DTU Management Engineering
August 23, 2018, 15:00, Room GC B110 (click here for the map)
For multiple reasons, including little data and computing power, and a traditionalist community, representations of models and people (or, "agents") in transport demand behavior have changed very little through the decades. Each agent is represented as a vector of characteristics, and each model is a function that combines such characteristics with contextual information (e.g. travel time and cost for different transport options, in a mode choice model). However, new paradigms exist that have been under-explored, including networked representations, logic programming, deep representations, natural language based. It is well-known that many scientific breakthroughs in history have come up from "seeing things in a different perspective". It is my belief that strong opportunities in re-representing travel behavior, particularly considering current grand challenges of causality, transferability, new data types, social interactions, and many more. In this talk, I give some examples of the above from earlier work of mine and others, in order to present a few ideas and on-going work. Ultimately, I want to stimulate an exciting discussion towards collaboration between DTU and EPFL on these pressing and, I will argue, very promising opportunities!
Francisco Pereira is Professor at the Technical University of Denmark (DTU) since August 2015, where he leads the Machine Learning for Mobility (MLM) group. MLM works on real-time traffic prediction, behavior modeling, advanced data collection technologies, big data and transport modelling. Previously, he was Senior Research Scientist at the MIT ITS Lab, based in both Boston and Singapore, as part of the Singapore-MIT Alliance for Research and Technology, Future Urban Mobility project (SMART/FM). He has a Masters and Phd from University of Coimbra, Portugal, on Computer Engineering and Artificial Intelligence. His research focus is on applying machine learning and pattern recognition to the context of transportation systems with the purpose of understanding and predicting mobility behavior, and optimizing the transportation system as a whole, but also on using concepts and methodologies from transportation (e.g. behavior modeling) to develop new machine learning research.