Author name cluster

Chuan Xu

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.

6 papers

1 author row

AIIM Journal 2026 Journal Article

Tackling small sample survival analysis via transfer learning: A study of colorectal cancer prognosis

Yonghao Zhao
Changtao Li
Chi Shu
Qingbin Wu
Hong Li
Chuan Xu
Tianrui Li
Ziqiang Wang

Survival prognosis is crucial for medical informatics. Practitioners often confront small-sized clinical data, especially cancer patient cases, which can be insufficient to induce useful patterns for survival predictions. This study deals with small sample survival analysis by leveraging transfer learning, a useful machine learning technique that can enhance the target analysis with related knowledge pre-learned from other data. We propose and develop various transfer learning methods designed for common survival models. For parametric models such as DeepSurv, Cox-CC (Cox-based neural networks), and DeepHit (end-to-end deep learning model), we apply standard transfer learning techniques like pretraining and fine-tuning. For non-parametric models such as Random Survival Forest, we propose a new transfer survival forest (TSF) model that transfers tree structures from source tasks and fine-tunes them with target data. We evaluated the transfer learning methods on colorectal cancer (CRC) prognosis. The source data are 27, 379 SEER CRC stage I patients, and the target data are 728 CRC stage I patients from the West China Hospital. When enhanced by transfer learning, Cox-CC’s C t d value was boosted from 0. 7868 to 0. 8111, DeepHit’s from 0. 8085 to 0. 8135, DeepSurv’s from 0. 7722 to 0. 8043, and RSF’s from 0. 7940 to 0. 8297 (the highest performance). All models trained with data as small as 50 demonstrated even more significant improvement. Conclusions: Therefore, the current survival models used for cancer prognosis can be enhanced and improved by properly designed transfer learning techniques. The source code used in this study is available at https: //github. com/YonghaoZhao722/TSF.

Details DOI

EAAI Journal 2025 Journal Article

A modeled study of driver visual attention driven by driving tasks

Chuan Xu
Bo Jiang
Yukun Wang
Yan Su

Visual attention is an indispensable component of driving, enabling drivers to swiftly identify critical objects within complex and dynamic traffic environments. Despite its significance, existing visual attention models predominantly focus on static or idealized driving scenarios, limiting their ability to capture attention distribution patterns in real-world, dynamic environments. Furthermore, most of these models rely heavily on data-driven approaches, extracting features exclusively from visual image data, while neglecting the profound influence of “the driver, the vehicle, and the road environment”. Consequently, these models frequently fail to effectively address the intricacies of practical driving scenarios. To bridge these gaps, this study introduces a driver visual attention prediction model that comprehensively incorporates the driving task, driver experience, and the impact of dynamic visual scenes. The proposed model leverages the advanced learning capabilities of Convolutional Neural Networks (CNN) and Vision Transformer (ViT), coupled with sequence modeling mechanisms, to effectively capture the nuanced attention allocation patterns of drivers in complex driving contexts. The model is meticulously designed to adapt to dynamically evolving driving task requirements. Experimental results demonstrate that the proposed model outperforms state-of-the-art (SOTA) visual attention prediction models across multiple benchmark evaluation metrics on the DR(eye)VE dataset, particularly excelling in dynamic driving conditions. Moreover, generalization experiments were conducted on the BDD-A and TDV datasets validate the model’s robustness and applicability across varied driving tasks and dynamic conditions.

Details DOI

AAAI Conference 2025 Conference Paper

Attribute Inference Attacks for Federated Regression Tasks

Francesco Diana
Othmane Marfoq
Chuan Xu
Giovanni Neglia
Frédéric Giroire
Eoin Thomas

Federated Learning (FL) enables multiple clients, such as mobile phones and IoT devices, to collaboratively train a global machine learning model while keeping their data localized. However, recent studies have revealed that the training phase of FL is vulnerable to reconstruction attacks, such as attribute inference attacks (AIA), where adversaries exploit exchanged messages and auxiliary public information to uncover sensitive attributes of targeted clients. While these attacks have been extensively studied in the context of classification tasks, their impact on regression tasks remains largely unexplored. In this paper, we address this gap by proposing novel model-based AIAs specifically designed for regression tasks in FL environments. Our approach considers scenarios where adversaries can either eavesdrop on exchanged messages or directly interfere with the training process. We benchmark our proposed attacks against state-of-the-art methods using real-world datasets. The results demonstrate a significant increase in reconstruction accuracy, particularly in heterogeneous client datasets, a common scenario in FL. The efficacy of our model-based AIAs makes them better candidates for empirically quantifying privacy leakage for federated regression tasks.

PDF Details DOI

EAAI Journal 2025 Journal Article

Graph Neural Network-based Deep Reinforcement Learning algorithm for Virtual Network Function forwarding graph embedding in Space–Air–Ground Integrated Network

Liang Liu
Tengxiang Jing
Siyuan Tan
Yujie Zhang
Yejun He
Chuan Xu

Space–Air–Ground Integrated Network (SAGIN) can provide seamless three-dimensional coverage and enhanced flexibility, and it has been recognized as the core architecture of future 6G networks. Software Defined Network (SDN) and Network Function Virtualization (NFV) are two enabling technologies of SAGIN, which can sequentially arrange multiple Virtual Network Functions (VNFs) into VNF Forwarding Graph (VNF-FG) to provide users with scalable and parallel network services. However, SAGIN exhibits significant dynamism and heterogeneity, and VNFs may be deployed in multiple different heterogeneous locations, which brings great challenges to the efficient embedding of VNF-FGs required for user service request flows. In this paper, we study the VNF-FG embedding problem by jointly considering the structural constraints, node and link resource constraints, and End-to-End (E2E) delay constraint of VNF-FG in SDN/NFV-enabled SAGIN. Specifically, we first design a three-layer SDN/NFV-enabled SAGIN architecture consisting of a global controller and distributed intra-domain SDN controllers. Then, we define the Delay and Cost Efficient Dynamic VNF-FG Embedding Problem (DCE-DVEP) and formulate it as an Integer Linear Programming (ILP) with the objective of minimizing the weighted sum of E2E delay and embedding cost of all VNF-FGs. Finally, a Graph Neural Network-based Deep Reinforcement learning Embedding (GNN-DRE) algorithm is proposed to solve the DCE-DVEP, which can more accurately capture the rich feature information from both SAGIN and VNF-FG by specifically integrating different GNN models and adopts Deep Reinforcement Learning (DRL) to make effective embedding decisions. The simulation results demonstrate that, compared with other baseline algorithms, the GNN-DRE can reduce E2E delay and embedding cost by about 8% and 13%, respectively.

Details DOI

TCS Journal 2020 Journal Article

Data collection in population protocols with non-uniformly random scheduler

Chuan Xu
Joffroy Beauquier
Janna Burman
Shay Kutten
Thomas Nowak

Contrary to many previous studies on population protocols using the uniformly random scheduler, we consider a more general non-uniform case. Here, pair-wise interactions between agents (moving and communicating devices) are assumed to be drawn non-uniformly at random. While such a scheduler is known to be relevant for modeling many practical networks, it is also known to make the formal analysis more difficult. This study concerns data collection, a fundamental problem in mobile sensor networks (one of the target networks of population protocols). In this problem, pieces of information given to the agents (e. g. , sensor readings) should be eventually delivered to a predefined sink node without loss or duplication. Following an idea of the known deterministic protocol TTF solving this problem, we propose an adapted version of it and perform a complete formal analysis of execution times in expectation and with high probability (w. h. p.). We further investigate the important issue of energy consumption. The goal is to improve TTF in terms of energy complexity, while still keeping good time complexities (in expectation and w. h. p.). Namely, we propose a new parameterized protocol for data collection, called lazy TTF, and present a study showing that an appropriate choice of the protocol parameters can considerably improve energy performances (compared to TTF), at a slight expense of time performance.

Details DOI

NeurIPS Conference 2020 Conference Paper

Throughput-Optimal Topology Design for Cross-Silo Federated Learning

Othmane Marfoq
Chuan Xu
Giovanni Neglia
Richard Vidal

Federated learning usually employs a client-server architecture where an orchestrator iteratively aggregates model updates from remote clients and pushes them back a refined model. This approach may be inefficient in cross-silo settings, as close-by data silos with high-speed access links may exchange information faster than with the orchestrator, and the orchestrator may become a communication bottleneck. In this paper we define the problem of topology design for cross-silo federated learning using the theory of max-plus linear systems to compute the system throughput---number of communication rounds per time unit. We also propose practical algorithms that, under the knowledge of measurable network characteristics, find a topology with the largest throughput or with provable throughput guarantees. In realistic Internet networks with 10~Gbps access links for silos, our algorithms speed up training by a factor 9 and 1. 5 in comparison to the master-slave architecture and to state-of-the-art MATCHA, respectively. Speedups are even larger with slower access links.

PDF Details