Machine learning for transportation research and applications /

Transportation is a combination of systems that presents a variety of challenges often too intricate to be addressed by conventional parametric methods. Increasing data availability and recent advancements in machine learning provide new methods to tackle challenging transportation problems. This te...

Full description

Bibliographic Details
Main Authors:	Wang, Yinhai (Author), Cui, Zhiyong, 1989- (Author), Ke, Ruimin (Author)
Corporate Author:	ScienceDirect (Online service)
Format:	eBook
Language:	English
Published:	Amsterdam : Elsevier, 2023.
Subjects:	Transportation > Data processing. Machine learning. Transport > Informatique. Apprentissage automatique. Machine learning Transportation > Data processing Electronic books.
Online Access:	Connect to the full text of this electronic book

Table of Contents:

Front Cover
Machine Learning for Transportation Research and Applications
Copyright
Contents
About the authors
1 Introduction
1.1 Background
1.1.1 Importance of transportation
1.1.2 Motivation
1.2 ML is promising for transportation research and applications
1.2.1 A brief history of ML
1.2.2 ML for transportation research and applications
1.3 Book organization
2 Transportation data and sensing
2.1 Data explosion
2.2 ITS data needs
2.3 Infrastructure-based data and sensing
2.3.1 Traffic flow detection
2.3.2 Travel time estimation
2.3.3 Traffic anomaly detection
2.3.4 Parking detection
2.4 Vehicle onboard data and sensing
2.4.1 Traffic near-crash detection
2.4.2 Road user behavior sensing
2.4.3 Road and lane detection
2.4.4 Semantic segmentation
2.5 Aerial sensing for ground transportation data
2.5.1 Road user detection and tracking
2.5.2 Advanced aerial sensing
2.5.3 UAV for infrastructure data collection
2.6 ITS data quality control and fusion
2.7 Transportation data and sensing challenges
2.7.1 Heterogeneity
2.7.2 High probability of sensor failure
2.7.3 Sensing in extreme cases
2.7.4 Privacy protection
2.8 Exercises
3 Machine learning basics
3.1 Categories of machine learning
3.1.1 Supervised vs. unsupervised learning
3.1.2 Generative vs. discriminative algorithms
3.1.3 Parametric vs. nonparametric modeling
3.2 Supervised learning
3.2.1 Linear regression
Problem setup
Solving the optimization problem
Vectorization
3.2.2 Logistic regression
Softmax regression
3.3 Unsupervised learning
3.3.1 Principal component analysis
3.3.2 Clustering
3.4 Key concepts in machine learning
3.4.1 Loss
3.4.2 Regularization
L1 vs. L2
3.4.3 Gradient descent vs. gradient ascent.
3.4.4 K-fold cross-validation
3.5 Exercises
3.5.1 Questions
4 Fully connected neural networks
4.1 Linear regression
4.2 Deep neural network fundamentals
4.2.1 Perceptron
4.2.2 Hidden layers
4.2.3 Activation functions
Sigmoid function
Tanh function
ReLU function
4.2.4 Loss functions
4.2.5 Back-propagation
Forward propagation
Backward propagation
4.2.6 Validation dataset
4.2.7 Underfitting or overfitting?
4.3 Transportation applications
4.3.1 Traffic prediction
4.3.2 Traffic sign image classification
4.4 Exercises
4.4.1 Questions
5 Convolution neural networks
5.1 Convolution neural network fundamentals
5.1.1 From fully connected layers to convolutions
5.1.2 Convolutions
5.1.3 Architecture
5.1.4 AlexNet
5.2 Case study: traffic video sensing
5.3 Case study: spatiotemporal traffic pattern learning
5.4 Case study: CNNs for data imputation
5.4.1 CNN-based imputation approach
5.4.2 Experiment
5.5 Exercises
6 Recurrent neural networks
6.1 RNN fundamentals
6.2 RNN variants and related architectures
6.2.1 Long short-term memory (LSTM) and gated recurrent units (GRU)
6.2.2 Bidirectional RNN
6.2.3 Sequence to sequence
6.3 RNN as a building block for transportation applications
6.3.1 RNN for road traffic prediction
Problem description
Network-wide traffic prediction
Traffic prediction algorithms
6.3.2 Traffic prediction with missing values
Problem definition
LSTM-based traffic prediction with missing values
6.4 Exercises
6.4.1 Questions
6.4.2 Project: predicting network-wide traffic using LSTM
Problem definition
Dataset preparation
Implement and fine-tune model
Model evaluation
7 Reinforcement learning
7.1 Reinforcement learning setting
7.1.1 Markov property.
7.1.2 Goal of reinforcement learning
7.1.3 Categories and terms in reinforcement learning
Model-free vs. model-based
Stationary vs. nonstationary
Deterministic policy vs. stochastic policy
Offline learning vs. online learning
Exploration vs. exploitation
Off-policy learning vs. on-policy learning
7.2 Value-based methods
7.2.1 Q-learning
7.2.2 Deep Q-networks
7.3 Policy gradient methods for deep RL
7.3.1 Stochastic policy gradient
7.3.2 Deterministic policy gradient
7.4 Combining policy gradient and Q-learning
7.4.1 Actor-critic methods
7.5 Case study 1: traffic signal control
7.5.1 Agent formulation
7.6 Case study 2: car following control
7.6.1 Agent formulation
7.6.2 Model and simulation settings
7.7 Case study 3: bus bunching control
7.7.1 Agent formulation
7.8 Exercises
7.8.1 Questions
8 Transfer learning
8.1 What is transfer learning
8.2 Why transfer learning
8.3 Definition
8.4 Transfer learning steps
8.5 Transfer learning types
8.5.1 Domain adaptation
8.5.2 Multi-task learning
8.5.3 Zero-shot learning
8.5.4 Few-shot learning
8.6 Case study: vehicle detection enhancement through transfer learning
8.7 Case study: parking information management and prediction system by attribute representation learning
8.7.1 Background
8.7.2 Methods
8.7.3 Results
8.8 Case study: transfer learning for nighttime traffic detection
8.9 Case study: simulation to real-world knowledge transfer for driving be- havior recognition
8.10 Exercises
9 Graph neural networks
9.1 Preliminaries
9.2 Graph neural networks
9.2.1 Spectral GNN
9.2.2 Spatial GNN
9.2.3 Attention-based GNNs
9.3 Case study 1: traffic graph convolutional network for traffic prediction
9.3.1 Problem definition
9.3.2 Method: traffic graph convolutional LSTM.
9.3.3 Results
9.4 Case study 2: graph neural network for traffic forecasting with missing values
9.4.1 Problem definition
9.4.2 Method: graph Markov network
9.4.3 Results
9.5 Case study 3: graph neural network (GNN) for vehicle keypoints' cor- rection
9.5.1 Problem definition
9.5.2 Method: graph neural network for keypoints correction
9.5.3 Results
9.6 Exercises
9.6.1 Questions
10 Generative adversarial networks
10.1 Generative adversarial network (GAN)
10.1.1 Binary classification
10.1.2 Original GAN formulation as binary classification
10.1.3 Objective (loss) function
10.1.4 Optimization algorithm
10.2 Case studies: GAN-based roadway traffic state estimation
10.2.1 Problem formulation
10.2.2 Model: generative adversarial architecture for spatiotemporal traffic- state estimation
10.2.3 Results
10.3 Case study: conditional GAN-based taxi hotspots prediction
10.3.1 Problem formulation
10.3.2 Model: LSTM-CGAN-based-hotspot prediction
10.3.3 Results
10.4 Case study: GAN-based pavement image data transferring
10.4.1 Problem formulation
10.4.2 Model: CycleGAN-based image style transfer
10.4.3 Results
10.5 Exercises
11 Edge and parallel artificial intelligence
11.1 Edge computing concept
11.2 Edge artificial intelligence
11.3 Parallel artificial intelligence
11.4 Federated learning concept
11.5 Federated learning methods
11.5.1 Horizontal federated learning
11.5.2 Vertical federated learning
11.6 Case study 1: parallel and edge AI in multi-task traffic surveillance
11.6.1 Motivations
11.6.2 Parallel edge computing system architecture
11.6.3 Algorithms and results
11.7 Case study 2: edge AI in vehicle near-crash detection
11.7.1 Motivations
11.7.2 Relative motion patterns in camera views for near-crashes.
11.7.3 Edge computing system architecture
11.7.4 Camera-parameter-free near-crash detection algorithm
11.7.5 Height or width
11.7.6 Modeling bounding box centers for horizontal motion pattern identi- fication
11.7.7 Experimental results
11.8 Case study 3: federated learning for vehicle trajectory prediction
11.8.1 Motivation
11.8.2 Methodology
11.8.3 Results
11.9 Exercises
12 Future directions
12.1 Future trends of deep learning technologies for transportation
12.2 The future of transportation with AI
12.3 Book extension and future plan
Bibliography
Index
Back Cover.