Searcharxiv⌕ Search

arXiv subjects

Qi Zhang

Publications and source records attributed to Qi Zhang.

At least 343 records · Page 19Linked to original sources
arXiv2025

BAPO: Stabilizing Off-Policy Reinforcement Learning for LLMs via Balanced Policy Optimization with Adaptive Clipping

Reinforcement learning (RL) has recently become the core paradigm for aligning and strengthening large language models (LLMs). Yet, applying RL in off-policy settings--where stale data from past policies are used for training--improves sample efficiency, but remains challenging: policy entropy declines sharply, optimization often becomes unstable and may even collapse. Through theoretical and empirical analysis, we identify two key insights: (i) an imbalance in optimization, where negative-advantage samples dominate the policy gradient, suppressing useful behaviors and risking gradient explosions; and (ii) the derived Entropy-Clip Rule, which reveals that the fixed clipping mechanism in PPO-like objectives systematically blocks entropy-increasing updates, thereby driving the policy toward over-exploitation at the expense of exploration. Building on these insights, we propose BAlanced Policy Optimization with Adaptive Clipping (BAPO), a simple yet effective method that dynamically adjusts clipping bounds to adaptively re-balance positive and negative contributions, preserve entropy, and stabilize RL optimization. Across diverse off-policy scenarios--including sample replay and partial rollout--BAPO achieves fast, stable, and data-efficient training. On AIME 2024 and AIME 2025 benchmarks, our 7B BAPO model surpasses open-source counterparts such as SkyWork-OR1-7B, while our 32B BAPO model not only achieves state-of-the-art results among models of the same scale but also outperforms leading proprietary systems like o3-mini and Gemini-2.5-Flash-Thinking.

cs.LG↗
arXiv2025

BokehDiff: Neural Lens Blur with One-Step Diffusion

We introduce BokehDiff, a novel lens blur rendering method that achieves physically accurate and visually appealing outcomes, with the help of generative diffusion prior. Previous methods are bounded by the accuracy of depth estimation, generating artifacts in depth discontinuities. Our method employs a physics-inspired self-attention module that aligns with the image formation process, incorporating depth-dependent circle of confusion constraint and self-occlusion effects. We adapt the diffusion model to the one-step inference scheme without introducing additional noise, and achieve results of high quality and fidelity. To address the lack of scalable paired data, we propose to synthesize photorealistic foregrounds with transparency with diffusion models, balancing authenticity and scene diversity.

cs.CV↗
arXiv2025

HGC-Avatar: Hierarchical Gaussian Compression for Streamable Dynamic 3D Avatars

Recent advances in 3D Gaussian Splatting (3DGS) have enabled fast, photorealistic rendering of dynamic 3D scenes, showing strong potential in immersive communication. However, in digital human encoding and transmission, the compression methods based on general 3DGS representations are limited by the lack of human priors, resulting in suboptimal bitrate efficiency and reconstruction quality at the decoder side, which hinders their application in streamable 3D avatar systems. We propose HGC-Avatar, a novel Hierarchical Gaussian Compression framework designed for efficient transmission and high-quality rendering of dynamic avatars. Our method disentangles the Gaussian representation into a structural layer, which maps poses to Gaussians via a StyleUNet-based generator, and a motion layer, which leverages the SMPL-X model to represent temporal pose variations compactly and semantically. This hierarchical design supports layer-wise compression, progressive decoding, and controllable rendering from diverse pose inputs such as video sequences or text. Since people are most concerned with facial realism, we incorporate a facial attention mechanism during StyleUNet training to preserve identity and expression details under low-bitrate constraints. Experimental results demonstrate that HGC-Avatar provides a streamable solution for rapid 3D avatar rendering, while significantly outperforming prior methods in both visual quality and compression efficiency.

cs.CV↗
arXiv2025

What's Wrong with Your Code Generated by Large Language Models? An Extensive Study

The increasing development of LLMs in code generation has drawn significant attention among researchers. To enhance LLM-based code generation ability, current efforts are predominantly directed towards collecting high-quality datasets and leveraging diverse training technologies. However, there is a notable lack of comprehensive studies examining the limitations and boundaries of existing methods. To bridge this gap, we conducted an extensive empirical study evaluating the performance of three leading closed-source LLMs and six popular open-source LLMs on three commonly used benchmarks. Our investigation, which evaluated the length, cyclomatic complexity and API number of the generated code, revealed that these LLMs face challenges in generating successful code for more complex problems, and tend to produce code that is shorter yet more complicated as compared to canonical solutions. Additionally, we developed a taxonomy of bugs for incorrect codes that includes three categories and ten sub-categories, and analyzed the root cause for common bug types. To better understand the performance of LLMs in real-world projects, we also manually created a real-world benchmark RWPB. We analyzed bugs on RWPB to highlight distinct differences in bug distributions between actual scenarios and existing benchmarks. Finally, we propose a novel training-free iterative method that introduces self-critique, enabling LLMs to critique and correct their generated code based on bug types and compiler feedback. Our comprehensive and extensive study provides insights into the current limitations of LLM-based code generation and opportunities for enhancing the accuracy and quality of the generated code.

cs.SE↗
arXiv2025

FTIN: Frequency-Time Integration Network for Inertial Odometry

Inertial odometry (IO) leverages inertial measurement unit (IMU) signals for cost-effective localization. However, high IMU sampling rates introduce substantial redundancy that impedes IO's ability to attend to salient components, thereby creating an information bottleneck. To address this challenge, we propose a cross-domain IO framework that fuses information from the frequency and time domains. Specifically, we exploit the global context and energy-compaction properties of frequency-domain representations to capture holistic motion patterns and alleviate the bottleneck. To the best of our knowledge, this is among the first attempts to incorporate frequency-domain feature processing into IO. Experimental results on multiple public datasets demonstrate the effectiveness of the proposed frequency--time-domain fusion strategy.

cs.RO↗
arXiv2025

CKANIO: Learnable Chebyshev Polynomials for Inertial Odometry

Inertial odometry (IO) relies exclusively on signals from an inertial measurement unit (IMU) for localization and offers a promising avenue for consumer grade positioning. However, accurate modeling of the nonlinear motion patterns present in IMU signals remains the principal limitation on IO accuracy. To address this challenge, we propose CKANIO, an IO framework that integrates Chebyshev based Kolmogorov-Arnold Networks (Chebyshev KAN). Specifically, we design a novel residual architecture that leverages the nonlinear approximation capabilities of Chebyshev polynomials within the KAN framework to more effectively model the complex motion characteristics inherent in IMU signals. To the best of our knowledge, this work represents the first application of an interpretable KAN model to IO. Experimental results on five publicly available datasets demonstrate the effectiveness of CKANIO.

cs.RO↗
arXiv2025

High-efficiency and long-distance quantum memory-assisted device-independent quantum secret sharing with single photon sources

Quantum secret sharing (QSS) plays a critical role in building the distributed quantum networks. Device-independent (DI) QSS provides the highest security level for QSS. However, the photon transmission loss and extremely low multipartite entanglement generation rate largely limit DI QSS's secure photon transmission distance (less than 1 km) and practical key generation efficiency. To address the above drawbacks, we propose the quantum memory-assisted (QMA) DI QSS protocol based on single photon sources (SPSs). The single photons from the SPSs are used to construct long-distance multipartite entanglement channels with the help of the heralded architecture. The heralded architecture enables our protocol to have an infinite secure photon transmission distance in theory. The QMA technology can not only increase the multi-photon synchronization efficiency, but also optimize the photon transmittance to maximize the construction efficiency of the multipartite entanglement channels. Our protocol achieves the practical key generation efficiency seven orders of magnitude higher than that of the existing DI QSS protocols based on cascaded spontaneous parametric down-conversion sources and six orders of magnitude higher than that of the DI QSS based on SPSs without QMA. Our protocol has modular characteristics and is feasible under the current experimental technical conditions. Combining with the advanced random key generation basis strategy, the requirement on experimental devices can be effectively reduced. Our protocol is expected to promote the development of long-distance and high-efficiency DI quantum network in the future.

quant-ph↗
arXiv2025

Digitized Counterdiabatic Quantum Feature Extraction

We introduce a Hamiltonian-based quantum feature extraction method that generates complex features via the dynamics of $k$-local many-body spins Hamiltonians, enhancing machine learning performance. Classical feature vectors are embedded into spin-glass Hamiltonians, where both single-variable contributions and higher-order correlations are represented through many-body interactions. By evolving the system under suitable quantum dynamics on IBM digital quantum processors with 156 qubits, the data are mapped into a higher-dimensional feature space via expectation values of low- and higher-order observables. This allows us to capture statistical dependencies that are difficult to access with standard classical methods. We assess the approach on high-dimensional, real-world datasets, including molecular toxicity classification and image recognition, and analyze feature importance to show that quantum-extracted features complement and, in many cases, surpass classical ones. The results suggest that combining quantum and classical feature extraction can provide consistent improvements across diverse machine learning tasks, indicating a reliable level of early quantum usefulness for near-term quantum devices in data-driven applications.

quant-ph↗
arXiv2025

Boosting Adversarial Transferability via Commonality-Oriented Gradient Optimization

Exploring effective and transferable adversarial examples is vital for understanding the characteristics and mechanisms of Vision Transformers (ViTs). However, adversarial examples generated from surrogate models often exhibit weak transferability in black-box settings due to overfitting. Existing methods improve transferability by diversifying perturbation inputs or applying uniform gradient regularization within surrogate models, yet they have not fully leveraged the shared and unique features of surrogate models trained on the same task, leading to suboptimal transfer performance. Therefore, enhancing perturbations of common information shared by surrogate models and suppressing those tied to individual characteristics offers an effective way to improve transferability. Accordingly, we propose a commonality-oriented gradient optimization strategy (COGO) consisting of two components: Commonality Enhancement (CE) and Individuality Suppression (IS). CE perturbs the mid-to-low frequency regions, leveraging the fact that ViTs trained on the same dataset tend to rely more on mid-to-low frequency information for classification. IS employs adaptive thresholds to evaluate the correlation between backpropagated gradients and model individuality, assigning weights to gradients accordingly. Extensive experiments demonstrate that COGO significantly improves the transfer success rates of adversarial attacks, outperforming current state-of-the-art methods.

cs.CV↗
arXiv2025

IONext: Unlocking the Next Era of Inertial Odometry

Researchers have increasingly adopted Transformer-based models for inertial odometry. While Transformers excel at modeling long-range dependencies, their limited sensitivity to local, fine-grained motion variations and lack of inherent inductive biases often hinder localization accuracy and generalization. Recent studies have shown that incorporating large-kernel convolutions and Transformer-inspired architectural designs into CNN can effectively expand the receptive field, thereby improving global motion perception. Motivated by these insights, we propose a novel CNN-based module called the Dual-wing Adaptive Dynamic Mixer (DADM), which adaptively captures both global motion patterns and local, fine-grained motion features from dynamic inputs. This module dynamically generates selective weights based on the input, enabling efficient multi-scale feature aggregation. To further improve temporal modeling, we introduce the Spatio-Temporal Gating Unit (STGU), which selectively extracts representative and task-relevant motion features in the temporal domain. This unit addresses the limitations of temporal modeling observed in existing CNN approaches. Built upon DADM and STGU, we present a new CNN-based inertial odometry backbone, named Next Era of Inertial Odometry (IONext). Extensive experiments on six public datasets demonstrate that IONext consistently outperforms state-of-the-art (SOTA) Transformer- and CNN-based methods. For instance, on the RNIN dataset, IONext reduces the average ATE by 10% and the average RTE by 12% compared to the representative model iMOT.

cs.CV↗
arXiv2025

VCoT-Grasp: Grasp Foundation Models with Visual Chain-of-Thought Reasoning for Language-driven Grasp Generation

Robotic grasping is one of the most fundamental tasks in robotic manipulation, and grasp detection/generation has long been the subject of extensive research. Recently, language-driven grasp generation has emerged as a promising direction due to its practical interaction capabilities. However, most existing approaches either lack sufficient reasoning and generalization capabilities or depend on complex modular pipelines. Moreover, current grasp foundation models tend to overemphasize dialog and object semantics, resulting in inferior performance and restriction to single-object grasping. To maintain strong reasoning ability and generalization in cluttered environments, we propose VCoT-Grasp, an end-to-end grasp foundation model that incorporates visual chain-of-thought reasoning to enhance visual understanding for grasp generation. VCoT-Grasp adopts a multi-turn processing paradigm that dynamically focuses on visual inputs while providing interpretable reasoning traces. For training, we refine and introduce a large-scale dataset, VCoT-GraspSet, comprising 167K synthetic images with over 1.36M grasps, as well as 400+ real-world images with more than 1.2K grasps, annotated with intermediate bounding boxes. Extensive experiments on both VCoT-GraspSet and real robot demonstrate that our method significantly improves grasp success rates and generalizes effectively to unseen objects, backgrounds, and distractors. More details can be found at https://zhanghr2001.github.io/VCoT-Grasp.github.io.

cs.RO↗
arXiv2025

Towards Economical Inference: Enabling DeepSeek's Multi-Head Latent Attention in Any Transformer-based LLMs

Multi-head Latent Attention (MLA) is an innovative architecture proposed by DeepSeek, designed to ensure efficient and economical inference by significantly compressing the Key-Value (KV) cache into a latent vector. Compared to MLA, standard LLMs employing Multi-Head Attention (MHA) and its variants such as Grouped-Query Attention (GQA) exhibit significant cost disadvantages. Enabling well-trained LLMs (e.g., Llama) to rapidly adapt to MLA without pre-training from scratch is both meaningful and challenging. This paper proposes the first data-efficient fine-tuning method for transitioning from MHA to MLA (MHA2MLA), which includes two key components: for partial-RoPE, we remove RoPE from dimensions of queries and keys that contribute less to the attention scores, for low-rank approximation, we introduce joint SVD approximations based on the pre-trained parameters of keys and values. These carefully designed strategies enable MHA2MLA to recover performance using only a small fraction (0.3% to 0.6%) of the data, significantly reducing inference costs while seamlessly integrating with compression techniques such as KV cache quantization. For example, the KV cache size of Llama2-7B is reduced by 92.19%, with only a 0.5% drop in LongBench performance.

cs.CL↗
arXiv2025

From Scores to Preferences: Redefining MOS Benchmarking for Speech Quality Reward Modeling

Assessing the perceptual quality of synthetic speech is crucial for guiding the development and refinement of speech generation models. However, it has traditionally relied on human subjective ratings such as the Mean Opinion Score (MOS), which depend on manual annotations and often suffer from inconsistent rating standards and poor reproducibility. To address these limitations, we introduce MOS-RMBench, a unified benchmark that reformulates diverse MOS datasets into a preference-comparison setting, enabling rigorous evaluation across different datasets. Building on MOS-RMBench, we systematically construct and evaluate three paradigms for reward modeling: scalar reward models, semi-scalar reward models, and generative reward models (GRMs). Our experiments reveal three key findings: (1) scalar models achieve the strongest overall performance, consistently exceeding 74% accuracy; (2) most models perform considerably worse on synthetic speech than on human speech; and (3) all models struggle on pairs with very small MOS differences. To improve performance on these challenging pairs, we propose a MOS-aware GRM that incorporates an MOS-difference-based reward function, enabling the model to adaptively scale rewards according to the difficulty of each sample pair. Experimental results show that the MOS-aware GRM significantly improves fine-grained quality discrimination and narrows the gap with scalar models on the most challenging cases. We hope this work will establish both a benchmark and a methodological framework to foster more rigorous and scalable research in automatic speech quality assessment.

cs.SD↗
arXiv2025

Query-Kontext: An Unified Multimodal Model for Image Generation and Editing

Unified Multimodal Models (UMMs) have demonstrated remarkable performance in text-to-image generation (T2I) and editing (TI2I), whether instantiated as assembled unified frameworks which couple powerful vision-language model (VLM) with diffusion-based generator, or as naive Unified Multimodal Models with an early fusion of understanding and generation modalities. We contend that in current unified frameworks, the crucial capability of multimodal generative reasoning which encompasses instruction understanding, grounding, and image referring for identity preservation and faithful reconstruction, is intrinsically entangled with high-fidelity synthesis. In this work, we introduce Query-Kontext, a novel approach that bridges the VLM and diffusion model via a multimodal ``kontext'' composed of semantic cues and coarse-grained image conditions encoded from multimodal inputs. This design delegates the complex ability of multimodal generative reasoning to powerful VLM while reserving diffusion model's role for high-quality visual synthesis. To achieve this, we propose a three-stage progressive training strategy. First, we connect the VLM to a lightweight diffusion head via multimodal kontext tokens to unleash the VLM's generative reasoning ability. Second, we scale this head to a large, pre-trained diffusion model to enhance visual detail and realism. Finally, we introduce a low-level image encoder to improve image fidelity and perform instruction tuning on downstream tasks. Furthermore, we build a comprehensive data pipeline integrating real, synthetic, and open-source datasets, covering diverse multimodal reference-to-image scenarios, including image generation, instruction-driven editing, customized generation, and multi-subject composition. Experiments show that our approach matches strong unified baselines and even outperforms task-specific state-of-the-art methods in several cases.

cs.CV↗
arXiv2025

TP-MVCC: Tri-plane Multi-view Fusion Model for Silkie Chicken Counting

Accurate animal counting is essential for smart farming but remains difficult in crowded scenes due to occlusions and limited camera views. To address this, we propose a tri-plane-based multi-view chicken counting model (TP-MVCC), which leverages geometric projection and tri-plane fusion to integrate features from multiple cameras onto a unified ground plane. The framework extracts single-view features, aligns them via spatial transformation, and decodes a scene-level density map for precise chicken counting. In addition, we construct the first multi-view dataset of silkie chickens under real farming conditions. Experiments show that TP-MVCC significantly outperforms single-view and conventional fusion comparisons, achieving 95.1\% accuracy and strong robustness in dense, occluded scenarios, demonstrating its practical potential for intelligent agriculture.

cs.CV↗
arXiv2025

Semantic Reformulation Entropy for Robust Hallucination Detection in QA Tasks

Reliable question answering with large language models (LLMs) is challenged by hallucinations, fluent but factually incorrect outputs arising from epistemic uncertainty. Existing entropy-based semantic-level uncertainty estimation methods are limited by sampling noise and unstable clustering of variable-length answers. We propose Semantic Reformulation Entropy (SRE), which improves uncertainty estimation in two ways. First, input-side semantic reformulations produce faithful paraphrases, expand the estimation space, and reduce biases from superficial decoder tendencies. Second, progressive, energy-based hybrid clustering stabilizes semantic grouping. Experiments on SQuAD and TriviaQA show that SRE outperforms strong baselines, providing more robust and generalizable hallucination detection. These results demonstrate that combining input diversification with multi-signal clustering substantially enhances semantic-level uncertainty estimation.

cs.CL↗
arXiv2025

MAIN: Mutual Alignment Is Necessary for instruction tuning

Instruction tuning has empowered large language models (LLMs) to achieve remarkable performance, yet its success heavily depends on the availability of large-scale, high-quality instruction-response pairs. To meet this demand, various methods have been developed to synthesize data at scale. However, current methods for scaling up data generation often overlook a crucial aspect: the alignment between instructions and responses. We hypothesize that the quality of instruction-response pairs is determined not by the individual quality of each component, but by the degree of mutual alignment. To address this, we propose a Mutual Alignment Framework (MAIN) which enforces coherence between instructions and responses through mutual constraints. We demonstrate that MAIN generalizes well across model architectures and sizes, achieving state-of-the-art performance on LLaMA, Mistral, and Qwen models across diverse benchmarks. This work underscores the critical role of instruction-response alignment in enabling generalizable and high-quality instruction tuning for LLMs. All code is available from our repository.

cs.CL↗