Research

Current Research Areas

My current research interests are principally in the following areas:

• Mobile And Wearable Computing (activity and gesture recognition, IoT-based sensing, wireless energy harvesting, novel wearable interfaces)  More...

• Urban And Socio-Physical Analytics and Smart City Services (spatiotemporal data mining, urban event and anomaly detection) Less...

Cities like Singapore have recently initiated heavy investments in smart city infrastructure, resulting in the generation of, and access to, unprecedented levels of urban informatics data. Examples of such data include cameras on highways (that upload real-time images of traffic), sensors on buses & taxis (that report their location and occupancy levels every few minutes) and social media content (via which city residents share text, video and images of their surrondings). In this broad body of work, we tackle the question: what types of insights, about a city’s neighborhoods and businesses, can be glean by fusing these diverse data sources? Selected activities include:

·       Multimodal Event Detection: We have been exploring the combined use of social media and urban sensing data (e.g., from buses, taxis and parking garages) to detect and spatiotemporally localize urban micro-events. The work broadly uses anomaly determination approaches to identify outliers across multiple such information sources. While this work is still in its early stages, we have shown [paper1] that different sensing modes have different discriminative capabilities—e.g., social media streams exhibit outliers shortly before or during an event, whereas transportation data sources exhibit more persistent outliers. More recently, we have further analyzed [paper2] the event-driven disruptions, as observed via different data sources (bus loading levels, telecom call records and taxi trips), and shown the importance of fusing the outliers detected across each source.

Predictive Mobile Crowdsourcing: Over the last few years, mobile crowdsourcing, where a pool of at-will workers performs location-specific micro-tasks, has created disruptions in many urban services—including transportation (e.g., Uber) and last-mile package delivery (e.g., Amazon Flex). Broadly speaking, this body of research is driven by a central question: Can such mobile crowdsourcing services be made more effective by better leveraging the predicted movement path and the behavioral preferences of workers? To improve worker productivity, we pioneered a “push” model of crowdsourcing [paper1, paper2], where the crowdsourcing platform proactively recommends tasks that maximize the task completion rate while minimizing a worker’s detour from her routine movement trajectory. To support experimental investigations into worker behavior during crowdsourcing, we have developed and deployed Ta$ker, a campus-based experimental mobile crowdsourcing platform used over 1000 student “workers”, on the SMU campus. Ta$ker has enabled us to develop and empirically validate a variety of crowd-sourcing related technologies: e.g., the use of task bundling [paper] to improve worker productivity, the use of dynamic peer offloading [paper] to significantly improve the task completion rate, and the development of location obfuscation strategies [paper] that enhance worker location privacy with only modest impact on productivity. 

·       Neighborhood Economic & Mobility Dynamics: In collaboration with researchers from Cambridge University, we explored  how to combine data from location-based social networks with urban transportation data, to first profile the commuter interaction pattern with different neighborhoods and their shops, and thereby predict the vibrancy and likely failure propensity of individual retail businesses. As part of this broader area, we are also collaborating with public agencies to apply similar analyses to understand the utilization and catchment of retail outlets, predict their viability and their likely impact on traffic patterns in individual neighborhoods. Overall, this body of work opens up new forms of smarter urban planning, driven by insights into neighborhood-level human mobility and behavioral dynamics.