arXiv subjects

Li Zhang

Publications and source records attributed to Li Zhang.

At least 343 records · Page 19Linked to original sources

arXiv2025

AdaptiveLLM: A Framework for Selecting Optimal Cost-Efficient LLM for Code-Generation Based on CoT Length

While Large Language Models (LLMs) have significantly advanced code generation efficiency, they face inherent challenges in balancing performance and inference costs across diverse programming tasks. Dynamically selecting the optimal LLM based on task difficulty and resource constraints offers a promising approach to achieve an optimal balance between efficiency and performance. However, existing model selection methods are resource-intensive and often neglect cost efficiency. Moreover, these approaches rely on human-annotated difficulty labels that are frequently inaccessible in real-world settings and may not align with the LLM's own assessment of task difficulty. In this paper, we introduce AdaptiveLLM, a framework that dynamically selects optimal LLMs for a given coding task by automatically assessing task difficulty. Our framework first estimates task difficulty using Chain-of-Thought lengths generated by reasoning model, clusters these into three difficulty levels via k-means, and fine-tunes CodeBERT to embed difficulty-aware features. A trained XGBoost classifier then selects the best model for each problem, optimizing the performance-cost trade-off. Experimental results show that AdaptiveLLM achieves a 7.86% improvement in pass@1 score while reducing resource consumption by 88.9% compared to baseline method ComplexityNet. When compared to a single model, AdaptiveLLM demonstrates an approximately 15% accuracy improvement, while maintaining the same level of cost consumption. Apart from that, the difficulty assessment using CoT provides more reliable selection criteria than human evaluation. Our replication package is available at https://github.com/cjhCoder7/AdaptiveLLM.

cs.SE↗

arXiv2025

Thermodynamic topological classes of the rotating, accelerating black holes

In this paper, we investigate the topological numbers for the rotating, accelerating neutral black hole and its AdS extension, as well as the rotating, accelerating charged black hole and its AdS extension. We find that the topological number of an asymptotically flat accelerating black hole consistently differs by one from that of its non-accelerating counterpart. Furthermore, we show that for an asymptotically AdS accelerating black hole, the topological number is reduced by one compared to its non-accelerating AdS counterpart. In addition, we demonstrate that within the framework of general relativity, the acceleration parameter and the negative cosmological constant each independently add one to the topological number. However, when both factors are present, their effects neutralize each other, resulting in no overall change to the topological number.

hep-th↗

arXiv2025

Measuring Faithfulness and Abstention: An Automated Pipeline for Evaluating LLM-Generated 3-ply Case-Based Legal Arguments

Large Language Models (LLMs) demonstrate potential in complex legal tasks like argument generation, yet their reliability remains a concern. Building upon pilot work assessing LLM generation of 3-ply legal arguments using human evaluation, this paper introduces an automated pipeline to evaluate LLM performance on this task, specifically focusing on faithfulness (absence of hallucination), factor utilization, and appropriate abstention. We define hallucination as the generation of factors not present in the input case materials and abstention as the model's ability to refrain from generating arguments when instructed and no factual basis exists. Our automated method employs an external LLM to extract factors from generated arguments and compares them against the ground-truth factors provided in the input case triples (current case and two precedent cases). We evaluated eight distinct LLMs on three tests of increasing difficulty: 1) generating a standard 3-ply argument, 2) generating an argument with swapped precedent roles, and 3) recognizing the impossibility of argument generation due to lack of shared factors and abstaining. Our findings indicate that while current LLMs achieve high accuracy (over 90%) in avoiding hallucination on viable argument generation tests (Tests 1 & 2), they often fail to utilize the full set of relevant factors present in the cases. Critically, on the abstention test (Test 3), most models failed to follow instructions to stop, instead generating spurious arguments despite the lack of common factors. This automated pipeline provides a scalable method for assessing these crucial LLM behaviors, highlighting the need for improvements in factor utilization and robust abstention capabilities before reliable deployment in legal settings. Link: https://lizhang-aiandlaw.github.io/An-Automated-Pipeline-for-Evaluating-LLM-Generated-3-ply-Case-Based-Legal-Arguments/

cs.CL↗

arXiv2025

On the Limit of Language Models as Planning Formalizers

Large Language Models have been found to create plans that are neither executable nor verifiable in grounded environments. An emerging line of work demonstrates success in using the LLM as a formalizer to generate a formal representation of the planning domain in some language, such as Planning Domain Definition Language (PDDL). This formal representation can be deterministically solved to find a plan. We systematically evaluate this methodology while bridging some major gaps. While previous work only generates a partial PDDL representation, given templated, and therefore unrealistic environment descriptions, we generate the complete representation given descriptions of various naturalness levels. Among an array of observations critical to improve LLMs' formal planning abilities, we note that most large enough models can effectively formalize descriptions as PDDL, outperforming those directly generating plans, while being robust to lexical perturbation. As the descriptions become more natural-sounding, we observe a decrease in performance and provide detailed error analysis.

cs.CL↗

arXiv2025

Research on E-Commerce Long-Tail Product Recommendation Mechanism Based on Large-Scale Language Models

As e-commerce platforms expand their product catalogs, accurately recommending long-tail items becomes increasingly important for enhancing both user experience and platform revenue. A key challenge is the long-tail problem, where extreme data sparsity and cold-start issues limit the performance of traditional recommendation methods. To address this, we propose a novel long-tail product recommendation mechanism that integrates product text descriptions and user behavior sequences using a large-scale language model (LLM). First, we introduce a semantic visor, which leverages a pre-trained LLM to convert multimodal textual content such as product titles, descriptions, and user reviews into meaningful embeddings. These embeddings help represent item-level semantics effectively. We then employ an attention-based user intent encoder that captures users' latent interests, especially toward long-tail items, by modeling collaborative behavior patterns. These components feed into a hybrid ranking model that fuses semantic similarity scores, collaborative filtering outputs, and LLM-generated recommendation candidates. Extensive experiments on a real-world e-commerce dataset show that our method outperforms baseline models in recall (+12%), hit rate (+9%), and user coverage (+15%). These improvements lead to better exposure and purchase rates for long-tail products. Our work highlights the potential of LLMs in interpreting product content and user intent, offering a promising direction for future e-commerce recommendation systems.

cs.IR↗

arXiv2025

DynaCode: A Dynamic Complexity-Aware Code Benchmark for Evaluating Large Language Models in Code Generation

The rapid advancement of large language models (LLMs) has significantly improved their performance in code generation tasks. However, existing code benchmarks remain static, consisting of fixed datasets with predefined problems. This makes them vulnerable to memorization during training, where LLMs recall specific test cases instead of generalizing to new problems, leading to data contamination and unreliable evaluation results. To address these issues, we introduce DynaCode, a dynamic, complexity-aware benchmark that overcomes the limitations of static datasets. DynaCode evaluates LLMs systematically using a complexity-aware metric, incorporating both code complexity and call-graph structures. DynaCode achieves large-scale diversity, generating up to 189 million unique nested code problems across four distinct levels of code complexity, referred to as units, and 16 types of call graphs. Results on 12 latest LLMs show an average performance drop of 16.8% to 45.7% compared to MBPP+, a static code generation benchmark, with performance progressively decreasing as complexity increases. This demonstrates DynaCode's ability to effectively differentiate LLMs. Additionally, by leveraging call graphs, we gain insights into LLM behavior, particularly their preference for handling subfunction interactions within nested code. Our benchmark and evaluation code are available at https://github.com/HWH-2000/DynaCode.

cs.CL↗

arXiv2025

Towards Robust Overlapping Speech Detection: A Speaker-Aware Progressive Approach Using WavLM

Overlapping Speech Detection (OSD) aims to identify regions where multiple speakers overlap in a conversation, a critical challenge in multi-party speech processing. This work proposes a speaker-aware progressive OSD model that leverages a progressive training strategy to enhance the correlation between subtasks such as voice activity detection (VAD) and overlap detection. To improve acoustic representation, we explore the effectiveness of state-of-the-art self-supervised learning (SSL) models, including WavLM and wav2vec 2.0, while incorporating a speaker attention module to enrich features with frame-level speaker information. Experimental results show that the proposed method achieves state-of-the-art performance, with an F1 score of 82.76\% on the AMI test set, demonstrating its robustness and effectiveness in OSD.

cs.SD↗

arXiv2025

R1-T1: Fully Incentivizing Translation Capability in LLMs via Reasoning Learning

Despite recent breakthroughs in reasoning-enhanced large language models (LLMs) like DeepSeek-R1, incorporating inference-time reasoning into machine translation (MT), where human translators naturally employ structured, multi-layered reasoning chain-of-thoughts (CoTs), is yet underexplored. Existing methods either design a fixed CoT tailored for a specific MT sub-task (e.g., literature translation), or rely on synthesizing CoTs unaligned with humans and supervised fine-tuning (SFT) prone to overfitting, limiting their adaptability to diverse translation scenarios. This paper introduces R1-Translator (R1-T1), a novel framework to achieve inference-time reasoning for general MT via reinforcement learning (RL) with human-aligned CoTs comprising six common patterns. Our approach pioneers three innovations: (1) extending reasoning-based translation to broader MT scenarios (e.g., multilingual MT, domain MT) unseen in the training phase; (2) formalizing six expert-curated CoT templates that mirror hybrid human strategies like context-aware paraphrasing and back translation; and (3) enabling self-evolving CoT discovery through RL. Both human and automatic evaluation results indicate a steady translation performance improvement in a total of 10+ languages and 40+ translation directions on Flores-101 test set and four domain-specific MT tasks, especially on the languages unseen from training.

cs.CL↗

arXiv2025

Distilling Textual Priors from LLM to Efficient Image Fusion

Multi-modality image fusion aims to synthesize a single, comprehensive image from multiple source inputs. Traditional approaches, such as CNNs and GANs, offer efficiency but struggle to handle low-quality or complex inputs. Recent advances in text-guided methods leverage large model priors to overcome these limitations, but at the cost of significant computational overhead, both in memory and inference time. To address this challenge, we propose a novel framework for distilling large model priors, eliminating the need for text guidance during inference while dramatically reducing model size. Our framework utilizes a teacher-student architecture, where the teacher network incorporates large model priors and transfers this knowledge to a smaller student network via a tailored distillation process. Additionally, we introduce spatial-channel cross-fusion module to enhance the model's ability to leverage textual priors across both spatial and channel dimensions. Our method achieves a favorable trade-off between computational efficiency and fusion quality. The distilled network, requiring only 10% of the parameters and inference time of the teacher network, retains 90% of its performance and outperforms existing SOTA methods. Extensive experiments demonstrate the effectiveness of our approach. The implementation will be made publicly available as an open-source resource.

cs.CV↗

arXiv2025

Q-Insight: Understanding Image Quality via Visual Reinforcement Learning

Image quality assessment (IQA) focuses on the perceptual visual quality of images, playing a crucial role in downstream tasks such as image reconstruction, compression, and generation. The rapid advancement of multi-modal large language models (MLLMs) has significantly broadened the scope of IQA, moving toward comprehensive image quality understanding that incorporates content analysis, degradation perception, and comparison reasoning beyond mere numerical scoring. Previous MLLM-based methods typically either generate numerical scores lacking interpretability or heavily rely on supervised fine-tuning (SFT) using large-scale annotated datasets to provide descriptive assessments, limiting their flexibility and applicability. In this paper, we propose Q-Insight, a reinforcement learning-based model built upon group relative policy optimization (GRPO), which demonstrates strong visual reasoning capability for image quality understanding while requiring only a limited amount of rating scores and degradation labels. By jointly optimizing score regression and degradation perception tasks with carefully designed reward functions, our approach effectively exploits their mutual benefits for enhanced performance. Extensive experiments demonstrate that Q-Insight substantially outperforms existing state-of-the-art methods in both score regression and degradation perception tasks, while exhibiting impressive zero-shot generalization to comparison reasoning tasks. Code will be available at https://github.com/lwq20020127/Q-Insight.

cs.CV↗

arXiv2025

Semi-Supervised Multi-Label Feature Selection with Consistent Sparse Graph Learning

In practical domains, high-dimensional data are usually associated with diverse semantic labels, whereas traditional feature selection methods are designed for single-label data. Moreover, existing multi-label methods encounter two main challenges in semi-supervised scenarios: (1). Most semi-supervised methods fail to evaluate the label correlations without enough labeled samples, which are the critical information of multi-label feature selection, making label-specific features discarded. (2). The similarity graph structure directly derived from the original feature space is suboptimal for multi-label problems in existing graph-based methods, leading to unreliable soft labels and degraded feature selection performance. To overcome them, we propose a consistent sparse graph learning method for multi-label semi-supervised feature selection (SGMFS), which can enhance the feature selection performance by maintaining space consistency and learning label correlations in semi-supervised scenarios. Specifically, for Challenge (1), SGMFS learns a low-dimensional and independent label subspace from the projected features, which can compatibly cross multiple labels and effectively achieve the label correlations. For Challenge (2), instead of constructing a fixed similarity graph for semi-supervised learning, SGMFS thoroughly explores the intrinsic structure of the data by performing sparse reconstruction of samples in both the label space and the learned subspace simultaneously. In this way, the similarity graph can be adaptively learned to maintain the consistency between label space and the learned subspace, which can promote propagating proper soft labels for unlabeled samples, facilitating the ultimate feature selection. An effective solution with fast convergence is designed to optimize the objective function. Extensive experiments validate the superiority of SGMFS.

cs.LG↗

arXiv2025

Vision to Specification: Automating the Transition from Conceptual Features to Functional Requirements

The translation of high-level abstract features into clear, and testable functional requirements (FRs) is a crucial step in software development, bridging the gap between user needs and technical specifications. In engineering practice, significant expert effort is needed for this translation. Our approach, EasyFR, streamlines the process by recommending Semantic Role Labeling (SRL) sequences for the given abstract features to guide Pre-trained Language Models (PLMs) in producing cohesive FR statements. By analyzing ten diverse datasets, we induce two variable SRL templates, each including two configurable parts. For concrete features, our proposed Key2Temp model can construct the appropriate variant of the SRL template by identifying a variable SRL template and placing the feature tokens in the appropriate slots. In this way, our approach reframes the process of requirement generation into a structured slot-filling activity. Experimental validation on four open datasets demonstrates that EasyFR outperforms three advanced Natural language generation (NLG) approaches, including GPT4, particularly when existing FRs are available for training. The positive influence of our SRL template variant recommendations is further confirmed through an ablation study. We believe that our results indicate a notable step forward in the realm of automated requirements synthesis, holding potential to improve the process of requirements specification in future software projects.

cs.SE↗

arXiv2025

Incorporating Verification Standards for Security Requirements Generation from Functional Specifications

In the current software driven era, ensuring privacy and security is critical. Despite this, the specification of security requirements for software is still largely a manual and labor intensive process. Engineers are tasked with analyzing potential security threats based on functional requirements (FRs), a procedure prone to omissions and errors due to the expertise gap between cybersecurity experts and software engineers. To bridge this gap, we introduce F2SRD (Function to Security Requirements Derivation), an automated approach that proactively derives security requirements (SRs) from functional specifications under the guidance of relevant security verification requirements (VRs) drawn from the well recognized OWASP Application Security Verification Standard (ASVS). F2SRD operates in two main phases: Initially, we develop a VR retriever trained on a custom database of FR and VR pairs, enabling it to adeptly select applicable VRs from ASVS. This targeted retrieval informs the precise and actionable formulation of SRs. Subsequently, these VRs are used to construct structured prompts that direct GPT4 in generating SRs. Our comparative analysis against two established models demonstrates F2SRD's enhanced performance in producing SRs that excel in inspiration, diversity, and specificity essential attributes for effective security requirement generation. By leveraging security verification standards, we believe that the generated SRs are not only more focused but also resonate stronger with the needs of engineers.

cs.SE↗

arXiv2025

Patient-Specific Dynamic Digital-Physical Twin for Coronary Intervention Training: An Integrated Mixed Reality Approach

Background and Objective: Precise preoperative planning and effective physician training for coronary interventions are increasingly important. Despite advances in medical imaging technologies, transforming static or limited dynamic imaging data into comprehensive dynamic cardiac models remains challenging. Existing training systems lack accurate simulation of cardiac physiological dynamics. This study develops a comprehensive dynamic cardiac model research framework based on 4D-CTA, integrating digital twin technology, computer vision, and physical model manufacturing to provide precise, personalized tools for interventional cardiology. Methods: Using 4D-CTA data from a 60-year-old female with three-vessel coronary stenosis, we segmented cardiac chambers and coronary arteries, constructed dynamic models, and implemented skeletal skinning weight computation to simulate vessel deformation across 20 cardiac phases. Transparent vascular physical models were manufactured using medical-grade silicone. We developed cardiac output analysis and virtual angiography systems, implemented guidewire 3D reconstruction using binocular stereo vision, and evaluated the system through angiography validation and CABG training applications. Results: Morphological consistency between virtual and real angiography reached 80.9%. Dice similarity coefficients for guidewire motion ranged from 0.741-0.812, with mean trajectory errors below 1.1 mm. The transparent model demonstrated advantages in CABG training, allowing direct visualization while simulating beating heart challenges. Conclusion: Our patient-specific digital-physical twin approach effectively reproduces both anatomical structures and dynamic characteristics of coronary vasculature, offering a dynamic environment with visual and tactile feedback valuable for education and clinical planning.

cs.CV↗

arXiv2025

Bidirectional Distillation: A Mixed-Play Framework for Multi-Agent Generalizable Behaviors

Population-population generalization is a challenging problem in multi-agent reinforcement learning (MARL), particularly when agents encounter unseen co-players. However, existing self-play-based methods are constrained by the limitation of inside-space generalization. In this study, we propose Bidirectional Distillation (BiDist), a novel mixed-play framework, to overcome this limitation in MARL. BiDist leverages knowledge distillation in two alternating directions: forward distillation, which emulates the historical policies' space and creates an implicit self-play, and reverse distillation, which systematically drives agents towards novel distributions outside the known policy space in a non-self-play manner. In addition, BiDist operates as a concise and efficient solution without the need for the complex and costly storage of past policies. We provide both theoretical analysis and empirical evidence to support BiDist's effectiveness. Our results highlight its remarkable generalization ability across a variety of cooperative, competitive, and social dilemma tasks, and reveal that BiDist significantly diversifies the policy distribution space. We also present comprehensive ablation studies to reinforce BiDist's effectiveness and key success factors. Source codes are available in the supplementary material.

cs.LG↗

arXiv2025

Revolutionizing Newcomers' Onboarding Process in OSS Communities: The Future AI Mentor

Onboarding newcomers is vital for the sustainability of open-source software (OSS) projects. To lower barriers and increase engagement, OSS projects have dedicated experts who provide guidance for newcomers. However, timely responses are often hindered by experts' busy schedules. The recent rapid advancements of AI in software engineering have brought opportunities to leverage AI as a substitute for expert mentoring. However, the potential role of AI as a comprehensive mentor throughout the entire onboarding process remains unexplored. To identify design strategies of this ``AI mentor'', we applied Design Fiction as a participatory method with 19 OSS newcomers. We investigated their current onboarding experience and elicited 32 design strategies for future AI mentor. Participants envisioned AI mentor being integrated into OSS platforms like GitHub, where it could offer assistance to newcomers, such as ``recommending projects based on personalized requirements'' and ``assessing and categorizing project issues by difficulty''. We also collected participants' perceptions of a prototype, named ``OSSerCopilot'', that implemented the envisioned strategies. They found the interface useful and user-friendly, showing a willingness to use it in the future, which suggests the design strategies are effective. Finally, in order to identify the gaps between our design strategies and current research, we conducted a comprehensive literature review, evaluating the extent of existing research support for this concept. We find that research is relatively scarce in certain areas where newcomers highly anticipate AI mentor assistance, such as ``discovering an interested project''. Our study has the potential to revolutionize the current newcomer-expert mentorship and provides valuable insights for researchers and tool designers aiming to develop and enhance AI mentor systems.

cs.SE↗

arXiv2025

Nonperiodic dynamic CT reconstruction using backward-warping INR with regularization of diffeomorphism (BIRD)

Dynamic computed tomography (CT) reconstruction faces significant challenges in addressing motion artifacts, particularly for nonperiodic rapid movements such as cardiac imaging with fast heart rates. Traditional methods struggle with the extreme limited-angle problems inherent in nonperiodic cases. Deep learning methods have improved performance but face generalization challenges. Recent implicit neural representation (INR) techniques show promise through self-supervised deep learning, but have critical limitations: computational inefficiency due to forward-warping modeling, difficulty balancing DVF complexity with anatomical plausibility, and challenges in preserving fine details without additional patient-specific pre-scans. This paper presents a novel INR-based framework, BIRD, for nonperiodic dynamic CT reconstruction. It addresses these challenges through four key contributions: (1) backward-warping deformation that enables direct computation of each dynamic voxel with significantly reduced computational cost, (2) diffeomorphism-based DVF regularization that ensures anatomically plausible deformations while maintaining representational capacity, (3) motion-compensated analytical reconstruction that enhances fine details without requiring additional pre-scans, and (4) dimensional-reduction design for efficient 4D coordinate encoding. Through various simulations and practical studies, including digital and physical phantoms and retrospective patient data, we demonstrate the effectiveness of our approach for nonperiodic dynamic CT reconstruction with enhanced details and reduced motion artifacts. The proposed framework enables more accurate dynamic CT reconstruction with potential clinical applications, such as one-beat cardiac reconstruction, cinematic image sequences for functional imaging, and motion artifact reduction in conventional CT scans.

cs.CV↗

arXiv2025

Transferable Adversarial Attacks on Black-Box Vision-Language Models

Vision Large Language Models (VLLMs) are increasingly deployed to offer advanced capabilities on inputs comprising both text and images. While prior research has shown that adversarial attacks can transfer from open-source to proprietary black-box models in text-only and vision-only contexts, the extent and effectiveness of such vulnerabilities remain underexplored for VLLMs. We present a comprehensive analysis demonstrating that targeted adversarial examples are highly transferable to widely-used proprietary VLLMs such as GPT-4o, Claude, and Gemini. We show that attackers can craft perturbations to induce specific attacker-chosen interpretations of visual information, such as misinterpreting hazardous content as safe, overlooking sensitive or restricted material, or generating detailed incorrect responses aligned with the attacker's intent. Furthermore, we discover that universal perturbations -- modifications applicable to a wide set of images -- can consistently induce these misinterpretations across multiple proprietary VLLMs. Our experimental results on object recognition, visual question answering, and image captioning show that this vulnerability is common across current state-of-the-art models, and underscore an urgent need for robust mitigations to ensure the safe and secure deployment of VLLMs.

cs.CV↗