Shiyuan Zhang

Yifeng Liu, Shiyuan Zhang, Yifan Zhang et al.

On-policy self-distillation, where a language model conditions on privileged context to supervise its own generations, is a promising source of dense supervision for sparse-reward reinforcement learning. Actually, it can be instantiated as an auxiliary full-vocabulary student-to-teacher reverse Kullback-Leibler divergence loss. We therefore propose SDPG, a self-distilled policy-gradient framework that combines group-relative verifier advantages with normalized standard deviation, exact full-vocabulary on-policy self-distillation, as well as reference-policy KL regularization. Empirically, SDPG improves stability and performance over RLVR and self-distillation baselines. The code is available at https://github.com/lauyikfung/SDPG.

Zeqing Wang, Shiyuan Zhang, Chengpei Tang et al.

Reasoning about temporal causality, particularly irreversible transformations of objects governed by real-world knowledge (e.g., fruit decay and human aging), is a fundamental aspect of human visual understanding. Unlike temporal perception based on simple event sequences, this form of reasoning requires a deeper comprehension of how object states change over time. Although the current powerful Vision-Language Models (VLMs) have demonstrated impressive performance on a wide range of downstream tasks, their capacity to reason about temporal causality remains underexplored. To address this gap, we introduce \textbf{TimeCausality}, a novel benchmark specifically designed to evaluate the causal reasoning ability of VLMs in the temporal dimension. Based on our TimeCausality, we find that while the current SOTA open-source VLMs have achieved performance levels comparable to closed-source models like GPT-4o on various standard visual question answering tasks, they fall significantly behind on our benchmark compared with their closed-source competitors. Furthermore, even GPT-4o exhibits a marked drop in performance on TimeCausality compared to its results on other tasks. These findings underscore the critical need to incorporate temporal causality into the evaluation and development of VLMs, and they highlight an important challenge for the open-source VLM community moving forward. Code and Data are available at \href{https://github.com/Zeqing-Wang/TimeCausality }{TimeCausality}.

Shiyuan Zhang, Jiale Du, Yuanwei Liu et al.

Base station (BS) deployment and operation are fundamental to network performance, yet they require accurate demand understanding, which remains difficult for operators. Cellular traffic in dense urban regions is well measured but highly dynamic, which undermines prediction-based management, whereas the scarcity of traffic measurements in emerging regions limits informed deployment decisions. Existing approaches therefore either depend on manual planning heuristics or use autoregressive predictors that fail to capture stochastic traffic variation. We present NetSpatial, a unified system for cellular planning and operation through spatially conditional traffic generation. NetSpatial exploits multimodal urban context, including satellite imagery and point of interest (POI) distributions, to learn how physical environment and functional semantics shape BS demand. It uses a multi-level flow-matching architecture that separates periodic structure from residual dynamics, enabling direct generation of long-horizon traffic sequences. NetSpatial supports two complementary decision scenarios, i.e., what-if analysis for deployment planning, which ranks candidate sites using generated traffic profiles, and what-to-do support for network operation, which uses generated traffic forecasts to guide BS sleep scheduling and load balancing. Experiments on real-world cellular traffic data show that NetSpatial reduces Jensen-Shannon Divergence (JSD) by 29.44% over the strongest baseline, generalizes across cities in zero-shot experiments, and enables up to 16.8% energy savings while maintaining over 80% quality of experience.

Shiyuan Zhang, Tong Li, Zhu Xiao et al.

Service-level mobile traffic prediction for individual users is essential for network efficiency and quality of service enhancement. However, current prediction methods are limited in their adaptability across different urban environments and produce inaccurate results due to the high uncertainty in personal traffic patterns, the lack of detailed environmental context, and the complex dependencies among different network services. These challenges demand advanced modeling techniques that can capture dynamic traffic distributions and rich environmental features. Inspired by the recent success of diffusion models in distribution modeling and Large Language Models (LLMs) in contextual understanding, we propose an LLM-Enhanced Spatio-temporal Diffusion Model (LSDM). LSDM integrates the generative power of diffusion models with the adaptive learning capabilities of transformers, augmented by the ability to capture multimodal environmental information for modeling service-level patterns and dynamics. Extensive evaluations on real-world service-level datasets demonstrate that the model excels in traffic usage predictions, showing outstanding generalization and adaptability. After incorporating contextual information via LLM, the performance improves by at least 2.83% in terms of the coefficient of determination. Compared to models of a similar type, such as CSDI, the root mean squared error can be reduced by at least 8.29%. The code and dataset will be available at: https://github.com/SoftYuaneR/LSDM.

Haipeng Fan, Shiyuan Zhang, Baohunesitu et al.

Autoregressive (AR) models have achieved unified and strong performance across both visual understanding and image generation tasks. However, removing undesired concepts from AR models while maintaining overall generation quality remains an open challenge. In this paper, we propose Erasure Autoregressive Model (EAR), a fine-tuning method for effective and utility-preserving concept erasure in AR models. Specifically, we introduce Windowed Gradient Accumulation (WGA) strategy to align patch-level decoding with erasure objectives, and Thresholded Loss Masking (TLM) strategy to protect content unrelated to the target concept during fine-tuning. Furthermore, we propose a novel benchmark, Erase Concept Generator and Visual Filter (ECGVF), aim at provide a more rigorous and comprehensive foundation for evaluating concept erasure in AR models. Specifically, we first employ structured templates across diverse large language models (LLMs) to pre-generate a large-scale corpus of target-replacement concept prompt pairs. Subsequently, we generate images from these prompts and subject them to rigorous filtering via a visual classifier to ensure concept fidelity and alignment. Extensive experimental results conducted on the ECGVF benchmark with the AR model Janus-Pro demonstrate that EAR achieves marked improvements in both erasure effectiveness and model utility preservation. Code is available at: https://github.com/immc-lab/ear/

Fei Wang, Jianwei Feng, Yinliang Zhao et al.

Recent years have witnessed the rapid development in the research topic of WiFi sensing that automatically senses human with commercial WiFi devices. This work falls into two major categories, i.e., the activity recognition and the indoor localization. The former work utilizes WiFi devices to recognize human daily activities such as smoking, walking, and dancing. The latter one, indoor localization, can be used for indoor navigation, location-based services, and through-wall surveillance. The key rationale behind this type of work is that people behaviors can influence the WiFi signal propagation and introduce specific patterns into WiFi signals, called WiFi fingerprints, which can be further explored to identify human activities and locations. In this paper, we propose a novel deep learning framework for joint activity recognition and indoor localization task using WiFi Channel State Information~(CSI) fingerprints. More precisely, we develop a system running standard IEEE 802.11n WiFi protocol, and collect more than 1400 CSI fingerprints on 6 activities at 16 indoor locations. Then we propose a dual-task convolutional neural network with 1-dimensional convolutional layers for the joint task of activity recognition and indoor localization. Experimental results and ablation study show that our approach achieves good performances in this joint WiFi sensing task. Data and code have been made publicly available at https://github.com/geekfeiw/apl.

Shiyuan Zhang, Qiwei Di, Xuheng Li et al.

Underdamped Langevin dynamics (ULD) is a widely-used sampler for Gibbs distributions $π\propto e^{-V}$, and is often empirically effective in high dimensions. However, existing non-asymptotic convergence guarantees for discretized ULD typically scale polynomially with the ambient dimension $d$, leading to vacuous bounds when $d$ is large. The main known dimension-free result concerns the randomized midpoint discretization in Wasserstein-2 distance (Liu et al.,2023), while dimension-independent guarantees for ULD discretizations in KL divergence have remained open. We close this gap by proving the first dimension-free KL divergence bounds for discretized ULD. Our analysis refines the KL local error framework (Altschuler et al., 2025) to a dimension-free setting and yields bounds that depend on $\mathrm{tr}(\mathbf{H})$, where $\mathbf{H}$ upper bounds the Hessian of $V$, rather than on $d$. As a consequence, we obtain improved iteration complexity for underdamped Langevin Monte Carlo relative to overdamped Langevin methods in regimes where $\mathrm{tr}(\mathbf{H})\ll d$.

Shiyuan Zhang, Weitong Zhang, Quanquan Gu

This paper investigates energy guidance in generative modeling, where the target distribution is defined as $q(\mathbf x) \propto p(\mathbf x)\exp(-β\mathcal E(\mathbf x))$, with $p(\mathbf x)$ being the data distribution and $\mathcal E(\mathcal x)$ as the energy function. To comply with energy guidance, existing methods often require auxiliary procedures to learn intermediate guidance during the diffusion process. To overcome this limitation, we explore energy-guided flow matching, a generalized form of the diffusion process. We introduce energy-weighted flow matching (EFM), a method that directly learns the energy-guided flow without the need for auxiliary models. Theoretical analysis shows that energy-weighted flow matching accurately captures the guided flow. Additionally, we extend this methodology to energy-weighted diffusion models and apply it to offline reinforcement learning (RL) by proposing the Q-weighted Iterative Policy Optimization (QIPO). Empirically, we demonstrate that the proposed QIPO algorithm improves performance in offline RL tasks. Notably, our algorithm is the first energy-guided diffusion model that operates independently of auxiliary models and the first exact energy-guided flow matching model in the literature.

Junwei Deng, Shiyuan Zhang, Jiaqi Ma

The advancement of generative AI has given rise to pressing copyright challenges, especially within the music industry. This paper focuses on the economic aspects of these challenges, emphasizing that the economic impact constitutes a central issue in the copyright arena. Furthermore, the complexity of the black-box generative AI technologies not only suggests but necessitates algorithmic solutions. Yet, such solutions have been largely missing, exacerbating regulatory hurdles in this landscape. We seek to address this gap by proposing viable royalty models for revenue sharing on AI music generation platforms. We start by examining existing royalty models utilized by platforms like Spotify and YouTube, and then discuss how to adapt them to the unique context of AI-generated music. A significant challenge emerging from this adaptation is the attribution of AI-generated music to influential copyrighted content in the training data. To this end, we present algorithmic solutions employing data attribution techniques. We also conduct a range of experiments to verify the effectiveness and robustness of these solutions. This research is one of the early attempts to integrate technical advancements with economic and legal considerations in the field of music generative AI, offering a computational copyright solution for the challenges posed by the opaque nature of AI technologies.

Hongyin Zhang, Shiyuan Zhang, Junxi Jin et al.

Vision-Language-Action (VLA) models based on flow matching have shown excellent performance in general-purpose robotic manipulation tasks. However, the action accuracy of these models on complex downstream tasks is unsatisfactory. One important reason is that these models rely solely on the post-training paradigm of imitation learning, which makes it difficult to have a deeper understanding of the distribution properties of data quality, which is exactly what Reinforcement Learning (RL) excels at. In this paper, we theoretically propose an offline RL post-training objective for VLA flow models and induce an efficient and feasible offline RL fine-tuning algorithm -- Adaptive Reinforced Flow Matching (ARFM). By introducing an adaptively adjusted scaling factor in the VLA flow model loss, we construct a principled bias-variance trade-off objective function to optimally control the impact of RL signal on flow loss. ARFM adaptively balances RL advantage preservation and flow loss gradient variance control, resulting in a more stable and efficient fine-tuning process. Extensive simulation and real-world experimental results show that ARFM exhibits excellent generalization, robustness, few-shot learning, and continuous learning performance.

Weiyi Wang, Junwei Deng, Yuzheng Hu et al.

Data attribution methods, which quantify the influence of individual training data points on a machine learning model, have gained increasing popularity in data-centric applications in modern AI. Despite a recent surge of new methods developed in this space, the impact of hyperparameter tuning in these methods remains under-explored. In this work, we present the first large-scale empirical study to understand the hyperparameter sensitivity of common data attribution methods. Our results show that most methods are indeed sensitive to certain key hyperparameters. However, unlike typical machine learning algorithms -- whose hyperparameters can be tuned using computationally-cheap validation metrics -- evaluating data attribution performance often requires retraining models on subsets of training data, making such metrics prohibitively costly for hyperparameter tuning. This poses a critical open challenge for the practical application of data attribution methods. To address this challenge, we advocate for better theoretical understandings of hyperparameter behavior to inform efficient tuning strategies. As a case study, we provide a theoretical analysis of the regularization term that is critical in many variants of influence function methods. Building on this analysis, we propose a lightweight procedure for selecting the regularization value without model retraining, and validate its effectiveness across a range of standard data attribution benchmarks. Overall, our study identifies a fundamental yet overlooked challenge in the practical application of data attribution, and highlights the importance of careful discussion on hyperparameter selection in future method development.

Shiyuan Zhang, Junwei Deng, Juhan Bae et al.

Training data attribution (TDA) plays a critical role in understanding the influence of individual training data points on model predictions. Gradient-based TDA methods, popularized by \textit{influence function} for their superior performance, have been widely applied in data selection, data cleaning, data economics, and fact tracing. However, in real-world scenarios where commercial models are not publicly accessible and computational resources are limited, existing TDA methods are often constrained by their reliance on full model access and high computational costs. This poses significant challenges to the broader adoption of TDA in practical applications. In this work, we present a systematic study of TDA methods under various access and resource constraints. We investigate the feasibility of performing TDA under varying levels of access constraints by leveraging appropriately designed solutions such as proxy models. Besides, we demonstrate that attribution scores obtained from models without prior training on the target dataset remain informative across a range of tasks, which is useful for scenarios where computational resources are limited. Our findings provide practical guidance for deploying TDA in real-world environments, aiming to improve feasibility and efficiency under limited access.

Haoye Chai, Shiyuan Zhang, Xiaoqian Qi et al.

Mobile traffic forecasting allows operators to anticipate network dynamics and performance in advance, offering substantial potential for enhancing service quality and improving user experience. However, existing models are often task-oriented and are trained with tailored data, which limits their effectiveness in diverse mobile network tasks of Base Station (BS) deployment, resource allocation, energy optimization, etc. and hinders generalization across different urban environments. Foundation models have made remarkable strides across various domains of NLP and CV due to their multi-tasking adaption and zero/few-shot learning capabilities. In this paper, we propose an innovative Foundation model for Mo}bile traffic forecasting (FoMo), aiming to handle diverse forecasting tasks of short/long-term predictions and distribution generation across multiple cities to support network planning and optimization. FoMo combines diffusion models and transformers, where various spatio-temporal masks are proposed to enable FoMo to learn intrinsic features of different tasks, and a contrastive learning strategy is developed to capture the correlations between mobile traffic and urban contexts, thereby improving its transfer learning capability. Extensive experiments on 9 real-world datasets demonstrate that FoMo outperforms current models concerning diverse forecasting tasks and zero/few-shot learning, showcasing a strong universality.

Fei Wang, Jinsong Han, Shiyuan Zhang et al.

We build CSI-Net, a unified Deep Neural Network~(DNN), to learn the representation of WiFi signals. Using CSI-Net, we jointly solved two body characterization problems: biometrics estimation (including body fat, muscle, water, and bone rates) and person recognition. We also demonstrated the application of CSI-Net on two distinctive pose recognition tasks: the hand sign recognition (fine-scaled action of the hand) and falling detection (coarse-scaled motion of the body).