Select a calendar:
Filter August Events by Event Type:
SUNMONTUEWEDTHUFRISAT
University Calendar
Events for August
-
PhD Defense - Dingxiong Deng
Wed, Aug 16, 2017 @ 01:00 PM - 03:00 PM
Thomas Lord Department of Computer Science
University Calendar
PhD Candidate: Dingxiong Deng
Committee: Cyrus Shahabi (Chair), Craig Knoblock and Ketan Savla.
Title: Spatiotemporal Traffic Forecasting in Road Networks
Time: August 16 (Wednesday) 1:00-3:00pm
Location: SAL 213
Abstract:
Real-time traffic forecasting from high-fidelity spatiotemporal traffic sensor datasets is an important problem for intelligent transportation systems and sustainability. However, it is challenging due to the complex topological dependencies and high dynamism associated with changing road conditions.
To address these challenges, we explore two different methods of incorporating the spatiotemporal correlations between sensors in the road network. We first propose a Latent Space Model for Road Networks (LSM-RN) that combines the spatial and temporal correlations of sensors. In particular, given a series of road network snapshots, we learn the attributes of sensors in latent spaces to estimate how traffic patterns are formed and evolved. We present an incremental online algorithm which sequentially and adaptively learns the latent attributes from the temporal graph changes, thus enabling real-time traffic forecasting. However, LSM-RN only utilizes the most recent graph snapshots as inputs and does not distinguish the underlying traffic situations, hence it does not perform well in long term traffic forecasting. To address this issue, we further explore the commonalities across multiple traffic sensors who behave the same in a specific traffic situation. We show that building models based on the shared traffic situations across sensors can help improve the prediction accuracy. We propose a Multi-Task Learning (MTL) framework that aims to first automatically identify the traffic situations and then simultaneously build one forecasting model for similar behaving sensors per traffic situation. We demonstrated that our proposed framework outperforms all the best traffic prediction approaches in both short and long term predictions, respectively.
Location: Henry Salvatori Computer Science Center (SAL) - 213
Audiences: Everyone Is Invited
Contact: Lizsl De Leon
