Jing Xiong
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Jing Xiong has a total of 91 co-authors. Here are the top co-authors by number of shared publications:
Ngai Wong
4 shared publicationsChaofan Tao
3 shared publicationsLingpeng Kong
The University of Hong Kong
3 shared publicationsZhongwei Wan
Ohio State University
3 shared publicationsChuanyang Zheng
The Chinese University of Hong Kong
2 shared publicationsHui Shen
2 shared publications
Latest Publications
- Open Access
- 20 April 2026
- Published by MDPI AG in Preprints.org
Speculative Decoding for Multimodal Models: A Survey
Multimodal generative models have demonstrated remarkable capabilities across diverse domains, from visual understanding and image generation to video processing, audio synthesis, and embodied control. These capabilities, however, incur substantial inference overhead due to autoregressive decoding or iterative generation, which is further compounded by modality-specific challenges such as extensive visual token redundancy, strict real-time latency constraints in robotic control, and prolonged sequential generation in text-to-image synthesis. Speculative decoding has emerged as a promising paradigm to accelerate inference without degrading output quality, yet existing surveys remain focused on text-only large language models. In this survey, we provide a systematic and comprehensive review of speculative decoding methods for multimodal models, spanning Vision-Language, Vision-Language-Action, Video-Language, Speech, Text-to-Image (Vision Auto-Regressive), and Diffusion models. We organize the literature into a unified taxonomy with two primary axes, covering the draft generation stage and the verification and acceptance stage, complemented by an analysis of inference framework support. Through this taxonomy, we identify recurring cross-modal design patterns, including token compression, KV cache optimization, target-informed transfer, drafter-target alignment, verification cost reduction, relaxed acceptance, and verify-to-draft feedback, and examine how successful techniques transfer across modalities. We further provide a systematic comparison of existing methods under both self-reported and standardized benchmarking settings. Finally, we discuss open challenges and outline future directions. We have also created a GitHub repository that organizes the papers featured in this survey at https://github.com/zyfzs0/Multimodal-Models-Speculative-Decoding-Survey, and we will actively maintain it as new research emerges. We hope this survey can serve as a valuable resource for researchers and practitioners working on accelerating multimodal inference.
- 20 April 2026
- Published by MDPI AG in Preprints.org
Open AccessSpeculative Decoding for Multimodal Models: A Survey
Multimodal generative models have demonstrated remarkable capabilities across diverse domains, from visual understanding and image generation to video processing, audio synthesis, and embodied control. These capabilities, however, incur substantial inference overhead due to autoregressive decoding or iterative generation, which is further compounded by modality-specific challenges such as extensive visual token redundancy, strict real-time latency constraints in robotic control, and prolonged sequential generation in text-to-image synthesis. Speculative decoding has emerged as a promising paradigm to accelerate inference without degrading output quality, yet existing surveys remain focused on text-only large language models. In this survey, we provide a systematic and comprehensive review of speculative decoding methods for multimodal models, spanning Vision-Language, Vision-Language-Action, Video-Language, Speech, Text-to-Image (Vision Auto-Regressive), and Diffusion models. We organize the literature into a unified taxonomy with two primary axes, covering the draft generation stage and the verification and acceptance stage, complemented by an analysis of inference framework support. Through this taxonomy, we identify recurring cross-modal design patterns, including token compression, KV cache optimization, target-informed transfer, drafter-target alignment, verification cost reduction, relaxed acceptance, and verify-to-draft feedback, and examine how successful techniques transfer across modalities. We further provide a systematic comparison of existing methods under both self-reported and standardized benchmarking settings. Finally, we discuss open challenges and outline future directions. We have also created a GitHub repository that organizes the papers featured in this survey at https://github.com/zyfzs0/Multimodal-Models-Speculative-Decoding-Survey, and we will actively maintain it as new research emerges. We hope this survey can serve as a valuable resource for researchers and practitioners working on accelerating multimodal inference.
- Open Access
OVD: On-policy Verbal Distillation
Knowledge distillation offers a promising path to transfer reasoning capabilities from large teacher models to efficient student models; however, existing token-level on-policy distillation methods require token-level alignment between the student and teacher models, which restricts the student model's exploration ability, prevent effective use of interactive environment feedback, and suffer from severe memory bottlenecks in reinforcement learning. We introduce On-policy Verbal Distillation (OVD), a memory-efficient framework that replaces token-level probability matching with trajectory matching using discrete verbal scores (0--9) from teacher models. OVD dramatically reduces memory consumption while enabling on-policy distillation from teacher models with verbal feedback, and avoids token-level alignment, allowing the student model to freely explore the output space. Extensive experiments on Web question answering and mathematical reasoning tasks show that OVD substantially outperforms existing methods, delivering up to +12.9% absolute improvement in average EM on Web Q&A tasks and a up to +25.7% gain on math benchmarks (when trained with only one random samples), while also exhibiting superior training efficiency. Our project page is available at https://OVD.github.io
Open AccessOVD: On-policy Verbal Distillation
Knowledge distillation offers a promising path to transfer reasoning capabilities from large teacher models to efficient student models; however, existing token-level on-policy distillation methods require token-level alignment between the student and teacher models, which restricts the student model's exploration ability, prevent effective use of interactive environment feedback, and suffer from severe memory bottlenecks in reinforcement learning. We introduce On-policy Verbal Distillation (OVD), a memory-efficient framework that replaces token-level probability matching with trajectory matching using discrete verbal scores (0--9) from teacher models. OVD dramatically reduces memory consumption while enabling on-policy distillation from teacher models with verbal feedback, and avoids token-level alignment, allowing the student model to freely explore the output space. Extensive experiments on Web question answering and mathematical reasoning tasks show that OVD substantially outperforms existing methods, delivering up to +12.9% absolute improvement in average EM on Web Q&A tasks and a up to +25.7% gain on math benchmarks (when trained with only one random samples), while also exhibiting superior training efficiency. Our project page is available at https://OVD.github.io
- Open Access
- 21 January 2026
- Published by MDPI AG in Preprints.org
Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models
Mechanistic Interpretability (MI) has emerged as a vital approach to demystify the opaque decision-making of Large Language Models (LLMs). However, existing reviews primarily treat MI as an observational science, summarizing analytical insights while lacking a systematic framework for actionable intervention. To bridge this gap, we present a practical survey structured around the pipeline: "Locate, Steer, and Improve." We formally categorize Localizing (diagnosis) and Steering (intervention) methods based on specific Interpretable Objects to establish a rigorous intervention protocol. Furthermore, we demonstrate how this framework enables tangible improvements in Alignment, Capability, and Efficiency, effectively operationalizing MI as an actionable methodology for model optimization. With actionable mechanistic interpretability evolving at a fast pace, we pledge to keep this survey up to date, ensuring it reflects the cutting-edge advances in this area.
- 21 January 2026
- Published by MDPI AG in Preprints.org
Open AccessLocate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models
Mechanistic Interpretability (MI) has emerged as a vital approach to demystify the opaque decision-making of Large Language Models (LLMs). However, existing reviews primarily treat MI as an observational science, summarizing analytical insights while lacking a systematic framework for actionable intervention. To bridge this gap, we present a practical survey structured around the pipeline: "Locate, Steer, and Improve." We formally categorize Localizing (diagnosis) and Steering (intervention) methods based on specific Interpretable Objects to establish a rigorous intervention protocol. Furthermore, we demonstrate how this framework enables tangible improvements in Alignment, Capability, and Efficiency, effectively operationalizing MI as an actionable methodology for model optimization. With actionable mechanistic interpretability evolving at a fast pace, we pledge to keep this survey up to date, ensuring it reflects the cutting-edge advances in this area.
- Open Access
MMFormalizer: Multimodal Autoformalization in the Wild
Autoformalization, which translates natural language mathematics into formal statements to enable machine reasoning, faces fundamental challenges in the wild due to the multimodal nature of the physical world, where physics requires inferring hidden constraints (e.g., mass or energy) from visual elements. To address this, we propose MMFormalizer, which extends autoformalization beyond text by integrating adaptive grounding with entities from real-world mathematical and physical domains. MMFormalizer recursively constructs formal propositions from perceptually grounded primitives through recursive grounding and axiom composition, with adaptive recursive termination ensuring that every abstraction is supported by visual evidence and anchored in dimensional or axiomatic grounding. We evaluate MMFormalizer on a new benchmark, PhyX-AF, comprising 115 curated samples from MathVerse, PhyX, Synthetic Geometry, and Analytic Geometry, covering diverse multimodal autoformalization tasks. Results show that frontier models such as GPT-5 and Gemini-3-Pro achieve the highest compile and semantic accuracy, with GPT-5 excelling in physical reasoning, while geometry remains the most challenging domain. Overall, MMFormalizer provides a scalable framework for unified multimodal autoformalization, bridging perception and formal reasoning. To the best of our knowledge, this is the first multimodal autoformalization method capable of handling classical mechanics (derived from the Hamiltonian), as well as relativity, quantum mechanics, and thermodynamics. More details are available on our project page: MMFormalizer.github.io
Open AccessMMFormalizer: Multimodal Autoformalization in the Wild
Autoformalization, which translates natural language mathematics into formal statements to enable machine reasoning, faces fundamental challenges in the wild due to the multimodal nature of the physical world, where physics requires inferring hidden constraints (e.g., mass or energy) from visual elements. To address this, we propose MMFormalizer, which extends autoformalization beyond text by integrating adaptive grounding with entities from real-world mathematical and physical domains. MMFormalizer recursively constructs formal propositions from perceptually grounded primitives through recursive grounding and axiom composition, with adaptive recursive termination ensuring that every abstraction is supported by visual evidence and anchored in dimensional or axiomatic grounding. We evaluate MMFormalizer on a new benchmark, PhyX-AF, comprising 115 curated samples from MathVerse, PhyX, Synthetic Geometry, and Analytic Geometry, covering diverse multimodal autoformalization tasks. Results show that frontier models such as GPT-5 and Gemini-3-Pro achieve the highest compile and semantic accuracy, with GPT-5 excelling in physical reasoning, while geometry remains the most challenging domain. Overall, MMFormalizer provides a scalable framework for unified multimodal autoformalization, bridging perception and formal reasoning. To the best of our knowledge, this is the first multimodal autoformalization method capable of handling classical mechanics (derived from the Hamiltonian), as well as relativity, quantum mechanics, and thermodynamics. More details are available on our project page: MMFormalizer.github.io
- Open Access
CTR-Sink: Attention Sink for Language Models in Click-Through Rate Prediction
Click-Through Rate (CTR) prediction, a core task in recommendation systems, estimates user click likelihood using historical behavioral data. Modeling user behavior sequences as text to leverage Language Models (LMs) for this task has gained traction, owing to LMs' strong semantic understanding and contextual modeling capabilities. However, a critical structural gap exists: user behavior sequences consist of discrete actions connected by semantically empty separators, differing fundamentally from the coherent natural language in LM pre-training. This mismatch causes semantic fragmentation, where LM attention scatters across irrelevant tokens instead of focusing on meaningful behavior boundaries and inter-behavior relationships, degrading prediction performance. To address this, we propose $\textit{CTR-Sink}$, a novel framework introducing behavior-level attention sinks tailored for recommendation scenarios. Inspired by attention sink theory, it constructs attention focus sinks and dynamically regulates attention aggregation via external information. Specifically, we insert sink tokens between consecutive behaviors, incorporating recommendation-specific signals such as temporal distance to serve as stable attention sinks. To enhance generality, we design a two-stage training strategy that explicitly guides LM attention toward sink tokens and a attention sink mechanism that amplifies inter-sink dependencies to better capture behavioral correlations. Experiments on one industrial dataset and two open-source datasets (MovieLens, Kuairec), alongside visualization results, validate the method's effectiveness across scenarios.Open AccessCTR-Sink: Attention Sink for Language Models in Click-Through Rate Prediction
Click-Through Rate (CTR) prediction, a core task in recommendation systems, estimates user click likelihood using historical behavioral data. Modeling user behavior sequences as text to leverage Language Models (LMs) for this task has gained traction, owing to LMs' strong semantic understanding and contextual modeling capabilities. However, a critical structural gap exists: user behavior sequences consist of discrete actions connected by semantically empty separators, differing fundamentally from the coherent natural language in LM pre-training. This mismatch causes semantic fragmentation, where LM attention scatters across irrelevant tokens instead of focusing on meaningful behavior boundaries and inter-behavior relationships, degrading prediction performance. To address this, we propose $\textit{CTR-Sink}$, a novel framework introducing behavior-level attention sinks tailored for recommendation scenarios. Inspired by attention sink theory, it constructs attention focus sinks and dynamically regulates attention aggregation via external information. Specifically, we insert sink tokens between consecutive behaviors, incorporating recommendation-specific signals such as temporal distance to serve as stable attention sinks. To enhance generality, we design a two-stage training strategy that explicitly guides LM attention toward sink tokens and a attention sink mechanism that amplifies inter-sink dependencies to better capture behavioral correlations. Experiments on one industrial dataset and two open-source datasets (MovieLens, Kuairec), alongside visualization results, validate the method's effectiveness across scenarios. - Open Access
- 1 January 2025
- Published by Association for Computational Linguistics (ACL)
- p. 4179-4199
UNComp: Can Matrix Entropy Uncover Sparsity? — A Compressor Design from an Uncertainty-Aware Perspective
Jing Xiong, Jianghan Shen, Fanghua Ye, Chaofan Tao, Zhongwei Wan, Jianqiao Lu, Xun Wu, Chuanyang Zheng, Zhijiang Guo, Min Yang, Lingpeng Kong, Ngai Wong. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.
- 1 January 2025
- Published by Association for Computational Linguistics (ACL)
- p. 4179-4199
Open AccessUNComp: Can Matrix Entropy Uncover Sparsity? — A Compressor Design from an Uncertainty-Aware Perspective
Jing Xiong, Jianghan Shen, Fanghua Ye, Chaofan Tao, Zhongwei Wan, Jianqiao Lu, Xun Wu, Chuanyang Zheng, Zhijiang Guo, Min Yang, Lingpeng Kong, Ngai Wong. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.
- Open Access
UncertaintyRAG: Span-Level Uncertainty Enhanced Long-Context Modeling for Retrieval-Augmented Generation
We present UncertaintyRAG, a novel approach for long-context Retrieval-Augmented Generation (RAG) that utilizes Signal-to-Noise Ratio (SNR)-based span uncertainty to estimate similarity between text chunks. This span uncertainty enhances model calibration, improving robustness and mitigating semantic inconsistencies introduced by random chunking. Leveraging this insight, we propose an efficient unsupervised learning technique to train the retrieval model, alongside an effective data sampling and scaling strategy. UncertaintyRAG outperforms baselines by 2.03% on LLaMA-2-7B, achieving state-of-the-art results while using only 4% of the training data compared to other advanced open-source retrieval models under distribution shift settings. Our method demonstrates strong calibration through span uncertainty, leading to improved generalization and robustness in long-context RAG tasks. Additionally, UncertaintyRAG provides a lightweight retrieval model that can be integrated into any large language model with varying context window lengths, without the need for fine-tuning, showcasing the flexibility of our approach.
Open AccessUncertaintyRAG: Span-Level Uncertainty Enhanced Long-Context Modeling for Retrieval-Augmented Generation
We present UncertaintyRAG, a novel approach for long-context Retrieval-Augmented Generation (RAG) that utilizes Signal-to-Noise Ratio (SNR)-based span uncertainty to estimate similarity between text chunks. This span uncertainty enhances model calibration, improving robustness and mitigating semantic inconsistencies introduced by random chunking. Leveraging this insight, we propose an efficient unsupervised learning technique to train the retrieval model, alongside an effective data sampling and scaling strategy. UncertaintyRAG outperforms baselines by 2.03% on LLaMA-2-7B, achieving state-of-the-art results while using only 4% of the training data compared to other advanced open-source retrieval models under distribution shift settings. Our method demonstrates strong calibration through span uncertainty, leading to improved generalization and robustness in long-context RAG tasks. Additionally, UncertaintyRAG provides a lightweight retrieval model that can be integrated into any large language model with varying context window lengths, without the need for fine-tuning, showcasing the flexibility of our approach.