DescriptionOne area of rapid growth in transportation systems is smart mobility services and connected autonomous vehicles. These technological achievements have led to the development of new modes of travel and data sources which are decentralized, on-demand and dynamic – they stand to benefit the most from data-driven analysis and deep learning. This seminar presentation explores how we can capture and model changes in travel behaviour by leveraging on recent deep learning optimization and modelling techniques. We developed a new framework connecting deep learning to discrete choice models by representing the unobserved heterogeneity using deep residual layers - shortcut connections that capture the correlation patterns in behavioural choice models. This lends the model the ability to estimate increasingly complex models while retaining key information about the model parameters and individual behaviour. We further discuss how an implementation of deep learning optimization in discrete choice modelling might have an effect on economic interpretability.
|Period||21 Oct 2019|
|Held at||Ecole Polytechnique Fédérale de Lausanne, Switzerland|
|Degree of Recognition||Local|