Abstract:Artificial Intelligence (AI) has been widely adopted in numerous fields and enabled various smart systems because of its strong ability to perform tasks, including prediction, event detection, and status estimation, among others. As one of the typical smart systems empowered by AI and Internet of Things (IoT) technologies, the smart transportation system has made dramatic progress for traditional transportation in numerous aspects, including autonomous driving, smart traffic lights, navigation, and traffic forecasting, among others. Deep learning is an essential component to enable such smart systems. Typically, specific deep learning models, e.g., Recurrent Neural Networks (RNN) and Convolutional Neural Networks (CNN), can be trained on collected transportation data for a particular task. However, traditional deep learning techniques rely on data sufficiency to build an effective model. Additionally, each trained model can only work on one single task. This has limited the efficacy of deep learning techniques in numerous application scenarios. To this end, multi-task learning (MTL) has been studied to train a single model that can work for multiple tasks. This technique effectively allows the learning model to expand to specific tasks with only a limited amount of data affiliated. In the meantime, MTL significantly reduces the training time of each task. The success of MTL requires that there are potential relationships among different tasks. Many tasks in the smart transportation system are related. For instance, traffic speed and vehicle volume estimations for each road are highly correlated. Based on this, research on applying MTL in smart transportation systems has been studied recently. This paper reviews the recent efforts to use MTL in smart transportation systems and conducts an extensive survey to provide insights. In particular, we categorize the MTL applications in smart transportation systems into traffic forecasting, traffic sign recognition, vehicle recognition, travel time estimation, road safety estimation, taxi demand prediction, and autonomous driving. Ultimately, we discuss challenges and future research directions in applying MTL in smart transportation systems.

Enhancing road surface recognition via optimal transport and metric learning in task-agnostic intelligent driving environments

An Adaptive Machine Learning Framework for Multi-Scenes Road Surface Weather Condition Monitoring

MCMCNet: A Semi-supervised Road Extraction Network for High-resolution Remote Sensing Images Via Multiple Consistency and Multi-task Constraints

Improving classification of road surface conditions via road area extraction and contrastive learning

Unifying Terrain Awareness Through Real-Time Semantic Segmentation

Road Surface State Recognition Based on SVM Optimization and Image Segmentation Processing

Multi-Traffic Scene Perception Based on Supervised Learning.

A Comprehensive Implementation of Road Surface Classification for Vehicle Driving Assistance: Dataset, Models, and Deployment

Robust Identification of Road Surface Condition Based on Ego-Vehicle Trajectory Reckoning

Road Detection Method Based on Online Learning

Lightweight Model Development for Forest Region Unstructured Road Recognition Based on Tightly Coupled Multisource Information

Domain Adaptive Road Perception Network of Autonomous Vehicles

Survey on Multi-Task Learning in Smart Transportation

Reinforced Wasserstein Training for Severity-Aware Semantic Segmentation in Autonomous Driving

Research on Road Scene Understanding of Autonomous Vehicles Based on Multi-Task Learning

An End-to-End Multi-Task Learning Model for Drivable Road Detection via Edge Refinement and Geometric Deformation

Traffic Sign Recognition Via Multi-Modal Tree-Structure Embedded Multi-Task Learning.

Integrated Road Information Perception Framework for Road Type Recognition and Adaptive Evenness Assessment

Multi-Modality Driven LoRA for Adverse Condition Depth Estimation

SA-MixNet: Structure-aware Mixup and Invariance Learning for Scribble-supervised Road Extraction in Remote Sensing Images

Road Recognition from Remote Sensing Imagery Using Incremental Learning