Author name cluster

Ling Xing

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

2 papers

1 author row

EAAI Journal 2025 Journal Article

A survey of federated learning-based gradient compression for internet of vehicles

Ling Xing
Zhaocheng Luo
Jianping Gao
Kaikai Deng
Honghai Wu
Huahong Ma

The Federated Learning (FL) paradigm in the Internet of Vehicles (IoV) leverages vehicular networks and intelligent technologies to integrate data from traffic devices, aiming to develop a smart transportation system with high throughput, low latency, privacy protection, and collaborative multi-party training. However, frequent exchanges of model parameters between numerous vehicle nodes and roadside units (RSUs) in FL lead to uplink channel overload, which poses a significant challenge to system development. To this end, existing works introduce Gradient Compression (GC) technologies, incorporating methods such as count sketch, dynamic adjustment, and sparse matrices. These methods help to reduce communication overhead, enhance transmission efficiency, and maintain FL model training accuracy, making GC a crucial solution for overcoming communication barriers in IoV-based FL systems. In this paper, we first investigate GC technologies, systematically categorizing them by Quantization, Sparsification, and Driving Strategy-orientated compression, followed by using metrics such as efficiency, load, and latency to evaluate them. Secondly, we compare the pros and cons of different compression technologies in addressing communication problems and analyze their key characteristics. Finally, considering the deep integration of IoV and FL, we explore future research directions of the FL framework for IoV, analyze potential challenges, and propose corresponding solutions in conjunction with current mainstream deep learning models.

Details DOI

NeurIPS Conference 2025 Conference Paper

Vision-centric Token Compression in Large Language Model

Ling Xing
Alex Jinpeng Wang
Rui Yan
Xiangbo Shu
Jinhui Tang

Real-world applications are stretching context windows to hundreds of thousand of tokens while Large Language Models (LLMs) swell from billions to trillions of parameters. This dual expansion send compute and memory costs skyrocketing, making $\textit{token compression}$ indispensable. We introduce Vision Centric Token Compression ($\textbf{Vist}$), a $\textit{slow–fast}$ compression framework that mirrors human reading: the $\textit{fast}$ path renders distant tokens into images, letting a $\textbf{frozen, lightweight vision encoder}$ skim the low-salience context; the $\textit{slow}$ path feeds the proximal window into the LLM for fine-grained reasoning. A Probability-Informed Visual Enhancement (PVE) objective masks high-frequency tokens during training, steering the Resampler to concentrate on semantically rich regions—just as skilled reader gloss over function words. On eleven in-context learning benchmarks, $\textbf{Vist}$ achieves the same accuracy with 2. 3$\times$ fewer tokens, cutting FLOPs by 16\% and memory by 50\%. This method delivers remarkable results, outperforming the strongest text encoder-based compression method CEPE by $\textbf{7. 6}$\% on average over benchmarks like TriviaQA, NQ, PopQA, NLUI, and CLIN, setting a new standard for token efficiency in LLMs. The project is at https: //github. com/CSU-JPG/VIST.

PDF Details