Wei Sun’s research while affiliated with IBM and other places

Moving objects, for example vehicles, vessels, people, are being instrumented with Internet of Things devices such as GPS, environmental sensors, mobile cameras that produce big streaming data with spatial-temporal features enabling application innovations in many industries for autonomous driving, usage based insurance, fleet management, etc. We introduce an analytics framework to streamline the insights discovery from such big data on top of geographic map by pulling together a list of commonly required high performance functions with service-oriented architecture. Supported by a service composition approach, comprehensive applications intelligently managing moving objects could be built effectively. Example applications in connected vehicle domains are presented to illustrate the effectiveness of such a framework.

A method and apparatus controls use of a computing resource by multiple tenants in DBaaS service. The method includes intercepting a task that is to access a computer resource, the task being an operating system process or thread; identifying a tenant that is in association with the task from the multiple tenants; determining other tasks of the tenant that access the computing resource; and controlling the use of the computing resource by the task, so that the total amount of usage of the computing resource by the task and the other tasks does not exceed the limit of usage of the computing resource for the tenant.

A method and apparatus for adjusting wheel diameter by acquiring a current wheel diameter distribution vector; calculating similarity scores between the current wheel diameter distribution vector and known wheel diameter distribution vector clusters to determine a potential wear rate of the current wheel diameter distribution vector; calculating similarity scores between candidate target wheel diameter distribution vectors and the known wheel diameter distribution vector clusters to determine a potential wear rates of the candidate target wheel diameter distribution vectors; and selecting a target wheel diameter distribution vector from the candidate target wheel diameter distribution vectors, such that the reduction of potential wear rate achieved by switching the current wheel diameter distribution vector to the target wheel diameter distribution vector meets a predetermined condition. Future potential wear rates and the amount of grinding required during train maintenance can be optimized through changing wheel diameter distribution vectors.

An apparatus for tuning a parameter of a Kalman filter in a wheel inspection for a vehicle is disclosed. The apparatus includes an association module that associates wheel inspection data of the vehicle with locations of corresponding wheels in the vehicle; an evaluation module that evaluates a wheel stable score of the vehicle based on the wheel inspection data and the association, where the wheel stable score indicates reliability of the wheel inspection; and a parameter tuning module that tunes a measurement error covariance of the Kalman filter according to the evaluated wheel stable score.

A system, method and computer program product for mining a rule including spatial information and non-spatial information by using a SAR (Spatial Association Rule) mining tool. The computing system is configured to construct an expanded spatial predicate transaction table for reference spatial objects and a generalized taxonomy for task-relevant spatial objects. The computing system is configured to run the SAR mining tool with the constructed expanded spatial predicate transaction and the generalized taxonomy. The computing system outputs, from the SAR mining tool, a set of generalized spatial association rules for the reference spatial objects. The generalized spatial association rule includes the spatial information and non-spatial information, associated with both the reference spatial objects and the task-relevant spatial objects.

A database system and a method of optimizing cross-database query. The database system includes a plurality of databases. At least two of the databases are stored with one or more data collections composed of tables with the same structure. A federated view of the data collection is created on each of the databases. A request routing layer for routing, in response to a query request crossing the data collections, the query request crossing the data collections to one of the at least two databases according to a predetermined routing rule, so as to query by using the federated view of the database.

A method and apparatus for transferring resources from a server to a client, wherein the resources are for an object in an application. According to the present invention, resources needed by the object are determined. Resources which have usages that are less than a threshold according to statistics are removed from the determined resources. The remainder of the determined resources, which are needed by the object, is transferred to the client. According to the present invention, users do not have to wait for a long time before they can see a web page.

A method, apparatus, and a computer program product for storing and accessing multi-tenancy data. The method includes the steps of: creating a plurality of table sets in one or more databases, wherein each table set is used to store data of a group of tenants selected from a plurality of tenants; accessing data of a tenant in a table set in response to receiving a data access request from the tenant; and recording relationships between the tenants and the table sets in a multi-tenancy metadata repository, wherein the step of accessing the data of the tenant comprises the steps of finding the table set by querying the metadata repository and accessing the data of the tenant in the table set based on the result received from the query of the metadata repository.

Criminal activities are a major risk factor for the well-being of society in many countries. Police patrol service is a critical instrument to combat the criminal activities with violent aspects. The police resource allocation for street patrolling is one of the most important tactical management activities, and it is important to continuously improve the patrolling strategies. Previous research mainly focuses on determining the important scores of locations based on hotspots analysis or identifying important routes based on the topology of road networks. There are several limitations in the current patrol planning studies. First, the patterns in the distribution of hotspots were rarely considered in the patrol planning. Second, some existing patrol optimization algorithms can lead to a predictable patroller. Third, the overall performance of multiple patrol activities is often not optimized. Finally, more efficient algorithms are needed for a real-time solution that serves large jurisdictions. To deal with these limitations, this chapter aims to integrate a spatial pattern identification approach with an efficient route optimization algorithm to produce randomized optimal patrol routes. A case study is provided to illustrate our approach, which shows that the proposed approach can improve the overall effectiveness of multiple unpredictable patrol activities and it is efficient enough to be used as a real-time solution.

Law enforcement agencies often rely on crime pattern identification techniques to support their tactical planning. K-function analysis has been one of the most popular crime pattern identification approaches. It has been integrated with point randomization procedures to identify the level of clustering in crimes. One limitation of this integration is that it can only differentiate between a complete random pattern and a clustered point pattern. It is well known that crimes only occur in populated area and the distribution of human population is spatially heterogeneous. A complete random pattern of crimes rarely occurs. The current K-function offers little insights on the clustering levels of crimes given our prior knowledge on the distribution of processes that may have influenced the occurrence of crimes. This study integrates two controlled point randomization procedures with K-function analysis to analyze crime patterns. These two approaches are compared against the complete random pattern and results indicate that the controlled point randomization procedures can reveal detailed information on the underlying processes for point patterns. It can also take into account the underlying processes for the crimes.