Abstract:Large models are widely used in intelligent transportation systems (ITS) and autonomous vehicles (AV) due to their excellent new capabilities such as intelligence emergence, domain application adaptive, and multi-task learning. By integrating massive amounts of data, vehicles and transportation systems based on large models can understand the real-world environment, simulate the reasoning process of human drivers, optimize traffic flow, and improve driving safety and efficiency. However, in the practical implementation of large models, there are three key research questions: (1) How can model reasoning be consistent with the target task? (2) How to improve the redundancy of structures and weights caused by complex models? (3) How to solve the problems of supercomputing power, high latency, and large throughput of large models in practical deployment? These challenges have stimulated the development of deployment techniques for large models. Although existing review articles discuss large model technologies from singular or partial perspectives, there remains a lack of comprehensive systematic investigations into the application and deployment of large models for ITS and AV. Therefore, this paper conducted a quantitative analysis of scientific literature, demonstrating the necessity and significance of studying large models for ITS and AV. Subsequently, it outlined the concept and characteristics of large models and provided a detailed summary of the frontier progress of large models for ITS and AV. Moreover, to bridge the gap between large model inference and target tasks, address structural and weight redundancies, and tackle challenges in practical deployment such as high computational power, latency, and throughput, it explored efficient deployment techniques to accelerate the rapid deployment of large models. Finally, it discussed the challenges and future trends. This paper aims to provide researchers and engineers with an understanding of the forefront advancements and future trends of large models to facilitate the rapid implementation of large models and accelerate their development in ITS and AV.

Network Architecture and Technologies for Large Generative Models

Large-Scale Deterministic Networks: Architecture, Enabling Technologies, Case Study and Future Directions

High-Speed Data Communication with Advanced Networks in Large Language Model Training

Large Language Models for Networking: Applications, Enabling Techniques, and Challenges

ScaleNet: Searching for the Model to Scale.

A Look Into Training Large Language Models on Next Generation Datacenters

Large Language Models for Networking: Workflow, Advances and Challenges

Delay Modeling for Heterogeneous Backhaul Technologies

Large Language Model (LLM)-enabled Graphs in Dynamic Networking

Diffusion Models as Network Optimizers: Explorations and Analysis

Large models for intelligent transportation systems and autonomous vehicles: A survey

A Comprehensive Study on Large-Scale Graph Training: Benchmarking and Rethinking

Communication-Efficient TeraByte-Scale Model Training Framework for Online Advertising

Generative AI in Data Center Networking: Fundamentals, Perspectives, and Case Study

Communication-Efficient Large-Scale Distributed Deep Learning: A Comprehensive Survey

When Large Language Models Meet Optical Networks: Paving the Way for Automation

Large-scale Urban Cellular Traffic Generation Via Knowledge-Enhanced GANs with Multi-Periodic Patterns

An Efficient 2D Method for Training Super-Large Deep Learning Models

Optimizing Task Placement and Online Scheduling for Distributed GNN Training Acceleration

Xnet: Modeling Network Performance with Graph Neural Networks

A Survey on Performance Modeling and Prediction for Distributed DNN Training