbitrary (source, destination) pair. In step, road network and collection trajectories, clustering method that identifies regions in network. If pair are connected maintain paths used these learn preference for travel between regions....

Correlated time series forecasting plays an essential role in many cyber-physical systems, where entities interact with each other over time. To enable accurate forecasting, it is to capture both the temporal dynamics and correlations among different entities. former, two popular types of models, recurrent neural networks (RNNs) convolution (TCNs), are employed. latter, a graph constructed reflect certain relationships then (GC) applied upon The state-of-the-art accuracy achieved by models...

We propose variational quasi-recurrent autoencoders (VQRAEs) to enable robust and efficient anomaly detection in time series unsupervised settings. The proposed VQRAEs employs a judiciously designed objective function based on divergences, including a, ß, and, -divergence, making it possible separate anomalies from normal data without the reliance labels, thus achieving robustness fully training. To better capture temporal dependencies data, are built upon neural networks, which employ...

Due to the sweeping digitalization of processes, increasingly vast amounts time series data are being produced. Accurate classification such facilitates decision making in multiple domains. State-of-the-art accuracy is often achieved by ensemble learning where results synthesized from base models. This characteristic implies that needs substantial computing resources, preventing their use resource-limited environments, as edge devices. To extend applicability learning, we propose LightTS...

Sensors in cyber-physical systems often capture interconnected processes and thus emit correlated time series (CTS), the forecasting of which enables important applications. The key to successful CTS is uncover temporal dynamics spatial correlations among series. Deep learning-based solutions exhibit impressive performance at discerning these aspects. In particular, automated forecasting, where design an optimal deep learning architecture automated, accuracy that surpasses what has been...