arXiv subjects

Yang Zhao

Publications and source records attributed to Yang Zhao.

At least 109 records · Page 6Linked to original sources

arXiv2025

How to Auto-optimize Prompts for Domain Tasks? Adaptive Prompting and Reasoning through Evolutionary Domain Knowledge Adaptation

Designing optimal prompts and reasoning processes for large language models (LLMs) on domain-specific tasks is both necessary and challenging in real-world applications. Determining how to integrate domain knowledge, enhance reasoning efficiency, and even provide domain experts with refined knowledge integration hints are particularly crucial yet unresolved tasks. In this research, we propose Evolutionary Graph Optimization for Prompting (EGO-Prompt), an automated framework to designing better prompts, efficient reasoning processes and providing enhanced causal-informed process. EGO-Prompt begins with a general prompt and fault-tolerant initial Semantic Causal Graph (SCG) descriptions, constructed by human experts, which is then automatically refined and optimized to guide LLM reasoning. Recognizing that expert-defined SCGs may be partial or imperfect and that their optimal integration varies across LLMs, EGO-Prompt integrates a novel causal-guided textual gradient process in two steps: first, generating nearly deterministic reasoning guidance from the SCG for each instance, and second, adapting the LLM to effectively utilize the guidance alongside the original input. The iterative optimization algorithm further refines both the SCG and the reasoning mechanism using textual gradients with ground-truth. We tested the framework on real-world public health, transportation and human behavior tasks. EGO-Prompt achieves 7.32%-12.61% higher F1 than cutting-edge methods, and allows small models to reach the performence of larger models at under 20% of the original cost. It also outputs a refined, domain-specific SCG that improves interpretability.

cs.AI↗

arXiv2025

Towards Physics-informed Spatial Intelligence with Human Priors: An Autonomous Driving Pilot Study

How to integrate and verify spatial intelligence in foundation models remains an open challenge. Current practice often proxies Visual-Spatial Intelligence (VSI) with purely textual prompts and VQA-style scoring, which obscures geometry, invites linguistic shortcuts, and weakens attribution to genuinely spatial skills. We introduce Spatial Intelligence Grid (SIG): a structured, grid-based schema that explicitly encodes object layouts, inter-object relations, and physically grounded priors. As a complementary channel to text, SIG provides a faithful, compositional representation of scene structure for foundation-model reasoning. Building on SIG, we derive SIG-informed evaluation metrics that quantify a model's intrinsic VSI, which separates spatial capability from language priors. In few-shot in-context learning with state-of-the-art multimodal LLMs (e.g. GPT- and Gemini-family models), SIG yields consistently larger, more stable, and more comprehensive gains across all VSI metrics compared to VQA-only representations, indicating its promise as a data-labeling and training schema for learning VSI. We also release SIGBench, a benchmark of 1.4K driving frames annotated with ground-truth SIG labels and human gaze traces, supporting both grid-based machine VSI tasks and attention-driven, human-like VSI tasks in autonomous-driving scenarios.

cs.CV↗

arXiv2025

Permutative Preference Alignment from Listwise Ranking of Human Judgments

Aligning Large Language Models (LLMs) with human preferences is crucial in ensuring desirable and controllable model behaviors. Current methods, such as Reinforcement Learning from Human Feedback (RLHF) and Direct Preference Optimization (DPO), rely on the Bradley-Terry (B-T) model to maximize the likelihood of pairwise choices. However, when multiple responses are available, the B-T model fails to guarantee an accurate list ranking of the responses. To address this issue, we propose Permutative Preference Alignment (PPA), a novel offline listwise approach that incorporates the Normalized Discounted Cumulative Gain (NDCG), a widely-used ranking metric, as an alternative training objective for LLM alignment. We develop an end-to-end alignment algorithm by approximating NDCG with a differentiable surrogate loss. Experiments demonstrate that PPA outperforms existing pairwise and listwise methods on evaluation sets and general benchmarks such as AlpacaEval. Furthermore, we show that NDCG-based approaches improve ranking accuracy more effectively than B-T-based methods and provide a theoretical explanation for this improvement.

cs.CL↗

arXiv2025

OneSearch: A Preliminary Exploration of the Unified End-to-End Generative Framework for E-commerce Search

Traditional e-commerce search systems employ multi-stage cascading architectures (MCA) that progressively filter items through recall, pre-ranking, and ranking stages. While effective at balancing computational efficiency with business conversion, these systems suffer from fragmented computation and optimization objective collisions across stages, which ultimately limit their performance ceiling. To address these, we propose \textbf{OneSearch}, the first industrial-deployed end-to-end generative framework for e-commerce search. This framework introduces three key innovations: (1) a Keyword-enhanced Hierarchical Quantization Encoding (KHQE) module, to preserve both hierarchical semantics and distinctive item attributes while maintaining strong query-item relevance constraints; (2) a multi-view user behavior sequence injection strategy that constructs behavior-driven user IDs and incorporates both explicit short-term and implicit long-term sequences to model user preferences comprehensively; and (3) a Preference-Aware Reward System (PARS) featuring multi-stage supervised fine-tuning and adaptive reward-weighted ranking to capture fine-grained user preferences. Extensive offline evaluations on large-scale industry datasets demonstrate OneSearch's superior performance for high-quality recall and ranking. The rigorous online A/B tests confirm its ability to enhance relevance in the same exposure position, achieving statistically significant improvements: +1.67% item CTR, +2.40% buyer, and +3.22% order volume. Furthermore, OneSearch reduces operational expenditure by 75.40% and improves Model FLOPs Utilization from 3.26% to 27.32%. The system has been successfully deployed across multiple search scenarios in Kuaishou, serving millions of users, generating tens of millions of PVs daily.

cs.IR↗

arXiv2025

Long-range bipartite entanglement in XXZ spin chains with the exponential and power-law long-range interactions

Long-range bipartite entanglement (LBE) and its distribution properties are studied in XXZ spin chains with the exponential and power-law long-range interactions (ELRIs and PLRIs). LBE quantified by two-qubit concurrence decays exponentially along with two-site distance in the infinite chain with ELRIs in the thermodynamic limit, and the long-range behavior of two-spin entanglement can detect the quantum phase transition and identify different quantum phases away from the critical point. Moreover, a fine-grained LBE distribution relation is obtained for the infinite XXZ spin chain. On the other hand, in the finite XXZ spin chain with the conventional PLRIs, the long-range concurrence decays algebraically and the total one is no longer monotonic along with the chain length. The total LBE distribution property can exhibit a piecewise function, which has a close relationship with the decaying mode and strength of PLRIs. These LBE relations can be regarded as the generalization of Koashi-Bužek-Imoto bound for the prototypical long-range XXZ model, having potential applications in quantum information processing.

quant-ph↗

arXiv2025

Quantum phase transition of sub-Ohmic spin-boson models: An approach by the multiple Davydov D2 Ansatz

The ground state properties and quantum phase transitions of sub-Ohmic spin-boson models are investigated using the multiple Davydov D2 Ansatz in conjunction with the variational principle. Three variants of the model are studied: (i) a single bath with diagonal coupling, (ii) two independent baths with diagonal and off-diagonal couplings, and (iii) a single bath with simultaneous diagonal and off-diagonal couplings. For the purely diagonal model, the multiple Davydov D2 Ansatz yields critical coupling strengths that are consistent with other methodologies, validating its accuracy and efficiency. In the two-bath model, the competition between diagonal and off-diagonal couplings drives a first-order transition for both symmetric and asymmetric spectral exponents, with von-Neumann entropy showing a continuous peak only under exact symmetry. Finally, for a single bath with simultaneous diagonal and off-diagonal couplings, we demonstrate that a rotational transformation maps the system to an equivalent purely diagonal model, enabling simpler and intuitive physical interpretation and reduced computational complexity.

quant-ph↗

arXiv2025

Survey of Large Language Models in Extended Reality: Technical Paradigms and Application Frontiers

Large Language Models (LLMs) have demonstrated remarkable capabilities in natural language understanding and generation, and their integration with Extended Reality (XR) is poised to transform how users interact with immersive environments. This survey provides a comprehensive review of recent developments at the intersection of LLMs and XR, offering a structured organization of research along both technical and application dimensions. We propose a taxonomy of LLM-enhanced XR systems centered on key technical paradigms -- such as interactive agent control, XR development toolkits, and generative scene synthesis -- and discuss how these paradigms enable novel capabilities in XR. In parallel, we examine how LLM-driven techniques support practical XR applications across diverse domains, including immersive education, clinical healthcare, and industrial manufacturing. By connecting these technical paradigms with application frontiers, our survey highlights current trends, delineates design considerations, and identifies open challenges in building LLM-augmented XR systems. This work provides insights that can guide researchers and practitioners in advancing the state of the art in intelligent XR experiences.

cs.HC↗

arXiv2025

VaseVQA-3D: Benchmarking 3D VLMs on Ancient Greek Pottery

Vision-Language Models (VLMs) have achieved significant progress in multimodal understanding tasks, demonstrating strong capabilities particularly in general tasks such as image captioning and visual reasoning. However, when dealing with specialized cultural heritage domains like 3D vase artifacts, existing models face severe data scarcity issues and insufficient domain knowledge limitations. Due to the lack of targeted training data, current VLMs struggle to effectively handle such culturally significant specialized tasks. To address these challenges, we propose the VaseVQA-3D dataset, which serves as the first 3D visual question answering dataset for ancient Greek pottery analysis, collecting 664 ancient Greek vase 3D models with corresponding question-answer data and establishing a complete data construction pipeline. We further develop the VaseVLM model, enhancing model performance in vase artifact analysis through domain-adaptive training. Experimental results validate the effectiveness of our approach, where we improve by 12.8% on R@1 metrics and by 6.6% on lexical similarity compared with previous state-of-the-art on the VaseVQA-3D dataset, significantly improving the recognition and understanding of 3D vase artifacts, providing new technical pathways for digital heritage preservation research. Code: https://github.com/AIGeeksGroup/VaseVQA-3D. Website: https://aigeeksgroup.github.io/VaseVQA-3D.

cs.CV↗

arXiv2025

A Davydov Ansatz approach to accurate system-bath dynamics in the presence of multiple baths with distinct temperatures

We perform benchmark simulations using the time-dependent variational approach with the multiple Davydov Ansatz (mDA) to study realtime nonequilibrium dynamics in a single qubit model coupled to two thermal baths with distinct temperatures. A broad region of the parameter space has been investigated, accompanied by a detailed analysis of the convergence behavior of the mDA method. In addition, we have compared our mD2 results to those from two widely adopted, numerically "exact" techniques: the methods of hierarchical equations of motion (HEOM) and the quasi-adiabatic path integral (QUAPI). It is found that the mDA approach in combination with thermal field dynamics yields numerically accurate, convergent results in nearly all regions of the parameter space examined, including those that pose serious challenges for QUAPI and HEOM. Our results reveal that mDA offers a highly adaptable framework capable of capturing long-time dynamics, even in challenging regimes where other methods face limitations. These findings underscore the potential of mDA as a versatile tool for exploring quantum thermodynamics, energy transfer processes, and non-equilibrium quantum systems.

quant-ph↗

arXiv2025

Autoregressive Adversarial Post-Training for Real-Time Interactive Video Generation

Existing large-scale video generation models are computationally intensive, preventing adoption in real-time and interactive applications. In this work, we propose autoregressive adversarial post-training (AAPT) to transform a pre-trained latent video diffusion model into a real-time, interactive video generator. Our model autoregressively generates a latent frame at a time using a single neural function evaluation (1NFE). The model can stream the result to the user in real time and receive interactive responses as controls to generate the next latent frame. Unlike existing approaches, our method explores adversarial training as an effective paradigm for autoregressive generation. This not only allows us to design an architecture that is more efficient for one-step generation while fully utilizing the KV cache, but also enables training the model in a student-forcing manner that proves to be effective in reducing error accumulation during long video generation. Our experiments demonstrate that our 8B model achieves real-time, 24fps, streaming video generation at 736x416 resolution on a single H100, or 1280x720 on 8xH100 up to a minute long (1440 frames). Visit our research website at https://seaweed-apt.com/2

cs.CV↗

arXiv2025

ImageDoctor: Diagnosing Text-to-Image Generation via Grounded Image Reasoning

The rapid advancement of text-to-image (T2I) models has increased the need for reliable human preference modeling, a demand further amplified by recent progress in reinforcement learning for preference alignment. However, existing approaches typically quantify the quality of a generated image using a single scalar, limiting their ability to provide comprehensive and interpretable feedback on image quality. To address this, we introduce ImageDoctor, a unified multi-aspect T2I model evaluation framework that assesses image quality across four complementary dimensions: plausibility, semantic alignment, aesthetics, and overall quality. ImageDoctor also provides pixel-level flaw indicators in the form of heatmaps, which highlight misaligned or implausible regions, and can be used as a dense reward for T2I model preference alignment. Inspired by the diagnostic process, we improve the detail sensitivity and reasoning capability of ImageDoctor by introducing a "look-think-predict" paradigm, where the model first localizes potential flaws, then generates reasoning, and finally concludes the evaluation with quantitative scores. Built on top of a vision-language model and trained through a combination of supervised fine-tuning and reinforcement learning, ImageDoctor demonstrates strong alignment with human preference across multiple datasets, establishing its effectiveness as an evaluation metric. Furthermore, when used as a reward model for preference tuning, ImageDoctor significantly improves generation quality -- achieving an improvement of 10% over scalar-based reward models.

cs.CV↗

arXiv2025

LLM-Enhanced Self-Evolving Reinforcement Learning for Multi-Step E-Commerce Payment Fraud Risk Detection

This paper presents a novel approach to e-commerce payment fraud detection by integrating reinforcement learning (RL) with Large Language Models (LLMs). By framing transaction risk as a multi-step Markov Decision Process (MDP), RL optimizes risk detection across multiple payment stages. Crafting effective reward functions, essential for RL model success, typically requires significant human expertise due to the complexity and variability in design. LLMs, with their advanced reasoning and coding capabilities, are well-suited to refine these functions, offering improvements over traditional methods. Our approach leverages LLMs to iteratively enhance reward functions, achieving better fraud detection accuracy and demonstrating zero-shot capability. Experiments with real-world data confirm the effectiveness, robustness, and resilience of our LLM-enhanced RL framework through long-term evaluations, underscoring the potential of LLMs in advancing industrial RL applications.

cs.LG↗

arXiv2025

Uniform 2D Target Generation via Inverse-designed Metasurfaces

We propose an inverse design framework for metasurfaces that achieves highly uniform two-dimensional intensity profiles across an on-demand shape. The optimization objective is formulated to enhance overall projection efficiency via the adjoint method, while a regularization term penalizes local deviations in field amplitude to suppress intensity non-uniformity. The regularization weight is adaptively tuned based on the current non-uniformity, enabling stable and efficient optimization. Compared with the widely used mean squared error (MSE) objective, our method yields superior performance in both intensity fidelity and uniformity. We also extend our framework to handle realistic Gaussian beam illumination by biasing the library. Simulation results confirm the effectiveness of our approach for generating high-quality, uniform field patterns.

physics.optics↗

arXiv2025

MANZANO: A Simple and Scalable Unified Multimodal Model with a Hybrid Vision Tokenizer

Unified multimodal Large Language Models (LLMs) that can both understand and generate visual content hold immense potential. However, existing open-source models often suffer from a performance trade-off between these capabilities. We present Manzano, a simple and scalable unified framework that substantially reduces this tension by coupling a hybrid image tokenizer with a well-curated training recipe. A single shared vision encoder feeds two lightweight adapters that produce continuous embeddings for image-to-text understanding and discrete tokens for text-to-image generation within a common semantic space. A unified autoregressive LLM predicts high-level semantics in the form of text and image tokens, with an auxiliary diffusion decoder subsequently translating the image tokens into pixels. The architecture, together with a unified training recipe over understanding and generation data, enables scalable joint learning of both capabilities. Manzano achieves state-of-the-art results among unified models, and is competitive with specialist models, particularly on text-rich evaluation. Our studies show minimal task conflicts and consistent gains from scaling model size, validating our design choice of a hybrid tokenizer.

cs.CV↗

arXiv2025

MIMO Channel Shaping and Rate Maximization Using Beyond-Diagonal RIS

This paper investigates the limits to which a passive Reconfigurable Intelligent Surface (RIS) can reshape a point-to-point Multiple-Input Multiple-Output (MIMO) channel in terms of singular values and their functions (e.g., achievable rate and harvestable power) for improved wireless performance. We depart from the Diagonal (D) scattering model and adopt a Beyond-Diagonal (BD) model that exploits element-wise connections for passive signal amplitude and phase manipulation. Specifically, analytical tight bounds are derived under typical RIS deployment scenarios to unveil the channel shaping potentials of BD-RIS regarding communication Degrees of Freedom (DoF), singular value spread, power gain, and capacity. An efficient numerical method is then proposed to optimize BD-RIS for any locally Lipschitz function of channel singular values, and showcased to characterize the achievable singular value region. As a side product, we tackle BD-RIS-aided MIMO rate maximization problem by a local-optimal Alternating Optimization (AO) approach and a low-complexity shaping approach. Results show that BD-RIS significantly improves the dynamic range of channel singular values and the tradeoff in manipulating them, thus offering enhanced data rate, harvestable power, and physical-layer security. These advantages become more pronounced when the number of RIS elements, group size, or MIMO dimensions increase. Of particular interest, BD-RIS is shown to activate multi-stream transmission and achieve the asymptotic DoF at much lower transmit power than D-RIS thanks to its proficiency in channel shaping.

eess.SP↗

arXiv2025

Upper bound of holographic entanglement entropy combinations

In this work, we develop a systematic formalism to evaluate the upper bound of a large family of holographic entanglement entropy combinations when fixing $n$ subsystems and fine-tuning one other subsystem. The upper bound configurations and values of these entropy combinations can be derived and classified. The upper bound of these entropy combinations reveals holographic $n+1$-partite entanglement that $n$ fixed subsystems participate in. In AdS$_3$/CFT$_2$, AdS$_4$/CFT$_3$, and even higher-dimensional holography, one can, in principle, find different formulas of upper bound values, reflecting the fundamental difference in entanglement structure in different dimensions.

hep-th↗

arXiv2025

Bias Analysis and Mitigation through Protected Attribute Detection and Regard Classification

Large language models (LLMs) acquire general linguistic knowledge from massive-scale pretraining. However, pretraining data mainly comprised of web-crawled texts contain undesirable social biases which can be perpetuated or even amplified by LLMs. In this study, we propose an efficient yet effective annotation pipeline to investigate social biases in the pretraining corpora. Our pipeline consists of protected attribute detection to identify diverse demographics, followed by regard classification to analyze the language polarity towards each attribute. Through our experiments, we demonstrate the effect of our bias analysis and mitigation measures, focusing on Common Crawl as the most representative pretraining corpus.

cs.CL↗

arXiv2025

Robust Optimization for Movable Antenna-aided Cell-Free ISAC with Time Synchronization Errors

The cell-free integrated sensing and communication (CF-ISAC) system, which effectively mitigates intra-cell interference and provides precise sensing accuracy, is a promising technology for future 6G networks. However, to fully capitalize on the potential of CF-ISAC, accurate time synchronization (TS) between access points (APs) is critical. Due to the limitations of current synchronization technologies, TS errors have become a significant challenge in the development of the CF-ISAC system. In this paper, we propose a novel CF-ISAC architecture based on movable antennas (MAs), which exploits spatial diversity to enhance communication rates, maintain sensing accuracy, and reduce the impact of TS errors. We formulate a worst-case sensing accuracy optimization problem for TS errors to address this challenge, deriving the worst-case Cramér-Rao lower bound (CRLB). Subsequently, we develop a joint optimization framework for AP beamforming and MA positions to satisfy communication rate constraints while improving sensing accuracy. A robust optimization framework is designed for the highly complex and non-convex problem. Specifically, we employ manifold optimization (MO) to solve the worst-case sensing accuracy optimization problem. Then, we propose an MA-enabled meta-reinforcement learning (MA-MetaRL) to design optimization variables while satisfying constraints on MA positions, communication rate, and transmit power, thereby improving sensing accuracy. The simulation results demonstrate that the proposed robust optimization algorithm significantly improves the accuracy of the detection and is strong against TS errors. Moreover, compared to conventional fixed position antenna (FPA) technologies, the proposed MA-aided CF-ISAC architecture achieves higher system capacity, thus validating its effectiveness.

eess.SP↗