Artificial I Intelligence f for S Smart Transp sportation Yan Liu Associate Professor Computer Science Department University of Southern California Yan Liu (USC) ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ICML Time Series Workshop 1
AI and Machine Learning ? Neural Networks Deep Learning Machine Learning: Reinforcement supervised, Learning unsupervised Yan Liu (USC) ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ICML Time Series Workshop 2
GPS Data Location Data and Floating-Car Trajectory Yan Liu (USC) ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ICML Time Series Workshop 3
Sensors Loop detector, camera, microphone, mobile sensors … Yan Liu (USC) ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ICML Time Series Workshop 4
Transportation AI Big data makes AI possible for transportation. Yan Liu (USC) ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ICML Time Series Workshop 5
Smart Transportation Brain Ride-sharing Company Data � � Route Planning Route Planning Government Data ETA Map Data Collaborators’ Data Pick-up locations Services Collection � � Crowd Sourced Data VR Navigation Demand-Supply Prediction Performance Measures D S e u Order Dispatch Congestion Diagnosis Platform m p Car Pooling Network Design Analysis Optimization p a Resource Allocation Traffic Simulation l n Multi-modal Accident Analysis y d Taxi Signal Control Freeway Control Express Ride-sharing Control Traffic Guidance Car Pool Services Premiere Incident Management …… AI Dispatch Yan Liu (USC) ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ICML Time Series Workshop 6
Outline • Traffic estimation and forecasting • Li et al. Diffusion Convolutional Recurrent Neural Network: Data-driven Traffic Forecasting, ICLR 2018 • Demand forecasting • Li et al, Spatiotemporal Multi-Graph Convolution for Ride-hailing Demand Forecasting, AAAI 2019 • Multi-rate multi-resolution forecasting/interpolation • Che et al, Hierarchical Deep Generative Models for Multi-Rate Multivariate Time Series, ICML 2018 Yan Liu (USC) ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ICML Time Series Workshop 7
Traffic Prediction • Input: road network and past T’ traffic speed observed at sensors • Output: traffic speed for the next T steps Output: Predictions Input: Observations ... 7:00 AM ... 8:00 AM 8:10AM, 8:20AM, …, 9:00 AM Yan Liu (USC) ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ICML Time Series Workshop 8
Existing Work • KNN-based models • Time series models • Seasonal Autoregressive Integrated Moving Average (S-ARIMA) • Support vector regression • Our prior work: • Latent space models: Dingxiong Deng et al, Latent Space Model for Road Networks to Predict Time-Varying Traffic. KDD, 2016 • Mixture LSTM: Y. Qi et al, Deep Learning: A Generic Approach for Extreme Condition Traffic Forecasting. SDM 2016 Yan Liu (USC) ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ICML Time Series Workshop 9
Challenges for Traffic Forecasting Complex Non-linear, non-stationary Spatial Dependency Temporal Dynamic Speed (mile/h) Yan Liu (USC) ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ICML Time Series Workshop 10
Challenges for Traffic Forecasting • Spatial relationship among traffic flow is non-Euclidean and directed Yan Liu (USC) ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ICML Time Series Workshop 11
Traffic Forecasting with Convolution on Graph • Model spatial dependency with proposed diffusion convolution on graph * Yagu guang g Li et al, Diffusion Convolutional Recurrent Neural Network: Data-dr driven n Traffic Forecasting ng. ICLR, 2018 Yan Liu (USC) ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ICML Time Series Workshop 12
Spatial Dependency in Traffic Prediction • Spatial dependency among traffic flow is no non-Eucl clidean and direct cted Close in Similar Euclidean traffic Sensor 1 Sensor 2 space speed !"#$ %&' ( ) → ( + ≠ !"#$ %&' ( ) → ( + Sensor 3 Yan Liu (USC) ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ICML Time Series Workshop 13
Spatial Dependency Modeling • Model the network of traffic sensors, i.e., loop detectors, as a directed graph • Graph ! = ($, &) • Vertices ( : o sensors • Adjacency matrix & : → weight between vertices 9 * +, = exp − dist 567 8 + , 8 , if dist 567 8 + , 8 , ≤ = : 9 dist 567 8 + , 8 , : road network distance from 8 + to 8 , , = : threshold to ensure sparsity, : 9 variance of all pairwise road network distances Yan Liu (USC) ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ICML Time Series Workshop 14
Problem Statement • Graph signal: ! " ∈ ℝ |&|×( , observation on ) at time * • + : number of vertices • , : feature dimension of each vertex. • Problem Statement : Learn a function -(·) to map 1 2 historical graph signals to future 1 graph signals ! 4 ! 489 ! 486 ! 456 7 89 - . … … Yan Liu (USC) ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ICML Time Series Workshop 15
Generalize Convolution to Graph • Diffusion convolution filter: combination of diffusion processes with different steps on the graph. Transition matrices of the diffusion process %2# 0 & :,) Learning complexity: 7 8 25 6 & :,) ⋆ + , - = / ! 0 3 4 01" + ! # + ! $ + … + ! % = ! " Example diffusion filter 0 Step 1 Step 2 Step K Step Centered at Diffusion Diffusion Diffusion Diffusion Min Max Filter weight ⋆ + : diffusion convolution, 9 : : diagonal out-degree matrix. Yan Liu (USC) ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ICML Time Series Workshop 16
Generalize Convolution to Graph • Diffusion convolution filter: combination of diffusion processes with different steps on the graph. Dual directional diffusion to model upstream and downstream separately %2# 0 + ! 0,$ 3 8 25 6 ⊺ 0 & :,) 25 6 & :,) ⋆ + , - = / ! 0,# 3 4 01" + ! # + ! $ + … + ! % = ! " Example diffusion filter 1 Step 2 Step 0 Step K Step Centered at Diffusion Diffusion Diffusion Diffusion Min Max ⋆ + : diffusion convolution, : ; : diagonal out-degree matrix, : < : diagonal in-degree matrix weight Yan Liu (USC) ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ICML Time Series Workshop 17
Advantage of Diffusion Convolution 123 . + 4 .,: 5 ; 27 8 ⊺ . % :,' 27 8 % :,' ⋆ ) * + = - 4 .,3 5 6 ./0 • Efficient • Learning complexity: ! " • Time complexity: ! " # , # number of edges • Expressive • Many popular convolution operations, including the ChebNet [Defferrard et al., NIPS ’16], can be seen as special cases of the diffusion convolution [Li et al. ICLR ’18]. ⋆ ) : diffusion convolution, = > : diagonal out-degree matrix, = ? : diagonal in-degree matrix * Defferrard, M et al, Convolutional Neural Networks on Graphs with Fast Localized Spectral Filtering, NIPS, 2016 * Yaguang Li et al. Diffusion Convolutional Recurrent Neural Network: Data-driven Traffic Forecasting, ICLR, 2018 Yan Liu (USC) ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ICML Time Series Workshop 18
Diffusion Convolutional Recurrent Neural Network • Diffusion Convolutional Recurrent Neural Network (DCRNN) • Model spatial dependency with diffusion convolution • Sequence to sequence learning with encoder-decoder framework * Yaguang Li et al. Diffusion Convolutional Recurrent Neural Network: Data-driven Traffic Forecasting, ICLR 2018 Yan Liu (USC) ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ARTIFICIAL INTELLIGENCE FOR SMART TRANSPORTATION ICML Time Series Workshop 19
Recommend
More recommend
