Generative adversarial networks (GANs) have succeeded in inducing cross-lingual word embeddings -maps of matching words across languages- without supervision. Despite these successes, GANs' performance for the difficult case of distant languages is still not satisfactory. These limitations have been explained by GANs' incorrect assumption that source and target embedding spaces are related by a single linear mapping and are approximately isomorphic. We assume instead that, especially across distant languages, the mapping is only piece-wise linear, and propose a multi-adversarial learning method. This novel method induces the seed cross-lingual dictionary through multiple mappings, each induced to fit the mapping for one subspace. Our experiments on unsupervised bilingual lexicon induction and cross-lingual document classification show that this method improves performance over previous single-mapping methods, especially for distant languages.

Distributed representations of words which map each word to a continuous vector have proven useful in capturing important linguistic information not only in a single language but also across different languages. Current unsupervised adversarial approaches show that it is possible to build a mapping matrix that align two sets of monolingual word embeddings together without high quality parallel data such as a dictionary or a sentence-aligned corpus. However, without post refinement, the performance of these methods' preliminary mapping is not good, leading to poor performance for typologically distant languages. In this paper, we propose a weakly-supervised adversarial training method to overcome this limitation, based on the intuition that mapping across languages is better done at the concept level than at the word level. We propose a concept-based adversarial training method which for most languages improves the performance of previous unsupervised adversarial methods, especially for typologically distant language pairs.

This paper describes the University of Geneva’s submission to the CoNLL 2017 shared task Multilingual Parsing from Raw Text to Universal Dependencies (listed as the CLCL (Geneva) entry). Our submitted parsing system is the grandchild of the first transition-based neural network dependency parser, which was the University of Geneva’s entry in the CoNLL 2007 multilingual dependency parsing shared task, with some improvements to speed and portability. These results provide a baseline for investigating how far we have come in the past ten years of work on neural network dependency parsing.

Global Positioning System technology has been widely used in vehicle tracking and road planning applications. An enormous amount of data concerning the trajectories of vehicles has been collected and stored for tracking purposes. A trajectory contains not only the footprints of a moving object but also additional information, such as speed and stopping points. Therefore, the large-scale trajectory data sets provide rich information and are currently attracting considerable attention; there have been many successful studies of event detection based on trajectory data. However, most of these studies have focused only on vehicles traveling in a road network and have note considered maritime trajectories. A maritime trajectory also contains auxiliary data (e.g., speed and rotation) in addition to the movements of a ship. However, ships are not bound to road networks, and consequently, it is difficult to apply traditional mining algorithms based on road networks. In addition, even if the amount of maritime trajectory data is very large, these data are also spatially sparse, which will significantly reduce the effectiveness of most existing mining algorithms. In this paper, we propose a new method of abnormal trajectory detection to address this problem. This method can detect abnormal vessel trajectories from Automatic Identification System (AIS), records for vessels via our feature learning algorithm. To reduce the search space, we invoke reference points as well as the Piecewise Linear Segmentation (PLS), algorithm to compress the trajectories without losing important information. A time-aware and spatially correlated collaborative algorithm is proposed to increase the density of the trajectories to improve the accuracy of the detection algorithm, which is based on Dynamic Time Warping (DTW). Finally, we report experiments conducted on a real-world data set, which demonstrate that the proposed detection method can detect anomalous trajectories effectively.

The last decade has witnessed the prevalence of sensor and GPS technologies that produce a high volume of trajectory data representing the motion history of moving objects. However some characteristics of trajectories such as variable lengths and asynchronous sampling rates make it difficult to fit into traditional database systems that are disk-based and tuple-oriented. Motivated by the success of column store and recent development of in-memory databases, we try to explore the potential opportunities of boosting the performance of trajectory data processing by designing a novel trajectory storage within main memory. In contrast to most existing trajectory indexing methods that keep consecutive samples of the same trajectory in the same disk page, we partition the database into frames in which the positions of all moving objects at the same time instant are stored together and aligned in main memory. We found this column-wise storage to be surprisingly well suited for in-memory computing since most frames can be stored in highly compressed form, which is pivotal for increasing the memory throughput and reducing CPU-cache miss. The independence between frames also makes them natural working units when parallelizing data processing on a multi-core environment. Lastly we run a variety of common trajectory queries on both real and synthetic datasets in order to demonstrate advantages and study the limitations of our proposed storage.

This package contains the system outputs from the CoNLL 2017 Shared Task in Multilingual Parsing from Raw Text to Universal Dependencies.

In the last decade, the trajectories data have been collected by many applications and such trajectories contain rich information that can be used to detect events especially for anomaly event detections. However, there are still many challenges on this problem, the major one is how to identify the similar trajectories on semantic level. In this work, we extract the nature features from raw trajectories and use them to do the semantic trajectory similarity search. To achieve this, we propose a PLS algorithm to detect such semantic similar trajectories efficiently and effectively. We also leverage the DBSCAN to help extract the information from large trajectory data. The results of our algorithm are demonstrated by the real world dataset.

Cloud storage is a kind of external storage which can provide by unlimited storage space with high availability and low cost on maintenance. On the other side, the size of geospatial data is too large and is increasing dramatically which makes such data is hard to be stored in the local data warehouse. Hence following the benefits of Cloud storage, such geospatial data is suitable to be stored in Cloud storage and managed by local data warehouse. However, there is a gap between Cloud storages and data warehouses built on traditional infrastructures, such as the mostly adopted massive parallel processing (MPP) based data warehouse. Therefore, in this paper, we propose a middleware-like architecture to connect MPP data warehouse and Cloud storage. It supports traditional geospatial data retrieving while integrating the Cloud storage lineage by a set of technical designs. Based on the prototype system and practical data, we demonstrate the appreciable performance and the flexibility for other third parties' development. Another major contribution of this paper is that we implement the prototype on open-source data warehouse and we make it open-sourced to public.

Traditional machine translation evaluation metrics such as BLEU and WER have been widely used, but these metrics have poor correlations with human judgements because they badly represent word similarity and impose strict identity matching. In this paper, we propose some modifications to the traditional measures based on word embeddings for these two metrics. The evaluation results show that our modifications significantly improve their correlation with human judgements.

An increasing amount of motion history data, which is called trajectory, is being collected from different sources such as GPS-enabled mobile devices, surveillance cameras and social networks. However it is hard to store and manage trajectory data in traditional database systems, since its variable lengths and asynchronous sampling rates do not fit disk-based and tuple-oriented structures, which are the fundamental structures of traditional database systems. We implement a novel trajectory storage system that is motivated by the success of column store and recent development of in-memory based databases. In this storage design, we try to explore the potential opportunities, which can boost the performance of query processing for trajectory data. To achieve this, we partition the trajectories into frames as column-oriented storage in order to store the sample points of a moving object, which are aligned by the time interval, within the main memory. Furthermore, the frames can be highly compressed and well structured to increase the memory utilization ratio and reduce the CPU-cache missing. It is also easier for parallelizing data processing on the multi-core server since the frames are mutually independent.

Owing to the development of cheap RAM-based storage technology, modern computing hardware can afford much larger main memory. Consequently, traditional database systems can be re-designed to store and manage all the data in main memory permanently. Such kind of in-memory database systems (IMDB) have attracted increasing attention from both academia and industry due to its outstanding performance in processing large amount of data. In this work, we will exploit the computational power of SAP HANA, the in-memory column-oriented data analytics platform designed by SAP, to support efficient query processing for moving object trajectories. We have tailored the frame-based data structure designed by our previous SharkDB project and made the trajectory data with variable lengths and sampling rates suitable for relational database model in SAP HANA. Extensive experiments based on large-scale real dataset have demonstrated superior performance of our frame-based design in processing a variant of queries.

Due to the prevalence of GPS-enabled devices and wireless communications technologies, spatial trajectories that describe the movement history of moving objects are being generated and accumulated at an unprecedented pace. Trajectory data in a database are intrinsically heterogeneous, as they represent discrete approximations of original continuous paths derived using different sampling strategies and different sampling rates. Such heterogeneity can have a negative impact on the effectiveness of trajectory similarity measures, which are the basis of many crucial trajectory processing tasks. In this paper, we pioneer a systematic approach to trajectory calibration that is a process to transform a heterogeneous trajectory dataset to one with (almost) unified sampling strategies. Specifically, we propose an anchor-based calibration system that aligns trajectories to a set of anchor points, which are fixed locations independent of trajectory data. After examining four different types of anchor points for the purpose of building a stable reference system, we propose a spatial-only geometry-based calibration approach that considers the spatial relationship between anchor points and trajectories. Then a more advanced spatial-only model-based calibration method is presented, which exploits the power of machine learning techniques to train inference models from historical trajectory data to improve calibration effectiveness. Afterward, since trajectory has temporal information, we extend these two spatial-only trajectory calibration algorithms to incorporate the temporal information, which can infer a proper time stamp to each anchor point of a calibrated trajectory. At last, we provide a solution to reduce cost, i.e., the number of trajectories that is necessary to be re-calibrated, of the updating of the reference system. Finally, we conduct extensive experiments using real trajectory datasets to demonstrate the effectiveness and efficiency of the proposed calibration system.

Route recommendation service has become a big business in industry since traveling is now an important part of our daily life. We can travel to unknown places by simply typing in our destination and then following recommendation service's guidance, that a pleasant trip desires them to provide a good route. However, previous research shows that even the routes recommended by the big-thumb service providers can deviate significantly from the routes travelled by experienced drivers since the many latent factors affect drivers' preferences and it is hard for a single route recommendation algorithm to model all of them. In this demo we will present the CrowPlanner system to leverage crowds' knowledge to improve the recommendation quality. It requests human workers to evaluate candidates routes recommended by different sources and methods, and determines the best route based on the feedbacks of these workers. In this demo, we first introduce the core component of our system for smart question generation, and then show several real route recommendation cases and the feedback of users.

The efficiency of spatial query processing is crucial for many applications such as location-based services. In spatial networks, queries like k-NN queries are all based on network distance evaluation. Classic solutions for these queries rely on network expansion and are not efficient enough for large networks. Some approaches have improved the query efficiency but brought considerable space cost for index. To address these problems, we propose a hierarchical graph partitioning based index named Partition Tree. It organizes the vertices of a spatial network into a hierarchy through a series of graph partitioning processes. Meanwhile precomputed distances are associated with this hierarchy to facilitate efficient query processing. Inspired by the observation that queries are usually invoked around objects of interest, we propose a query-oriented optimization on top of the Partition Tree. It uses a cost model to evaluate the influence of the object distribution and partitioning topology on the query efficiency. Then a cost-efficient graph partitioning method is developed based on this cost model. Experimental results on real datasets demonstrate that our proposed index and algorithms have superior performance over the state-of-the-art approaches and are scalable to large spatial networks.

The last decade has witnessed the prevalence of sensor and GPS technologies that produce a high volume of trajectory data representing the motion history of moving objects. However some characteristics of trajectories such as variable lengths and asynchronous sampling rates make it difficult to fit into traditional database systems that are disk-based and tuple-oriented. Motivated by the success of column store and recent development of in-memory databases, we try to explore the potential opportunities of boosting the performance of trajectory data processing by designing a novel trajectory storage within main memory. In contrast to most existing trajectory indexing methods that keep consecutive samples of the same trajectory in the same disk page, we partition the database into frames in which the positions of all moving objects at the same time instant are stored together and aligned in main memory. We found this column-wise storage to be surprisingly well suited for in-memory computing since most frames can be stored in highly compressed form, which is pivotal for increasing the memory throughput and reducing CPU-cache miss. The independence between frames also makes them natural working units when parallelizing data processing on a multi-core environment. Lastly we run a variety of common trajectory queries on both real and synthetic datasets in order to demonstrate advantages and study the limitations of our proposed storage.

This paper addresses the problem of searching the k aggregate farthest neighbours (AkFN query in short) on road networks. Given a query point set, AkFN is aimed at finding the top-k points from a dataset with the largest aggregate network distance. The challenge of the AkFN query on the road network is how to reduce the number of network distance evaluation which is an expensive operation. In our work, we propose a three-phase solution, including clustering points in dataset, network distance bound pre-computing and searching. By organizing the objects into compact clusters and pre-calculating the network distance bound from clusters to a set of reference points, we can effectively prune a large fraction of clusters without probing each individual point inside. Finally, we demonstrate the efficiency of our proposed approaches by extensive experiments on a real Point- of-Interest (POI) dataset.

The last decade has witnessed the prevalence of sensor and GPS technologies that produce a high volume of trajectory data representing the motion history of moving objects. However some characteristics of trajectories such as variable lengths and asynchronous sampling rates make it difficult to fit into traditional database systems that are disk-based and tuple-oriented. Motivated by the success of column store and recent development of in-memory databases, we try to explore the potential opportunities of boosting the performance of trajectory data processing by designing a novel trajectory storage within main memory. In contrast to most existing trajectory indexing methods that keep consecutive samples of the same trajectory in the same disk page, we partition the database into frames in which the positions of all moving objects at the same time instant are stored together and aligned in main memory. We found this column-wise storage to be surprisingly well suited for in-memory computing since most frames can be stored in highly compressed form, which is pivotal for increasing the memory throughput and reducing CPU-cache miss. The independence between frames also makes them natural working units when parallelizing data processing on a multi-core environment. Lastly we run a variety of common trajectory queries on both real and synthetic datasets in order to demonstrate advantages and study the limitations of our proposed storage.

Due to the prevalence of GPS-enabled devices and wireless communications technologies, spatial trajectories that describe the movement history of moving objects are being generated and accumulated at an unprecedented pace. Trajectory data in a database are intrinsically heterogeneous, as they represent discrete approximations of original continuous paths derived using different sampling strategies and different sampling rates. Such heterogeneity can have a negative impact on the effectiveness of trajectory similarity measures, which are the basis of many crucial trajectory processing tasks. In this paper, we pioneer a systematic approach to trajectory calibration that is a process to transform a heterogeneous trajectory dataset to one with (almost) unified sampling strategies. Specifically, we propose an anchor-based calibration system that aligns trajectories to a set of anchor points, which are fixed locations independent of trajectory data. After examining four different types of anchor points for the purpose of building a stable reference system, we propose a geometry-based calibration approach that considers the spatial relationship between anchor points and trajectories. Then a more advanced model-based calibration method is presented, which exploits the power of machine learning techniques to train inference models from historical trajectory data to improve calibration effectiveness. Finally, we conduct extensive experiments using real trajectory datasets to demonstrate the effectiveness and efficiency of the proposed calibration system.

The last decade has witnessed the prevalence of sensor and GPS technologies that produce a sheer volume of trajectory data representing the motion history of moving objects. Measuring similarity between trajectories is undoubtedly one of the most important tasks in trajectory data management since it serves as the foundation of many advanced analyses such as similarity search, clustering, and classication. In this light, tremendous eorts have been spent on this topic, which results in a large number of trajectory similarity measures. Generally, each individual work introducing a new distance measure has made specic

The last decade has witnessed the prevalence of sensor and GPS technologies that produce a sheer volume of trajectory data representing the motion history of moving objects. Measuring similarity between trajectories is undoubtedly one of the most important tasks in trajectory data management since it serves as the foundation of many advanced analyses such as similarity search, clustering, and classification. In this light, tremendous efforts have been spent on this topic, which results in a large number of trajectory similarity measures. Generally, each individual work introducing a new distance measure has made specific claims on the superiority of their proposal. However, for most works, the experimental study was focused on demonstrating the efficiency of the search algorithms, leaving the effectiveness aspect unverified empirically. In this paper, we conduct a comparative experimental study on the effectiveness of six widely used trajectory similarity measures based on a real taxi trajectory dataset. By applying a variety of transformations we designed for each original trajectory, our experimental observations demonstrate the advantages and drawbacks of these similarity measures in different circumstances.