Junxiao Xue
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Junxiao Xue has a total of 131 co-authors. Here are the top co-authors by number of shared publications:
- Xuecheng Wu
Xi'an Jiaotong University, Xi'an, China
8 shared publications - Mingliang Xu
Zhengzhou University
6 shared publications - Hao Wang
Xi'an Jiaotong University, Xi'an, China
4 shared publications - Heli Sun
Xi'an Jiaotong University, Xi'an, China
4 shared publications - Liang He
Xi'an Jiaotong University, Xi'an, China
4 shared publications - Xinyi Yin
Zhengzhou University, Zhengzhou, China
4 shared publications
Latest Publications
- 18 November 2026
- Published by Elsevier in Information Fusion
FedSKT: A Selective Knowledge Transfer Scheme for Federated Learning via End-Edge-Cloud Collaboration
- 18 November 2026
- Published by Elsevier in Information Fusion
FedSKT: A Selective Knowledge Transfer Scheme for Federated Learning via End-Edge-Cloud Collaboration
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 8915-8924
Motif-FPD: Towards Free-Form Natural Language-Driven Functional Protein Design via Motif-Centric LLM
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 8915-8924
Open AccessMotif-FPD: Towards Free-Form Natural Language-Driven Functional Protein Design via Motif-Centric LLM
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 883-892
Explaining Chain-of-Thought Reasoning in Vision-Language Models via Interactions
Recent studies have extended chain-of-thought (CoT) reasoning to Vision-Language Models (VLMs) to improve their performance in complex multimodal tasks. However, the underlying mechanisms of how VLMs learn inference patterns from CoT reasoning to process multimodal input information remain in a black-box state. To this end, we define and quantify the inference patterns learned by VLMs from CoT prompts, which help us analyze the representation characteristics of VLMs and reveal the mechanisms underlying CoT reasoning in VLMs. Specifically, we disentangle the output score of a VLM into a set of interactions between input variables in the CoT prompt and take these interactions as detailed inference patterns used by the VLM. Based on the interaction patterns, we analyze several classical VLMs and draw the following new insights on CoT reasoning in VLMs. (1) The interaction patterns used by VLMs exhibit high complexity, which means that VLMs can comprehend and internalize complex logical chains constructed by CoT prompts. (2) The interaction patterns used by VLMs manifest a high degree of adjacency, which implies that VLMs can truly learn and understand the adjacent logic underlying CoT prompts. (3) The text-based interaction patterns used by VLMs change accordingly when the visual information in input samples changes, which indicates that VLMs can model cross-modal inference patterns from CoT prompts. The code is available at https://github.com/Error404-stack1024/CoT-Interaction.
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 883-892
Open AccessExplaining Chain-of-Thought Reasoning in Vision-Language Models via Interactions
Recent studies have extended chain-of-thought (CoT) reasoning to Vision-Language Models (VLMs) to improve their performance in complex multimodal tasks. However, the underlying mechanisms of how VLMs learn inference patterns from CoT reasoning to process multimodal input information remain in a black-box state. To this end, we define and quantify the inference patterns learned by VLMs from CoT prompts, which help us analyze the representation characteristics of VLMs and reveal the mechanisms underlying CoT reasoning in VLMs. Specifically, we disentangle the output score of a VLM into a set of interactions between input variables in the CoT prompt and take these interactions as detailed inference patterns used by the VLM. Based on the interaction patterns, we analyze several classical VLMs and draw the following new insights on CoT reasoning in VLMs. (1) The interaction patterns used by VLMs exhibit high complexity, which means that VLMs can comprehend and internalize complex logical chains constructed by CoT prompts. (2) The interaction patterns used by VLMs manifest a high degree of adjacency, which implies that VLMs can truly learn and understand the adjacent logic underlying CoT prompts. (3) The text-based interaction patterns used by VLMs change accordingly when the visual information in input samples changes, which indicates that VLMs can model cross-modal inference patterns from CoT prompts. The code is available at https://github.com/Error404-stack1024/CoT-Interaction.
- 25 September 2026
- Published by Springer Nature
- p. 288-305
FED-Bench: A Cross-Granular Benchmark for Disentangled Evaluation of Facial Expression Editing
- 25 September 2026
- Published by Springer Nature
- p. 288-305
FED-Bench: A Cross-Granular Benchmark for Disentangled Evaluation of Facial Expression Editing
- 23 September 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 334
FedCAFE: A federated cross-attention feature exchange framework for heterogeneous large language model fine-tuning
- 23 September 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 334
FedCAFE: A federated cross-attention feature exchange framework for heterogeneous large language model fine-tuning
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-16
Toward Comprehensive Interactive Change Understanding in Remote Sensing: A Large-Scale Dataset and Dual-Granularity Enhanced VLM
Remote sensing change understanding (RSCU) is essential for analyzing remote sensing images and understanding how human activities affect the environment. However, existing datasets lack deep understanding and interactions in the diverse change captioning, counting, and localization tasks. To tackle these gaps, we construct ChangeIMTI, a new large-scale interactive multi-task instruction dataset that encompasses four complementary tasks including change captioning, binary change classification, change counting, and change localization. Building upon this new dataset, we further design a novel vision-guided vision-language model (ChangeVG) with dual-granularity awareness for bi-temporal remote sensing images (i.e., two remote sensing images of the same area at different times). The introduced visionguided module is a dual-branch architecture that synergistically combines fine-grained spatial feature extraction with high-level semantic summarization. These enriched representations further serve as the auxiliary prompts to guide large vision-language models (VLMs) (e.g., Qwen2.5-VL-7B) during instruction tuning, thereby facilitating the hierarchical cross-modal learning. We extensively conduct experiments across four tasks to demonstrate the superiority of our approach. Remarkably, on the change captioning task, our method outperforms the strongest method Semantic-CC by 1.39 points on the comprehensive S∗m metric, which integrates the semantic similarity and descriptive accuracy to provide an overall evaluation of change caption. Moreover, we also perform a series of ablation studies to examine the critical components of our method. The source code and associated data for this work are publicly available at Github.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-16
Toward Comprehensive Interactive Change Understanding in Remote Sensing: A Large-Scale Dataset and Dual-Granularity Enhanced VLM
Remote sensing change understanding (RSCU) is essential for analyzing remote sensing images and understanding how human activities affect the environment. However, existing datasets lack deep understanding and interactions in the diverse change captioning, counting, and localization tasks. To tackle these gaps, we construct ChangeIMTI, a new large-scale interactive multi-task instruction dataset that encompasses four complementary tasks including change captioning, binary change classification, change counting, and change localization. Building upon this new dataset, we further design a novel vision-guided vision-language model (ChangeVG) with dual-granularity awareness for bi-temporal remote sensing images (i.e., two remote sensing images of the same area at different times). The introduced visionguided module is a dual-branch architecture that synergistically combines fine-grained spatial feature extraction with high-level semantic summarization. These enriched representations further serve as the auxiliary prompts to guide large vision-language models (VLMs) (e.g., Qwen2.5-VL-7B) during instruction tuning, thereby facilitating the hierarchical cross-modal learning. We extensively conduct experiments across four tasks to demonstrate the superiority of our approach. Remarkably, on the change captioning task, our method outperforms the strongest method Semantic-CC by 1.39 points on the comprehensive S∗m metric, which integrates the semantic similarity and descriptive accuracy to provide an overall evaluation of change caption. Moreover, we also perform a series of ablation studies to examine the critical components of our method. The source code and associated data for this work are publicly available at Github.
- 26 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (6) , 11472-11492
Space Computing Constellation: System Architecture, Implementations, and Challenges
Low Earth Orbit (LEO) satellite constellations have experienced rapid growth in recent years, driven by their potential to deliver global, high-bandwidth Internet services with low latency. Beyond connectivity, LEO constellations also offer promising opportunities to enable in-orbit processing of space-native data to support a wide range of emerging space applications. In this context, the concept of space computing has been proposed, a paradigm that seamlessly integrates networking and computing to provide computing-as-a-service anytime and anywhere in space. However, the inherent characteristics of satellite constellations, such as dynamic network topologies, constrained system resources, and the harsh space environment, pose significant challenges in achieving this vision. This paper outlines the system architecture and the key enabling technologies for space computing, including spaceborne computers, laser communications, spaceborne router, distributed operating systems, and onboard AI. We also present the implementation of an open space computing platform, the 3-Body computing constellation, along with the in-orbit experimental results that demonstrate the advantages of multi-satellite distributed computing. Furthermore, we outline future research directions essential for advancing toward a truly interconnected, autonomous, and intelligent space computing system.
- 26 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (6) , 11472-11492
Space Computing Constellation: System Architecture, Implementations, and Challenges
Low Earth Orbit (LEO) satellite constellations have experienced rapid growth in recent years, driven by their potential to deliver global, high-bandwidth Internet services with low latency. Beyond connectivity, LEO constellations also offer promising opportunities to enable in-orbit processing of space-native data to support a wide range of emerging space applications. In this context, the concept of space computing has been proposed, a paradigm that seamlessly integrates networking and computing to provide computing-as-a-service anytime and anywhere in space. However, the inherent characteristics of satellite constellations, such as dynamic network topologies, constrained system resources, and the harsh space environment, pose significant challenges in achieving this vision. This paper outlines the system architecture and the key enabling technologies for space computing, including spaceborne computers, laser communications, spaceborne router, distributed operating systems, and onboard AI. We also present the implementation of an open space computing platform, the 3-Body computing constellation, along with the in-orbit experimental results that demonstrate the advantages of multi-satellite distributed computing. Furthermore, we outline future research directions essential for advancing toward a truly interconnected, autonomous, and intelligent space computing system.
- Open Access
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 3-11
AD-AVSR: Asymmetric Dual-stream Enhancement for Robust Audio-Visual Speech Recognition
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 3-11
Open AccessAD-AVSR: Asymmetric Dual-stream Enhancement for Robust Audio-Visual Speech Recognition
- Open Access
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 13692-13699
HOLA: Enhancing Audio-visual Deepfake Detection via Hierarchical Contextual Aggregations and Efficient Pre-training
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 13692-13699
Open AccessHOLA: Enhancing Audio-visual Deepfake Detection via Hierarchical Contextual Aggregations and Efficient Pre-training