Ziliang Wang

Hao Zhu, Jia Li, Cuiyun Gao et al. · pku

Large language models (LLMs) have achieved remarkable progress in code understanding tasks. However, they demonstrate limited performance in vulnerability detection and struggle to distinguish vulnerable code from patched code. We argue that LLMs lack understanding of security specifications -- the expectations about how code should behave to remain safe. When code behavior differs from these expectations, it becomes a potential vulnerability. However, such knowledge is rarely explicit in training data, leaving models unable to reason about security flaws. We propose VulInstruct, a specification-guided approach that systematically extracts security specifications from historical vulnerabilities to detect new ones. VulInstruct constructs a specification knowledge base from two perspectives: (i) General specifications from high-quality patches across projects, capturing fundamental safe behaviors; and (ii) Domain-specific specifications from repeated violations in particular repositories relevant to the target code. VulInstruct retrieves relevant past cases and specifications, enabling LLMs to reason about expected safe behaviors rather than relying on surface patterns. We evaluate VulInstruct under strict criteria requiring both correct predictions and valid reasoning. On PrimeVul, VulInstruct achieves 45.0% F1-score (32.7% improvement) and 37.7% recall (50.8% improvement) compared to baselines, while uniquely detecting 24.3% of vulnerabilities -- 2.4x more than any baseline. In pair-wise evaluation, VulInstruct achieves 32.3% relative improvement. VulInstruct also discovered a previously unknown high-severity vulnerability (CVE-2025-56538) in production code, demonstrating practical value for real-world vulnerability discovery. All code and supplementary materials are available at https://github.com/zhuhaopku/VulInstruct-temp.

Xuesi Wang, Junjie Wang, Ziliang Wang et al.

Semantic IDs represent items as shared discrete token sequences and have become a practical tool for recommendation and retrieval. Yet it remains difficult to tell why a tokenizer fails: poor quality may come from codebook underutilization, unstable decision boundaries, or geometric distortion of the embedding space. This paper develops a quantitative framework for diagnosing these failures through expected codeword overlap and effective codebook capacity. The former measures expected codeword confusion under retrieval-time perturbation, while the latter converts that confusion into an effective number of usable, well-separated codes. The framework links semantic boundary confusion to both code usage imbalance and Euclidean geometric constraints. As a proof of concept, we present Decoupled Residual Quantization (DRQ), which separates continuous geometry reconstruction from discrete distribution matching. Experiments on a large-scale industrial dataset show that Semantic ID quality is multi-objective: symbolic robustness, reconstruction fidelity, and behavior-aware soft matching each stress different aspects of a tokenizer. These downstream observations are based on one proprietary industrial dataset, so they should be read as a case study rather than a universal benchmark claim.

Ziliang Wang, Gaoyun Lin, Xuesi Wang et al.

Generative Recommendation (GR) reframes retrieval and ranking as autoregressive decoding over Semantic IDs (SIDs), unifying the multi-stage pipeline into a single model. Yet a fundamental expressive gap persists: discriminative models score items with direct feature access enabling explicit user-item crossing, whereas GR decodes over compact SID tokens without item-side signal. We formalize this via Bayes' theorem: ranking by p(y|f,u) is equivalent to ranking by p(f|y,u), which factorizes autoregressively over item features, showing that a generative model with full feature access matches its discriminative counterpart, with any practical gap stemming solely from incomplete feature coverage.We propose UniRec with Chain-of-Attribute (CoA) as its core mechanism. CoA prefixes each SID sequence with structured attribute tokens:category, seller, brand, before decoding the SID, recovering the item-side feature crossing that discriminative models exploit. Since items sharing identical attributes cluster in adjacent SID regions, attribute conditioning yields a measurable per-step entropy reduction H(s_k|s<k,a) < H(s_k|s<k), narrowing the search space and stabilizing beam search. We further address two deployment challenges: Capacity-constrained SID introduces exposure-weighted capacity penalties into residual quantization to suppress token collapse and the Matthew effect; Conditional Decoding Context (CDC) combines Task-Conditioned BOS with hash-based Content Summaries to inject scenario signals at each decoding step. A joint RFT and DPO framework aligns the model with business objectives beyond distribution matching.Experiments show UniRec outperforms the strongest baseline by +22.6% HR@50 overall and +15.5% on high-value orders. Deployed on Shopee's e-commerce platform, online A/B tests confirm significant gains in PVCTR (+5.37%), orders (+4.76%), and GMV (+5.60%).

Ziliang Wang, Xuhui Zheng, Kang An et al.

Efficient multi-hop reasoning requires Large Language Models (LLMs) based agents to acquire high-value external knowledge iteratively. Previous work has explored reinforcement learning (RL) to train LLMs to perform search-based document retrieval, achieving notable improvements in QA performance, but underperform on complex, multi-hop QA resulting from the sparse rewards from global signal only. To address this gap in existing research, we introduce StepSearch, a framework for search LLMs that trained with step-wise proximal policy optimization method. It consists of richer and more detailed intermediate search rewards and token-level process supervision based on information gain and redundancy penalties to better guide each search step. We constructed a fine-grained question-answering dataset containing sub-question-level search trajectories based on open source datasets through a set of data pipeline method. On standard multi-hop QA benchmarks, it significantly outperforms global-reward baselines, achieving 11.2% and 4.2% absolute improvements for 3B and 7B models over various search with RL baselines using only 19k training data, demonstrating the effectiveness of fine-grained, stepwise supervision in optimizing deep search LLMs. Our code will be released on https://github.com/Zillwang/StepSearch.

Zezeng Li, Weimin Wang, Ziliang Wang et al.

This paper presents a novel point cloud compression method COT-PCC by formulating the task as a constrained optimal transport (COT) problem. COT-PCC takes the bitrate of compressed features as an extra constraint of optimal transport (OT) which learns the distribution transformation between original and reconstructed points. Specifically, the formulated COT is implemented with a generative adversarial network (GAN) and a bitrate loss for training. The discriminator measures the Wasserstein distance between input and reconstructed points, and a generator calculates the optimal mapping between distributions of input and reconstructed point cloud. Moreover, we introduce a learnable sampling module for downsampling in the compression procedure. Extensive results on both sparse and dense point cloud datasets demonstrate that COT-PCC outperforms state-of-the-art methods in terms of both CD and PSNR metrics. Source codes are available at \url{https://github.com/cognaclee/PCC-COT}.

Ziliang Wang, Xiaohong Zhang, Kechi Zhang et al.

Individual objects, whether users or services, within a specific region often exhibit similar network states due to their shared origin from the same city or autonomous system (AS). Despite this regional network similarity, many existing techniques overlook its potential, resulting in subpar performance arising from challenges such as data sparsity and label imbalance. In this paper, we introduce the regional-based dual latent state learning network(R2SL), a novel deep learning framework designed to overcome the pitfalls of traditional individual object-based prediction techniques in Quality of Service (QoS) prediction. Unlike its predecessors, R2SL captures the nuances of regional network behavior by deriving two distinct regional network latent states: the city-network latent state and the AS-network latent state. These states are constructed utilizing aggregated data from common regions rather than individual object data. Furthermore, R2SL adopts an enhanced Huber loss function that adjusts its linear loss component, providing a remedy for prevalent label imbalance issues. To cap off the prediction process, a multi-scale perception network is leveraged to interpret the integrated feature map, a fusion of regional network latent features and other pertinent information, ultimately accomplishing the QoS prediction. Through rigorous testing on real-world QoS datasets, R2SL demonstrates superior performance compared to prevailing state-of-the-art methods. Our R2SL approach ushers in an innovative avenue for precise QoS predictions by fully harnessing the regional network similarities inherent in objects.

Ziliang Wang, Xiaohong Zhang, Ze Shi Li et al.

Accurate Quality of Service (QoS) prediction is essential for enhancing user satisfaction in web recommendation systems, yet existing prediction models often overlook feature noise, focusing predominantly on label noise. In this paper, we present the Probabilistic Deep Supervision Network (PDS-Net), a robust framework designed to effectively identify and mitigate feature noise, thereby improving QoS prediction accuracy. PDS-Net operates with a dual-branch architecture: the main branch utilizes a decoder network to learn a Gaussian-based prior distribution from known features, while the second branch derives a posterior distribution based on true labels. A key innovation of PDS-Net is its condition-based noise recognition loss function, which enables precise identification of noisy features in objects (users or services). Once noisy features are identified, PDS-Net refines the feature's prior distribution, aligning it with the posterior distribution, and propagates this adjusted distribution to intermediate layers, effectively reducing noise interference. Extensive experiments conducted on two real-world QoS datasets demonstrate that PDS-Net consistently outperforms existing models, achieving an average improvement of 8.91% in MAE on Dataset D1 and 8.32% on Dataset D2 compared to the ate-of-the-art. These results highlight PDS-Net's ability to accurately capture complex user-service relationships and handle feature noise, underscoring its robustness and versatility across diverse QoS prediction environments.

Shaowei Zhang, Faqiang Qian, Yan Chen et al.

Emotion Recognition in Conversation (ERC) has become a fundamental capability for large language models (LLMs) in human-centric interaction. Beyond accurate recognition, coherent emotional expression is also crucial, yet both are limited by the scarcity and static nature of high-quality annotated data. In this work, we propose SELF-EMO, a self-evolution framework grounded in the hypothesis that better emotion prediction leads to more consistent emotional responses. We introduce two auxiliary tasks, emotional understanding and emotional expression, and design a role-based self-play paradigm where the model acts as both an emotion recognizer and a dialogue responder. Through iterative interactions, the model generates diverse conversational trajectories, enabling scalable data generation. To ensure quality, we adopt a data flywheel mechanism that filters candidate predictions and responses using a smoothed IoU-based reward and feeds selected samples back for continuous self-improvement without external supervision. We further develop SELF-GRPO, a reinforcement learning algorithm that stabilizes optimization with multi-label alignment rewards and group-level consistency signals. Experiments on IEMOCAP, MELD, and EmoryNLP show that SELF-EMO achieves state-of-the-art performance, improving accuracy by +6.33% on Qwen3-4B and +8.54% on Qwen3-8B, demonstrating strong effectiveness and generalization.

Xuhui Zheng, Kang An, Ziliang Wang et al.

Multimodal pre-training remains constrained by the descriptive bias of image-caption pairs, leading models to favor surface linguistic cues over grounded visual understanding. We introduce MMRPT, a masked multimodal reinforcement pre-training framework that strengthens visual reasoning in MLLMs. We are the first to incorporate reinforcement learning directly into the pre-training of large vision-language models, enabling learning signals that reward visual grounding rather than caption imitation. MMRPT constructs masked multimodal data by estimating sentence-level visual dependency via attention over visual tokens and masking highly vision-dependent segments; the model reconstructs these spans through vision-grounded reasoning guided by a semantic-visual reward. Experiments show consistent zero-shot gains across diverse benchmarks and substantially improved robustness under supervised fine-tuning, demonstrating that reinforcement-driven masked reasoning provides a more reliable and generalizable pre-training objective for multimodal models.

Ke Xu, Ziliang Wang, Wei Zheng et al.

Cross-domain recommendation (CDR) methods are proposed to tackle the sparsity problem in click through rate (CTR) estimation. Existing CDR methods directly transfer knowledge from the source domains to the target domain and ignore the heterogeneities among domains, including feature dimensional heterogeneity and latent space heterogeneity, which may lead to negative transfer. Besides, most of the existing methods are based on single-source transfer, which cannot simultaneously utilize knowledge from multiple source domains to further improve the model performance in the target domain. In this paper, we propose a centralized-distributed transfer model (CDTM) for CDR based on multi-source heterogeneous transfer learning. To address the issue of feature dimension heterogeneity, we build a dual embedding structure: domain specific embedding (DSE) and global shared embedding (GSE) to model the feature representation in the single domain and the commonalities in the global space,separately. To solve the latent space heterogeneity, the transfer matrix and attention mechanism are used to map and combine DSE and GSE adaptively. Extensive offline and online experiments demonstrate the effectiveness of our model.

Ziliang Wang, Xiaohong Zhang, Ze Shi Li et al.

Accurate prediction of Quality of Service (QoS) metrics is fundamental for selecting and managing cloud based services. Traditional QoS models rely on manual feature engineering and yield only point estimates, offering no insight into the confidence of their predictions. In this paper, we propose QoSBERT, the first framework that reformulates QoS prediction as a semantic regression task based on pre trained language models. Unlike previous approaches relying on sparse numerical features, QoSBERT automatically encodes user service metadata into natural language descriptions, enabling deep semantic understanding. Furthermore, we integrate a Monte Carlo Dropout based uncertainty estimation module, allowing for trustworthy and risk-aware service quality prediction, which is crucial yet underexplored in existing QoS models. QoSBERT applies attentive pooling over contextualized embeddings and a lightweight multilayer perceptron regressor, fine tuned jointly to minimize absolute error. We further exploit the resulting uncertainty estimates to select high quality training samples, improving robustness in low resource settings. On standard QoS benchmark datasets, QoSBERT achieves an average reduction of 11.7% in MAE and 6.7% in RMSE for response time prediction, and 6.9% in MAE for throughput prediction compared to the strongest baselines, while providing well calibrated confidence intervals for robust and trustworthy service quality estimation. Our approach not only advances the accuracy of service quality prediction but also delivers reliable uncertainty quantification, paving the way for more trustworthy, data driven service selection and optimization.

Zezeng Li, Zhihui Qi, Weimin Wang et al.

Quad meshes are essential in geometric modeling and computational mechanics. Although learning-based methods for triangle mesh demonstrate considerable advancements, quad mesh generation remains less explored due to the challenge of ensuring coplanarity, convexity, and quad-only meshes. In this paper, we present Point2Quad, the first learning-based method for quad-only mesh generation from point clouds. The key idea is learning to identify quad mesh with fused pointwise and facewise features. Specifically, Point2Quad begins with a k-NN-based candidate generation considering the coplanarity and squareness. Then, two encoders are followed to extract geometric and topological features that address the challenge of quad-related constraints, especially by combining in-depth quadrilaterals-specific characteristics. Subsequently, the extracted features are fused to train the classifier with a designed compound loss. The final results are derived after the refinement by a quad-specific post-processing. Extensive experiments on both clear and noise data demonstrate the effectiveness and superiority of Point2Quad, compared to baseline methods under comprehensive metrics.

FaQiang Qian, WeiKun Zhang, Ziliang Wang et al.

Shaping powerful LLMs to be beneficial and safe is central to AI alignment. We argue that post-training alignment is fundamentally a unified Preference Learning problem, involving two modalities: demonstrated preferences (e.g., Supervised Fine-Tuning, SFT) and comparative preferences (e.g., Reinforcement Learning, RL).The standard sequential pipeline-SFT followed by RL-is flawed due to a critical distributional mismatch: SFT uses static expert data, but as the policy evolves, its generation distribution drifts, making SFT knowledge brittle. Subsequent RL then explores without direct access to the rich, ground-truth knowledge in expert demonstrations, leading to inefficient, ungrounded updates. This separation prevents mutual regularization between data sources. To address this, we reframe alignment as a constrained optimization problem and propose Unified Adversarial Preference Learning (UniAPL),a novel framework that dynamically aligns the policy's distribution with the expert's. UniAPL implements a single-stage unified training objective, jointly learning from mixed batches of SFT and preference data. In every gradient step, dense expert demonstrations directly ground and regularize online exploration, inherently resolving distributional mismatch and maximizing data synergy.We evaluate UniAPL on instruction-following tasks using Qwen3-235B-Instruct-2507 as the teacher. Our models match or exceed strong GRPO baselines: +5.77% on Qwen3-0.6B (matching a 32B model) and +3.75% on Qwen3-4B,even outperforming the teacher. Analyses of response length and log-probability distributions confirm that UniAPL outputs closely mimic expert demonstrations, achieving both stronger performance and better behavioral alignment.

Weixin Xu, Ziliang Wang

Medical ultrasound image segmentation presents a formidable challenge in the realm of computer vision. Traditional approaches rely on Convolutional Neural Networks (CNNs) and Transformer-based methods to address the intricacies of medical image segmentation. Nevertheless, inherent limitations persist, as CNN-based methods tend to disregard long-range dependencies, while Transformer-based methods may overlook local contextual information. To address these deficiencies, we propose a novel Feature Aggregation Module (FAM) designed to process two input features from the preceding layer. These features are seamlessly directed into two branches of the Convolution and Cross-Attention Parallel Module (CCAPM) to endow them with different roles in each of the two branches to help establish a strong connection between the two input features. This strategy enables our module to focus concurrently on both long-range dependencies and local contextual information by judiciously merging convolution operations with cross-attention mechanisms. Moreover, by integrating FAM within our proposed Spatial-Channel Regulation Module (SCRM), the ability to discern salient regions and informative features warranting increased attention is enhanced. Furthermore, by incorporating the SCRM into the encoder block of the UNet architecture, we introduce a novel framework dubbed Spatial-Channel Regulation Network (SCRNet). The results of our extensive experiments demonstrate the superiority of SCRNet, which consistently achieves state-of-the-art (SOTA) performance compared to existing methods.

Yuting Lu, Ziliang Wang, Weixin Xu et al.

Clinical deployment requires segmentation models to stay stable under distribution shifts and perturbations. The mainstream solution is adversarial training (AT) to improve robustness; however, AT often brings a clean--robustness trade-off and high training/tuning cost, which limits scalability and maintainability in medical imaging. We propose \emph{Layer-wise Noise-Guided Selective Wavelet Reconstruction (LNG-SWR)}. During training, we inject small, zero-mean noise at multiple layers to learn a frequency-bias prior that steers representations away from noise-sensitive directions. We then apply prior-guided selective wavelet reconstruction on the input/feature branch to achieve frequency adaptation: suppress noise-sensitive bands, enhance directional structures and shape cues, and stabilize boundary responses while maintaining spectral consistency. The framework is backbone-agnostic and adds low additional inference overhead. It can serve as a plug-in enhancement to AT and also improves robustness without AT. On CT and ultrasound datasets, under a unified protocol with PGD-$L_{\infty}/L_{2}$ and SSAH, LNG-SWR delivers consistent gains on clean Dice/IoU and significantly reduces the performance drop under strong attacks; combining LNG-SWR with AT yields additive gains. When combined with adversarial training, robustness improves further without sacrificing clean accuracy, indicating an engineering-friendly and scalable path to robust segmentation. These results indicate that LNG-SWR provides a simple, effective, and engineering-friendly path to robust medical image segmentation in both adversarial and standard training regimes.

Ziliang Wang, Kang An, Xuhui Zheng et al.

While search-augmented large language models (LLMs) exhibit impressive capabilities, their reliability in complex multi-hop reasoning remains limited. This limitation arises from three fundamental challenges: decomposition errors, where tasks are incorrectly broken down; retrieval missing, where key evidence fails to be retrieved; and reasoning errors, where flawed logic propagates through the reasoning chain. A single failure in any of these stages can derail the final answer. We propose Erasable Reinforcement Learning (ERL), a novel framework that transforms fragile reasoning into a robust process. ERL explicitly identifies faulty steps, erases them, and regenerates reasoning in place, preventing defective logic from propagating through the reasoning chain. This targeted correction mechanism turns brittle reasoning into a more resilient process. Models trained with ERL, termed ESearch, achieve substantial improvements on HotpotQA, MuSiQue, 2Wiki, and Bamboogle, with the 3B model achieving +8.48% EM and +11.56% F1, and the 7B model achieving +5.38% EM and +7.22% F1 over previous state-of-the-art(SOTA) results. These findings suggest that erasable reinforcement learning provides a powerful paradigm shift for robust multi-step reasoning in LLMs.

Xingyi Li, Han Zhang, Ziliang Wang et al.

Point cloud registration is a key step in robotic perception tasks, such as Simultaneous Localization and Mapping (SLAM). It is especially challenging in conditions with sparse points and heavy noise. Traditional registration methods, such as Iterative Closest Point (ICP) and Normal Distributions Transform (NDT), often have difficulties in achieving a robust and accurate alignment under these conditions. In this paper, we propose a registration framework based on moment matching. In particular, the point clouds are regarded as i.i.d. samples drawn from the same distribution observed in the source and target frames. We then match the generalized Gaussian Radial Basis moments calculated from the point clouds to estimate the rigid transformation between two frames. Moreover, such method does not require explicit point-to-point correspondences among the point clouds. We further show the consistency of the proposed method. Experiments on synthetic and real-world datasets show that our approach achieves higher accuracy and robustness than existing methods. In addition, we integrate our framework into a 4D Radar SLAM system. The proposed method significantly improves the localization performance and achieves results comparable to LiDAR-based systems. These findings demonstrate the potential of moment matching technique for robust point cloud registration in sparse and noisy scenarios.

Guangming Wang, Honghao Zeng, Ziliang Wang et al.

Real-time 3D human pose estimation is crucial for human-computer interaction. It is cheap and practical to estimate 3D human pose only from monocular video. However, recent bone splicing based 3D human pose estimation method brings about the problem of cumulative error. In this paper, the concept of virtual bones is proposed to solve such a challenge. The virtual bones are imaginary bones between non-adjacent joints. They do not exist in reality, but they bring new loop constraints for the estimation of 3D human joints. The proposed network in this paper predicts real bones and virtual bones, simultaneously. The final length of real bones is constrained and learned by the loop constructed by the predicted real bones and virtual bones. Besides, the motion constraints of joints in consecutive frames are considered. The consistency between the 2D projected position displacement predicted by the network and the captured real 2D displacement by the camera is proposed as a new projection consistency loss for the learning of 3D human pose. The experiments on the Human3.6M dataset demonstrate the good performance of the proposed method. Ablation studies demonstrate the effectiveness of the proposed inter-frame projection consistency constraints and intra-frame loop constraints.

Wenyan Zhang, Ling Xu, Meng Yan et al.

In recent years, the number of online services has grown rapidly, invoke the required services through the cloud platform has become the primary trend. How to help users choose and recommend high-quality services among huge amounts of unused services has become a hot issue in research. Among the existing QoS prediction methods, the collaborative filtering(CF) method can only learn low-dimensional linear characteristics, and its effect is limited by sparse data. Although existing deep learning methods could capture high-dimensional nonlinear features better, most of them only use the single feature of identity, and the problem of network deepening gradient disappearance is serious, so the effect of QoS prediction is unsatisfactory. To address these problems, we propose an advanced probability distribution and location-aware ResNet approach for QoS Prediction(PLRes). This approach considers the historical invocations probability distribution and location characteristics of users and services, and first use the ResNet in QoS prediction to reuses the features, which alleviates the problems of gradient disappearance and model degradation. A series of experiments are conducted on a real-world web service dataset WS-DREAM. The results indicate that PLRes model is effective for QoS prediction and at the density of 5%-30%, which means the data is sparse, it significantly outperforms a state-of-the-art approach LDCF by 12.35%-15.37% in terms of MAE.