交通规划模型的机器学习：基于出行选择图（TCG）的方法.pdf

1、Machine Learning for Transportation Planning Models:A Travel Choice Graph(TCG)-Based ApproachPresented by:Mohammad Abbasi,Ph.D.Arizona State UniversityAuthors Arizona State University Mohammad Abbasi,Xuesong(Simon)Zhou Acknowledgements Xin Wu(Postdoctoral Researcher at Villanova University)Ram Pendy

2、ala(School Director and Professor at Arizona State University)Taehooie Kim(Research Scientist at UrbanSim Inc.)Conference on Innovations in Travel Analysis and PlanningMotivations1.How to develop a unified modeling structure to seamlessly integrate traditionalstatistical inference models and new mac

3、hine learning techniques?2.How to adapt automatic differentiation methods(which are building blocks ofML)to greatly enhance the computational efficiency of large-scale econometricmodel estimation to better uncover complex behavioral parameters and latentvariables?3.How to theoretically define the co

4、nsistency of transportation demand-side andtraffic supply-side models,while providing numerically stable solutionsconsistent with multi-data sources from real-world systems?Conference on Innovations in Travel Analysis and PlanningBasic Concepts(4-step with different choice sets)Conference on Innovat

5、ions in Travel Analysis and Planning4-step process in transportation modelingBackground of artificial neural networksConference on Innovations in Travel Analysis and PlanningBrief history of deep learningMcCulloch-Pitts(1943)Neuron of brain functionRumelhart et al.(1986)Multi-layer distributed repre

6、sentationBack propagation algorithmHinton et al.(2006)The last ten years Convolutional networksRosenblatt(1956)Perceptron ADA-LINEStochastic gradient descentFigure source:http:/galaxy.agh.edu.pl/vlsi/AI/backp_t_en/backprop.htmlOur vision:Connected Travel Choice GraphConference on Innovations in Trav