Authors: Sean Ahearn*, Hunter College - City University
Topics: Geographic Information Science and Systems, Quantitative Methods
Keywords: trajectory, Recurrent Neural Network,
Session Type: Paper
Start / End Time: 1:45 PM / 3:00 PM
Room: Governors Square 10, Sheraton, Concourse Level
Presentation File: No File Uploaded
The developments in AI, especially Deep Learning, have radically changed the way we conceptualize and solve analytic problems in image classification, voice recognition and translation, to name a few. This revolution is the result of a convergence of access to vast amounts of data for training, advancements in Neural Network architecture, faster computing and new approaches to learning systems. A key question is: how do these new developments relate to geographic problems and analysis which deal with space, time and scale? In the analysis of trajectories, the Recurrent Neural Network (RNN) appears to be the most relevant NN architecture for analyzing and modeling trajectories in space and time. The RNN handles sequences and can retain contextual information but unlike the Hidden Markov Model, it doesn’t assume a Markov condition where the current state is solely dependent upon the previous state. The RNN’s predominant use has been Natural Language Processing, however, recently it has been applied to spatial temporal trajectories. This talk will explore how this particular type of NN can apply to trajectory analysis and how it compares with existing approaches.