I am an Associate Professor of Electrical Engineering and Computer Science (EECS) at the Lassonde School of Engineering, York University. I am faculty member of the Data Mining Lab.
- Trajectory Data Mining: Advances in location acquisition and tracking devices have given rise to the generation of enormous trajectory data consisting of spatial and temporal information of moving objects, such as persons, vehicles or animals. The primary focus of our research is on discovery of network patterns and dynamics through mining trajectory data streams. This describes a special type of trajectory mining task that seeks to efficiently discover pair-wise relationships (interactions) among moving objects over time. Mining trajectory data streams to find interesting network patterns is of increased research interest due to a broad range of useful applications, including analysis of transportation systems and location-based social networks.
- Network Representation Learning: With a growing number of networks – social, technological, biological – becoming available and representing an ever increasing amount of information, the ability to easily and effectively perform large-scale network mining and analysis is key to revealing the underlying dynamics of these networks, not easily observable before. Traditional approaches to network mining and analysis inherit a number of limitations; typically algorithms don't scale well (due to ineffective representation of network data) and require domain-expertise. More recently and to address the aforementioned limitations, there is a fast-growing interest in learning low-dimensional and continuous representations of networks. Representing networks into low-dimensional spaces occurs in an agnostic way (without domain-expertise) and has the potential to improve the performance of many data mining tasks that now need to operate in lower dimensions. Network mining can support a variety of applications in diverse disciplines and has the potential to impact different industries.
- Streaming & Dynamic Graph Mining: Large-scale graph mining and analysis is key to revealing the underlying dynamics of networks, not easily observable before. The conventional computational approach for performing graph/network analysis assumes there is a static network topology (and/or data) that is provided as input to a graph algorithm, which (always) terminates (i.e., produces an outcome or fails). Analyzing massive graphs via classical algorithms casts its own unique challenges (e.g, memory/time overhead), but the conventional approach is mostly insufficient for many modern data processing needs. Over the last years, there has been considerable interest in designing algorithms for processing graphs in the data stream model, where the input is defined by a stream of graph data (e.g., a stream of edges), and the graph algorithm, aware of these changes, must be able to accept the changes faster than a naive re-computation of an algorithm on static graphs. Algorithms in this model must operate under specific constraints: (i) the input stream must be processed in the order it arrives, and (ii) the processing can only use a limited amount of memory.