数据来源:ArXiv Domain

LLM Domain Papers

1. Enhancing Extubation Failure Prediction with LLM-Derived Features from Respiratory Therapy Clinical Notes

Abstract:Invasive mechanical ventilation is a lifesaving therapy, but timely, safe discontinuation is essential to preventing extubation failure (EF) and related risks to health. We present a novel approach to EF prediction that leverages features classified in free-text respiratory therapy notes using a large language model and logistic regression pipeline. Applied to a patient cohort from University of Washington Medicine, our method identifies clinically meaningful EF-related features that improve EF prediction performance when included alongside structured patient data. We further highlight how differences in target populations in prior EF prediction studies, such as heterogenous inclusion criteria and EF definition, can lead to systematic differences in model performance and hinder generalizability between studies.

中文摘要

摘要:侵入性机械通气是一种挽救生命的治疗方法,但及时、安全的撤机对于防止拔管失败(EF)及相关健康风险至关重要。我们提出了一种利用大型语言模型和逻辑回归流程,对自由文本呼吸治疗记录中分类特征进行分析的新型EF预测方法。应用于华盛顿大学医学中心的患者队列,我们的方法能够识别具有临床意义的EF相关特征,当这些特征与结构化患者数据一起使用时,可提高EF预测性能。我们还强调,先前EF预测研究中目标人群的差异,如入选标准和EF定义的异质性,可能导致模型性能的系统性差异,并阻碍研究之间的可推广性。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17532 (timeout of 15000ms exceeded)

Authors: Izzy Chaiken, Aditya Khowal, Neha A. Sathe, Mark M. Wurfel, Lucy Lu Wang

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2609.17532.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17532

Published: 2026-09-18T01:27:58.489Z


2. Faking Good and Faking Bad in LLMs: Response Distortion Across Dark Triad Personality Traits

Abstract:Social desirability and impression management are pervasive sources of response distortion in human personality assessment, yet their effects on Large Language Models (LLMs) remain underexplored. This study investigates whether contemporary LLMs systematically modulate the expression of Dark Triad traits (Machiavellianism, narcissism, and psychopathy) under fake-good and fake-bad conditions. Seven state-of-the-art models were evaluated across two ecologically relevant contexts: employment selection and forensic evaluation, in which socially desirable or undesirable incentives were conveyed through contextual framing. Trait expression was measured using standard psychometric scoring procedures and compared with self-assessment baselines at both aggregate and item levels. Results revealed systematic and condition-consistent response modulation. Most models reduced Dark Triad scores under fake-good conditions and increased them under fake-bad conditions, although the magnitude and consistency of these effects varied across traits and models. Machiavellianism and narcissism showed the strongest and most coherent shifts, whereas psychopathy displayed greater heterogeneity. Context also influenced responses, with employment scenarios generally producing larger effects than forensic scenarios. An additional experiment showed that explicit fake-bad instructions generated substantially stronger distortions than contextual framing alone. The results suggest that personality-related outputs should be interpreted in light of the motivational and situational context in which they are elicited. More broadly, they highlight the value of psychometric paradigms for evaluating susceptibility to response distortion, impression management, and context-dependent behavioral shifts, with important implications for LLM benchmarking, alignment evaluation, and robustness assessment.

中文摘要

摘要:社会期望性和印象管理是人类人格评估中普遍存在的响应失真来源,但其对大型语言模型(LLMs)的影响仍未得到充分研究。本研究探讨了当面对虚假良好(fake-good)和虚假不良(fake-bad)情境时,当代LLMs是否会系统性地调节黑暗三联性特质(权谋主义、 自恋和精神病倾向)的表达。研究评估了七种最先进的模型在两个生态相关的情境下的表现:就业筛选和法医评估,其中社会期望性或不期望性激励通过情境框架进行传达。特质表达通过标准心理测量评分程序进行测量,并在整体和条目层面与自我评估基线进行比较。结果显示了系统性且与情境一致的响应调节。大多数模型在虚假良好情境下降低黑暗三联性评分,而在虚假不良情境下则提高评分,尽管这种效应的幅度和一致性在不同特质和模型之间存在差异。权谋主义和自恋表现出最强且最连贯的变化,而精神病倾向则显示出更大的异质性。情境也影响了响应,就业场景通常产生的效应大于法医场景。另一项实验显示,明确的虚假不良指令比仅通过情境框架产生的失真明显更强。研究结果表明,人格相关输出的解释应考虑其被引发时的动机和情境背景。更广泛地讲,这些结果强调了心理测量范式在评估对响应失真、印象管理和情境依赖行为变化的敏感性方面的价值,对LLM的基准测试、对齐评估和稳健性评估具有重要意义。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17534 (HTTP 429)

Authors: Victoria Popa, Guglielmo Cola, Caterina Senette, Maurizio Tesconi

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2609.17534.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17534

Published: 2026-09-18T01:27:58.489Z


3. DANTINOX: A Unified Framework for Multi-Paradigm Language Modeling

Abstract:Language generation research increasingly spans three paradigms: autoregressive decoding, discrete masked diffusion, and continuous flow-matching. Comparing them is difficult because each lives in a separate codebase, so measured differences often reflect implementation details rather than the paradigms themselves. We present DantinoX, an open-source JAX/Flax library in which a single modular Transformer backbone serves all three paradigms. Switching the generation paradigm, attention mechanism, or hardware topology requires only a configuration change, while the backbone architecture, tokenizer, initialization strategy, and training infrastructure remain consistent. This enables controlled cross-paradigm comparisons within one API for training, streaming inference, and benchmarking.

中文摘要

摘要:语言生成研究正越来越多地跨越三种范式:自回归解码、离散掩码扩散和连续流匹配。比较它们很困难,因为每种范式都存在于独立的代码库中,因此测量到的差异往往反映的是实现细节而不是范式本身。我们提出了 DantinoX,这是一个开源的 JAX/Flax 库,其中单个模块化 Transformer 主干适用于所有三种范式。切换生成范式、注意力机制或硬件拓扑仅需更改配置,而主干架构、分词器、初始化策略和训练基础设施保持一致。这使得在同一 API 下进行训练、流式推理和基准测试时能够控制跨范式的比较。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17535 (HTTP 429)

Authors: Marco Simoni, Aleksandar Fontana, Giulio Rossolini, Andrea Saracino

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2609.17535.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17535

Published: 2026-09-18T01:27:58.489Z


4. Think Before You Comfort: Reflective Cognitive Alignment for Protocol-Grounded Elderly Stimulation Agents

Abstract:Cognitive Stimulation Therapy (CST) offers non-pharmacological support for elders with cognitive impairment, yet scalability remains constrained by reliance on trained facilitators and severe data scarcity, particularly for privacy-sensitive, low-resource languages such as Cantonese. While Large Language Models (LLMs) show promise for automated companionship, they often struggle to balance empathetic engagement with adherence to cognitive stimulation guidelines. We propose a framework addressing these challenges along two complementary axes. First, STaR-CS (Style-Transfer and Role-Conditioned Cognitive Stimulation) synthesizes multi-party dialogues through facilitator style modeling and structured skeleton extraction, mitigating data barriers. Building upon this corpus, the Reflective Cognitive Alignment (RCA) framework models stimulation interactions as a sequential decision process, integrating Protocol-Constrained Chain-of-Cognition (PC-CoC) for structured reasoning and Inference-Time Value Alignment (IVA) for principled response selection based on safety and engagement goals. Evaluations across six backbone LLMs and two independent judges show that RCA consistently improves protocol adherence, safety, and group facilitation over standard prompting baselines. Our code is available at this https URL.

中文摘要

摘要:认知刺激疗法(CST)为认知受损的老年人提供非药物支持,但其可扩展性仍受限于对训练有素的促动者的依赖以及严重的数据匮乏,尤其是对隐私敏感、资源匮乏的语言,如粤语。虽然大型语言模型(LLMs)在自动化陪伴方面显示出潜力,但它们通常在保持同理心互动与遵循认知刺激指南之间难以取得平衡。我们提出了一个沿两条互补轴线应对这些挑战的框架。首先,STaR-CS(基于风格迁移与角色条件的认知刺激)通过促动者风格建模和结构化骨架提取生成多方对话,从而缓解数据障碍。在此语料库的基础上,反思性认知对齐(RCA)框架将刺激性互动建模为一个序列决策过程,整合了受协议约束的认知链(PC-CoC)进行结构化推理,以及推理时价值对齐(IVA)以基于安全性和互动目标进行原则性回应选择。在六个基础LLM和两位独立评审的评估中,RCA在协议遵循、安全性和小组促进方面均优于标准提示基线。我们的代码可通过此 https URL 获取。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17536 (HTTP 429)

Authors: Jiyue Jiang, Ziyi Li, He Hu, Sheng Wang, Yuhan Chen, Yanyu Chen, Jingqi Zhou, Pengan Chen, Fei Ma, Irwin King, Yu Li, Chuan Wu

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2609.17536.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17536

Published: 2026-09-18T01:27:58.489Z


5. Relation Before Entity: Deferred Commitment in Language Model Factual Recall

Abstract:We ask whether relation-type information (e.g., capital-of) and entity-specific information (e.g., France to Paris) become causally active at the final-token position at the same depth during recall. Using four complementary causal diagnostics across four decoder-only models and eight prompt families, we find a robust temporal asymmetry: relation information becomes generation-controlling before entity information does. Relation onset precedes entity onset by 10-16 tested layers (31-44% of network depth) at threshold 0.4, with the ordering holding across all 16 model-threshold combinations for thresholds 0.2-0.5. Critically, entity information is not absent early: entity-token patching succeeds at 90-100% in early layers. Instead, entity commitment to generation is deferred: entity information is available at the entity-token position but becomes generation-controlling at the final token only after being routed there.

中文摘要

摘要:我们研究了关系类型信息(例如“首都属于”)和实体特定信息(例如“法国到巴黎”)在回忆时是否在最终标记位置的同一层深度上同时发挥因果作用。通过在四种仅解码器模型和八类提示上使用四种互补的因果诊断方法,我们发现了稳健的时间不对称性:关系信息比实体信息更早控制生成。在阈值为0.4时,关系信息的起始时间比实体信息早10-16个测试层(占网络深度的31-44%),且在阈值0.2-0.5的所有16个模型-阈值组合中均保持该顺序。关键是,实体信息在早期并非缺失:在早期层上对实体标记进行修补成功率达到90-100%。相反,实体对生成的承诺被延迟:实体信息在实体标记位置是可用的,但只有在被路由到最终标记后才成为生成控制信息。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17537 (HTTP 429)

Authors: Divyansh Agarwal

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2609.17537.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17537

Published: 2026-09-18T01:27:58.489Z


6. From Pixels to Pairs: A Comprehensive Benchmark of LLM-Based Key-Value Extraction in Noisy Document Settings

Abstract:Large language models (LLMs) are increasingly used for structured information extraction from documents, yet their behavior under realistic OCR noise remains poorly understood. We present a systematic benchmark of open-source instruction-tuned LLMs for key-value pair (KVP) extraction under both clean-text and noisy OCR conditions. We evaluate representative decoder-only models (Gemma, Mistral, Qwen2.5, LLaMA 3, and DeepSeek) on the FUNSD, CORD, and SROIE benchmarks using both Gold-text annotations and OCR outputs from PaddleOCR, EasyOCR, and Tesseract. A unified evaluation protocol isolates the effects of input quality, model design, and prompting under consistent conditions. The results show that modern LLMs act as strong semantic extractors when high-quality text is available, in some cases approaching supervised layout-aware systems. Under OCR noise, however, performance degrades substantially and performance gaps between models narrow as input corruption increases. Across all datasets, extraction performance is governed by two factors: semantic reasoning over text and preservation of textual fidelity under OCR noise. While larger models improve results on clean text, these gains diminish under noisy inputs, where OCR quality becomes the dominant factor. We also identify recurring failure modes, including key-value misalignment, hallucination, and numeric corruption. Our findings highlight the gap between clean-text evaluation and real-world deployment, emphasizing the need to jointly improve OCR quality, structural reasoning, and LLM-based semantic modeling.

中文摘要

摘要:大型语言模型(LLM)越来越多地被用于从文档中结构化信息提取,但它们在现实OCR噪声下的行为仍不充分。我们提出了系统基准测试,适用于开源指令调优LLM在纯文本和噪声OCR条件下键值对(KVP)提取。我们利用Gold文本注释和PaddleOCR、EasyOCR和Tesseract的OCR输出,在FUNSD、CORD和SROIE基准测试中评估了代表性仅限译码器的模型(Gemma、Mistral、Qwen2.5、LLaMA 3和DeepSeek)。统一的评估协议在一致条件下分离输入质量、模型设计和提示对输入的影响。结果表明,现代LLM在高质量文本可用时表现为强语义提取器,在某些情况下接近监督式布局感知系统。然而,在OCR噪声下,性能显著下降,模型间性能差距随着输入损坏增加而缩小。在所有数据集中,提取性能受两个因素控制:文本上的语义推理和OCR噪声下的文本忠实度保持。虽然较大模型在干净文本下能改善结果,但在噪声输入下,OCR质量成为主导因素,这些提升会减弱。我们还识别了反复出现的失败模式,包括键值错位、幻觉和数字损坏。我们的发现凸显了干净文本评估与现实应用之间的差距,强调了共同提升OCR质量、结构推理和基于LLM的语义建模的必要性。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17538 (HTTP 429)

Authors: Zahra Anvari, Vassilis Athitsos

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2609.17538.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17538

Published: 2026-09-18T01:27:58.489Z


7. MudawanSn: A Gold-Standard Wolof-Arabic Parallel Corpus for Machine Translation

Abstract:We present MudawanSn, a gold-standard resource of 1,271 sentence-aligned pairs manually translated from Wolof into Modern Standard Arabic (MSA). The source texts are drawn from the MasakhaNER corpus and cover politics, society, religion, and sports in Senegalese news discourse. Although multilingual resources such as FLORES-200 and NTREX include both Wolof and Arabic, no publicly available parallel corpus is specifically designed for the Wolof-Modern Standard Arabic language pair. We describe the corpus construction protocol, sentence alignment procedure, and quality-control workflow. We benchmark four machine translation systems spanning three architectural families: NLLB-200 (600M), mT5-base, and two AfriNLLB variants, showing that fine-tuning on MudawanSn yields substantial improvements in both translation directions. The best-performing model, AfriNLLB-12, achieves 7.76 BLEU and 30.72 chrF++ for Wolof-to-Arabic, and 8.75 BLEU and 33.08 chrF++ for Arabic-to-Wolof. The corpus is released under the CC BY-NC license and is publicly available on Hugging Face and GitHub.

中文摘要

摘要:我们介绍了 MudawanSn,这是一个黄金标准资源,包含 1,271 对经过人工从沃洛夫语翻译成现代标准阿拉伯语(MSA)的句子对。源文本取自 MasakhaNER 语料库,涵盖塞内加尔新闻话语中的政治、社会、宗教和体育内容。虽然 FLORES-200 和 NTREX 等多语言资源包含沃洛夫语和阿拉伯语,但目前没有专门为沃洛夫语-现代标准阿拉伯语语言对设计的公开平行语料库。我们描述了语料库构建协议、句子对齐程序及质量控制工作流程。我们对四种跨越三类架构的机器翻译系统进行了基准测试:NLLB-200(600M)、mT5-base 以及两个 AfriNLLB 变体,结果表明,在 MudawanSn 上进行微调能显著提升双向翻译性能。表现最佳的模型 AfriNLLB-12,在沃洛夫语到阿拉伯语翻译中实现了 7.76 BLEU 和 30.72 chrF++,在阿拉伯语到沃洛夫语翻译中实现了 8.75 BLEU 和 33.08 chrF++。该语料库在 CC BY-NC 许可证下发布,并可在 Hugging Face 和 GitHub 上公开获取。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17539 (HTTP 429)

Authors: Mouhamed Mbaye, Thierno Diop

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2609.17539.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17539

Published: 2026-09-18T01:27:58.489Z


8. Register Bias in Complexity-Based Large Language Model Routing

Abstract:Large language model services increasingly route each query to one of several models of differing capability, using a cheap estimate of query complexity to send easy queries to small models and hard queries to large ones. I show that this routing step is not register neutral: text written in a non-standard English register, African American English or the English of second-language writers, is systematically assigned a lower-capacity tier than a meaning-equivalent standard-English version of the same query. The effect is driven by a specific, common routing signal, input length, because non-standard registers omit function words and thus look shorter and therefore simpler; other complexity signals do not carry it. I demonstrate the disparity on 37,704 authentic learner sentence pairs and on a controlled parallel corpus. I then measure the quality consequence on a device, edge, and cloud model ladder and find that the harm is driven by pervasive model bias, every tier, including a frontier cloud model, answers non-standard-register queries significantly less accurately, while the marginal quality cost of the routing decision itself is not significant on this benchmark. Complexity-based routing thus compounds the exposure of the users that the models already serve worst.

中文摘要

摘要:大型语言模型服务越来越多地将每个查询路由到若干能力不同的模型之一,使用查询复杂性的粗略估计将简单查询发送到小模型,将复杂查询发送到大模型。我表明,这个路由步骤并非对语体中立:用非标准英语语体、非裔美国人英语或第二语言作者英语写的文本,系统性地被分配到容量较低的层级,而相同查询的意义等效标准英语版本则被分配到更高层级。这一效应由一个特定且常见的路由信号驱动,即输入长度,因为非标准语体省略了功能词,因此看起来更短,从而看起来更简单;其他复杂度信号并没有表现出这种现象。我在 37,704 对真实学习者句子中以及一个受控平行语料库中展示了这种差异。然后,我在设备端、边缘计算和云模型梯队上测量了质量影响,发现这种伤害是由普遍存在的模型偏差驱动的:每个层级,包括前沿云模型,对于非标准语体查询的回答准确性显著较低,而路由决策本身的边际质量成本在这个基准测试中并不显著。因此,基于复杂度的路由进一步加大了模型对其服务最差的用户群体的影响。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17542 (HTTP 429)

Authors: Simran Koul

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2609.17542.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17542

Published: 2026-09-18T01:27:58.489Z


9. Large Language Models Versus Physicians in Traditional Chinese Medicine: A Real-World Clinical Case Evaluation

Abstract:Large language models (LLMs) are increasingly being explored for clinical applications, yet their assessment for real-world traditional Chinese medicine (TCM) practice remains limited We constructed a clinical case library comprising 349 de-identified outpatient cases from 62 hospitals and evaluated 16 LLMs and a comparator cohort of 60 practicing TCM physicians using 60 representative cases selected from this library. Model outputs and physician reports were anonymized and scored by five senior TCM experts across nine diagnostic and therapeutic dimensions. Cutting-edge general-purpose LLMs achieved higher expert scores than the physician comparators, particularly for medical advice, treatment principles and selected diagnostic tasks. However, prescription-level analyses revealed discrepancies in herb selection, dosage, and treatment strategy, and qualitative safety review identified hallucinations and undesirable template-driven outputs. These findings highlight the potential of LLMs for TCM decision support while underscoring the need for physician oversight, safety constraints and prospective clinical evaluation.

中文摘要

摘要:大型语言模型(LLMs)在临床应用中的探索日益增多,但其在真实世界中对传统中医(TCM)实践的评估仍然有限。我们构建了一个临床病例库,包括来自62家医院的349例去标识化门诊病例,并使用从该库中选取的60例代表性病例评估了16个LLM模型和60名在职中医师的对照队列。模型输出和医生报告均进行了匿名处理,并由五名资深中医专家在九个诊断和治疗维度上进行评分。前沿通用LLMs在专家评分中优于医生对照,尤其在医疗建议、治疗原则和特定诊断任务上表现突出。然而,处方级分析显示在草药选择、剂量和治疗策略上存在差异,定性安全评估还发现了幻觉现象和不理想的模板驱动输出。这些发现突显了LLMs在中医决策支持中的潜力,同时强调了医生监督、安全约束和前瞻性临床评估的必要性。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17544 (HTTP 429)

Authors: Jiacheng Xie, Xiaoting Tang, Yang Yu, Jinpu Li, Shouli Li, Congcong Jing, Yantao Yang, Zhiyong Zhao, Ziyang Zhang, Qilin Song, Guanghui An, Dong Xu

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2609.17544.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17544

Published: 2026-09-18T01:27:58.489Z


Abstract:In this position paper, we argue that legal LLMs’ hallucinations should be evaluated as a failure of legal warrant rather than as factual inaccuracy or citation failure. We define claim-authority warrant as the context-sensitive relation between a consequential legal claim and authority that exists, applies to the relevant jurisdiction, is current for the date of analysis, has the legal status represented by the system, and supports the proposition asserted. Warranted legal generation is the broader system behavior that answers, narrows, asks, warns, corrects a false premise, or abstains according to that relation. The falsifiable prediction is that warrant metrics reveal material failures that answer accuracy, citation existence, generic attribution, LegalHalBench-style statute relevance, and CitaLaw-style sentence-citation alignment can miss. We sharpen this claim with a side-by-side comparison item and a small, reproducible pilot over public-rule tests. We then specify benchmark records, claim boundaries, support labels, mixed response-policy scoring, risk weights, annotation reliability reporting, and jurisdiction-specific authority ontologies. The result is a concrete research agenda for evaluating legal AI systems by whether their consequential claims are licensed by law.

中文摘要

摘要:在这篇立场论文中,我们认为法律大语言模型(LLM)的幻觉应被评估为法律授权的失败,而非事实错误或引用失败。我们将“主张-权威授权”定义为一种上下文敏感的关系,即一个具有影响力的法律主张与现存的、适用于相关司法管辖区的、在分析日期有效、具有系统所表示的法律地位并支持所主张命题的权威之间的关系。被授权的法律生成是更广泛的系统行为,它根据这种关系进行回答、缩小范围、提问、警示、纠正错误前提或节制。可证伪的预测是,授权指标能够揭示实际上的失败,而这些失败可能是回答的准确性、引用存在、通用归属、LegalHalBench风格的法规相关性以及CitaLaw风格的句子-引用对齐等方法无法发现的。我们通过并列对比条目和在公共规则测试上的一项小型、可重复的试点研究来强化这一论点。然后,我们具体说明了基准记录、主张边界、支持标签、混合响应策略评分、风险权重、注释可靠性报告以及特定司法管辖区的权威本体。其结果是一个具体的研究议程,用于评估法律人工智能系统是否其关键主张获得法律授权。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17546 (HTTP 429)

Authors: Maksym Taranukhin, Vered Shwartz

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2609.17546.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17546

Published: 2026-09-18T01:27:58.489Z


Agent Domain Papers

1. Making AI-Assisted Claims Independently Challengeable: Publication Authority and a Protocol for Falsifiable Publication Records

Abstract:AI-assisted claims can appear authoritative when evidence, analysis, human authorization, presentation, and correction history refer to different states. Provenance, attestation, and transparency expose history but alone do not specify the publication transition examined here. We develop Publication Authority as an exact-state, non-transferable, single-use publication capability and instantiate it in PAC-2026 (Publication-Accountability Calculus), a machine-readable AIJIM Protocol candidate. We evaluate its fourth bounded semantic freeze (SF-4), a fixed-profile specification designed for replaceable bindings. Six obligations govern evidence, runs and artifacts, measurement disclosure, authorization, surface correspondence, and lifecycle continuity. Each yields a target-bound witness, localized counterexample, or localized unverifiability; none can compensate for another. Only a fresh, complete all-pass record derives the permit consumed by one atomic publication transition. We use identity vectors, adversarial cases, finite models, and historical implementations. Ten models explored 110,764 safe reachable states; 76 unsafe configurations produced the expected violation or observer countermodel. A reader surface passing its correspondence check cannot authorize publication unless the accepted record admits that surface. SF-4 separates evidence horizon from verification time and rejects an authentic but causally invalid authorization. A historical predecessor path reproduced 17 frozen authorization-successor outcomes. A later in-house, instance-blind test of known case classes matched all 183 scored expectations; same-host package execution reproduced its 240 archived observations. Results support internal coherence, bounded safety, fault sensitivity, and limited constructibility, but not factual truth, general refinement, blind interoperability, field efficacy, or standards status.

中文摘要

摘要:当证据、分析、人工授权、呈现和修正历史涉及不同状态时,人工智能辅助的声明可能看起来具有权威性。来源、证明和透明度揭示了历史,但单独不足以指定此处所研究的出版转换。我们提出了出版权威(Publication Authority),作为一种精确状态、不可转让、一次性使用的出版能力,并在 PAC-2026(出版责任演算,Publication-Accountability Calculus)中实现,它是机器可读的 AIJIM 协议候选方案。我们评估了其第四个有界语义冻结(SF-4),这是一种为可替换绑定设计的固定配置规范。六项义务管理证据、运行和工件、测量披露、授权、表面对应和生命周期连续性。每一项都产生目标绑定的见证、本地化反例或本地化不可验证性;没有一项可以替代另一项。只有一个新的、完整的全通过记录才能导出由一次原子出版转换消耗的许可。我们使用身份向量、对抗性案例、有限模型和历史实现。十个模型探索了110,764个安全可达状态;76个不安全配置产生了预期的违规或观察者反模型。除非被接受的记录承认该表面,否则读者表面通过其对应性检查并不能授权出版。SF-4 将证据视界与验证时间分离,并拒绝真实但因果无效的授权。一个历史前身路径复制了17个冻结的授权-成功者结果。后来内部、实例盲测试已知案例类别,匹配了所有183个评分预期;同主机软件包执行复现了其240条存档观察结果。结果支持内部一致性、有界安全性、故障敏感性和有限构造性,但不支持事实真实性、通用优化、盲互操作性、现场效能或标准地位。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17631 (HTTP 429)

Authors: Torsten Olivi Tiltack, Yifei Dong, Kun Yu, Xu Wang, Wei Liu, Jianlong Zhou, Ren Ping Liu, Fang Chen

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.17631.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17631

Published: 2026-09-18T01:28:53.527Z


2. EvolveTrade: Experience-Driven Policy Refinement for Self-Evolving LLM Trading Agents

Abstract:Large language model (LLM) trading agents can combine market data, news, and executable analysis, but their behavior is often controlled by static hand-written tool-use policies that are fixed before deployment. This limits their ability to adapt how they gather evidence, invoke tools, verify signals, and manage risk under changing market regimes. We introduce EvolveTrade, a self-evolving framework that treats the system prompt of a tool-using trading agent as a text-parameterized policy. After each update interval, a Policy Agent revises this policy using accumulated decision traces and realized portfolio feedback, while keeping the backbone LLM fixed. The updated policy is then used for the next batch of trading decisions, enabling the agent to refine its information-acquisition and portfolio-construction procedure over time. Experiments across multiple market regimes and two LLM backbones show that EvolveTrade often improves Sharpe Ratio and Cumulative Return over fixed-policy LLM baselines, achieving the improved SR and CR in most evaluated settings. Behavioral analyses further show that self-evolved policies increase code-mediated analysis and activate regime-relevant computations; case-level policy-to-return attributions trace how policy-induced allocation changes contribute to realized return differences. These results suggest that adapting the reusable procedure governing tool use is a key direction for building more robust LLM trading agents.

中文摘要

摘要:大型语言模型(LLM)交易代理可以结合市场数据、新闻和可执行分析,但它们的行为通常由在部署前固定的静态手写工具使用策略所控制。这限制了它们在不断变化的市场环境下调整证据收集、调用工具、验证信号和管理风险的能力。我们提出了EvolveTrade,一个自我进化框架,它将工具使用交易代理的系统提示视为文本参数化策略。在每个更新间隔后,策略代理会利用累积的决策轨迹和已实现的投资组合反馈来修订该策略,同时保持基础LLM不变。更新后的策略随后用于下一批交易决策,使代理能够随着时间的推移优化其信息获取和投资组合构建流程。在多种市场环境和两种LLM骨干模型上的实验证明,EvolveTrade通常能在固定策略LLM基线之上提升夏普比率和累计收益,在大多数评估设置中实现更高的SR和CR。行为分析进一步显示,自我进化的策略增加了代码介导的分析,并激活与市场环境相关的计算;案例级策略到收益归因跟踪了策略引起的资产配置变化对已实现收益差异的贡献。这些结果表明,调整管理工具使用的可重复流程是构建更稳健LLM交易代理的关键方向。

LLM Analysis

Q: 这篇论文试图解决什么问题? 该论文旨在解决\*\*大型语言模型(LLM)交易代理中工具使用策略静态固化\*\*的核心问题。具体而言,现有LLM交易代理虽然具备整合市场数据、新闻与可执行分析的能力,但其决策程序(即“策略”)通常在部署前由人工预定义并以固定系统提示的形式冻结,无法根据实际交易经验与市场环境的变化进行自我调整。 该问题可进一步分解为以下几个层面: - \*\*静态策略对非平稳市场的适应性不足\*\* 金融市场具有显著的\*\*非平稳性\*\*(non-stationarity),不同市场机制(如震荡、急跌反弹、稳步上行)要求差异化的信息获取、信号验证与风险管理方式。然而,静态策略迫使代理在截然不同的市场环境中重复使用相同的分析范式与工具调用模式,导致其在变化的市场条件下难以维持稳健表现。 - \*\*经验反馈无法沉淀为可复用的决策程序\*\* 现有方法即使允许代理调用外部工具(如价格检索、新闻搜索、代码解释器),其“何时、为何、如何”调用这些资源的操作规程在部署后保持不变。代理每日产生的大量决策轨迹(包括资产级推理与投资组合反馈)未能被系统性利用来修订其可复用的工具使用流程。 - \*\*分析词汇与风险控制的僵化\*\* 如论文所示,固定策略导致代理被锁定在一套固定的分析指标中(例如始终计算20日收益、年化波动率等),而极少激活与当前市场机制真正相关的计算(如下跌行情中的风险价值(VaR)与信号归一化,或上行趋势中的指数移动平均(EMA)与相对强弱指标(RSI))。代理缺乏机制去扩展或调整其分析工具箱。 为应对上述挑战,论文提出 \*\*EvolveTrade\*\* 框架,其核心思想是: > 将交易代理的系统提示视为一个\*\*文本参数化的策略\*\* π_t ,并引入一个独立的策略代理(Policy Agent),在每个交易批次结束后,基于累积的决策轨迹 h_t 与实现的投资组合反馈 g_t 在线修订该策略: >


π(kN+1) = f(update)(π_(kN), B_k)

其中 B_k 为第 k 个批次的历史记录集合。通过这种方式,代理能够在保持底层LLM参数与工具接口完全冻结的前提下,使其信息获取、证据验证与投资组合构建程序随经验持续演化。 Q2: 有哪些相关研究? 根据论文第2节,相关研究主要分布于以下两个方向: ### 1. LLM 金融交易代理(LLM Agents for Financial Trading) 该方向探索如何为 LLM 配备记忆、工具、多模态信号与多智能体协作能力,以支持金融决策。 - **记忆与角色增强系统** FinMem(Yu et al., 2025b)引入分层记忆以跨会话保留市场观察;TradingAgents、StockAgent、QuantAgent 与 TradExpert(Xiao et al., 2025; Zhang et al., 2024a; Xiong et al., 2025a; Ding et al., 2024)则通过专门化的代理角色、模拟交易环境或固定分析管道来组织决策。 - **多模态与实时数据平台** FinAgent 及相关金融代理平台(Zhang et al., 2024b; Han et al., 2025; Yang et al., 2024b)整合多模态输入、实时数据、领域工具与反思式决策机制。 - **基准评测** AI-Trader、LiveTradeBench 与 FinAgentBench(Fan et al., 2025; Yu et al., 2025a; Choi et al., 2025)在更贴近现实的金融信息与市场条件下评估 LLM 代理。 - **可靠性研究** 近期工作开始关注 LLM 交易中的可靠性问题,包括噪声源信任、虚假股票代码记忆与时间泄漏(Li et al., 2026; Jeon and Lee, 2026; Benhenda, 2026)。 **与 EvolveTrade 的区别**:上述系统虽然证明了 LLM 可整合异构金融证据,但其观察通道、提示模板、角色结构或工具工作流通常在部署前即被固定, governing 信息获取、证据验证与风险控制的**操作规程**基本保持静态。 ### 2. 自我进化 LLM 代理(Self-evolving LLM Agents) 该方向研究 LLM 代理如何基于自身的执行轨迹、反馈或任务结果进行部署后改进。 - **工具使用基础** ReAct(Yao et al., 2023)与 Toolformer(Schick et al., 2023)等早期工作实现了推理与外部动作的交错,并探索了 API 调用时机。 - **提示与管道优化** 一类通用方法通过验证信号、类梯度反馈或进化搜索来修订指令或 LM 程序(Pryzant et al., 2023; Yang et al., 2024a; Khattab et al., 2024; Yuksekgonul et al., 2025; Agrawal et al., 2026; Choi et al., 2026; Zhang et al., 2025)。 - **代理级自我进化** 更进一步的代理级方法将成功与失败转化为反思记忆、可复用工作流或推理策略(Shinn et al., 2023; Wang et al., 2024, 2025; Ouyang et al., 2026; Pan et al., 2026)。 - **交易领域的策略进化** ATLAS 与 SHARP(Papadakis et al., 2026; Chen et al., 2026)尝试从市场反馈中调整提示或结构化策略;AlphaQuanter 与 FLAG-Trader(Deng et al., 2026; Xiong et al., 2025b)则表明学习的工具编排或策略优化可改善固定多代理基线。 **与 EvolveTrade 的区别**:上述方法或优化推理指令,或维护经验记忆,或离线学习工具使用策略,而非**持续地在线精炼**已部署代理在信息收集、验证与行动过程中使用的**操作规程**。EvolveTrade 的独特之处在于将系统提示视为**文本参数化的工具使用策略** π_t ,并基于代理自身的工具调用轨迹、交易决策与投资组合反馈进行在线更新:

π(kN+1) = f(update)(π_(kN), B_k)
Q3: 论文如何解决这个问题? 论文通过提出 **EvolveTrade** 框架,将 LLM 交易代理的系统提示视为一个**文本参数化的可进化策略**,并设计了一套基于实现交易经验的在线策略精炼机制。具体解决方案包含以下层面: ### 1. 将系统提示形式化为时变策略 论文首先将交易代理的决策环境形式化。在第 t 个交易日开始时,代理的状态为:

St = (π_t, t, P(t-1), T)
其中 πt 是当日生效的工具使用策略(以自然语言系统提示实现), P(t-1) 为先前投资组合状态, T 为可用工具集(价格检索、新闻搜索、Python 代码解释器)。与静态基线不同,此处 π_t 被显式建模为随时间变化的变量,允许在不同交易日持有不同的策略文本。 ### 2. 构建可审计的策略反馈记录 为消除金融市场中延迟且嘈杂的反馈信号带来的歧义,论文将决策轨迹与事后反馈配对,形成精炼记录:

h_t = (w_t, d_t^(dec))

R_t = (h_t, g_t)

  • h_t 包含代理提交的资产配置 w_t 及每支资产的推理依据 d_t^(dec) (即代理为设定该权重所使用的证据与论证) - g_t 为事后反馈,记录当日投资组合收益、各资产收益及仓位变化 该记录 R_t 使得策略修订能够基于**代理具体做了什么**以及**市场随后如何回应**进行归因,而非仅依赖孤立的盈亏数字。 ### 3. 在线策略自我进化 EvolveTrade 以固定的 N 个交易日为批次进行策略更新。设第 k 个批次覆盖交易日 (k-1)N+1, dots, kN ,则该批次内的策略保持恒定:

π((k-1)N+1) = ·s = π(kN)
批次结束时,系统收集该批次的全部记录:
Bk = R((k-1)N+1), dots, R(kN)
随后,一个独立的 **Policy Agent**(由相同或另一 LLM 实现)执行基于语言的精炼步骤:
π
(kN+1) = f(update)(π(kN), B_k)

f(update)(π, B) = LLM(I(update), π, B)
其中 I(update) 为固定的更新指令,要求 Policy Agent: - 分析 B_k 中的记录,识别当前提示中哪些具体条款导致了良好或糟糕的表现 - 将观察到的行为与提示中的具体行项关联,执行**有依据的编辑**(grounded edits),而非泛泛重写 - 输出完整的更新后策略文本 ### 4. 策略内容的可进化维度 策略精炼不限于高层交易逻辑,而是涵盖工具使用的完整操作规程。一次精炼可能改变: - **工具选择与优先级**:例如增加代码解释器调用频率,减少重复性新闻检索 - **查询 formulation**:如调整价格回溯窗口长度或新闻检索关键词 - **信号解释与交叉验证**:例如引入”新闻信号需经价格确认后方可增仓”的规则 - **风险控制与仓位调整**:如在不同市场机制下激活风险价值(VaR)、信号归一化或趋势跟踪指标(EMA、RSI、SMA) - **输出结构**:要求代理在最终输出中包含可审计的目标权重与再平衡动作 ### 5. 保持基础能力冻结以实现模块化进化 在整个过程中,底层 LLM 参数 pθ 与工具接口 T 始终保持冻结。唯一被学习的对象是**策略文本本身**。这种设计使得框架可应用于任何已部署的 LLM 交易代理,无需重新训练模型或修改工具实现,从而将”市场经验的沉淀”与”基础推理能力”解耦。 通过上述机制,EvolveTrade 使交易代理能够在不同市场机制(震荡、急跌反弹、上行趋势)中逐步激活此前被静态策略抑制的分析工具与风险控制程序,实现从固定分析词汇到**机制相关计算**的自适应扩展。 Q4: 论文做了哪些实验? 论文在第6节及附录中设计并执行了多组实验,涵盖不同市场机制、模型主干与消融条件,具体可分为以下几个层面: ## 1. 实验设置与评测环境 - **交易环境**:基于 LiveTradeBench 构建的日收盘回测模拟。每个交易日收盘后,代理提交目标权重,系统按收盘价再平衡组合。 - **投资范围**:15只美股蓝筹股(如 AAPL、MSFT、NVDA、JPM 等)及现金。 - **评测窗口**:每月为一个在线评测周期,代理每日观察市场、提交配置、接收组合反馈,并按固定间隔更新策略。 - **模型主干**:分别使用 **GPT-5-mini** 与 **Gemini-2.5-Flash** 作为交易代理与策略代理的底层模型。 - **策略更新间隔**:默认每 N=5 个交易日执行一次策略精炼。 - **市场机制**:为检验跨机制泛化性,实验覆盖了六种不同的后知识截断市场机制: - GPT-5-mini:2025年1月(震荡)、2025年4月(急跌反弹)、2025年9月(上行趋势) - Gemini-2.5-Flash:2025年11月(看跌)、2026年2月(震荡)、2026年4月(看涨) ## 2. 基线方法 实验对比了规则基线与多组 LLM 基线,以隔离工具访问与策略进化的效应: **规则基线** - SPY、Buy and Hold (B&H)、MACD、KDJ&RSI、ZMR、SMA **LLM 基线** - **Static Base Agent**:固定观察窗口(价格+新闻),无工具调用。 - **Static Tool-Calling (TC) Agent**:具备完整工具集(价格、新闻、代码解释器),但策略文本固定。 - **EvolveBase**:无工具,但允许策略自我进化,以隔离“纯策略进化”在无工具场景下的效应。 - **EvolveStrategy**:策略代理仅更新高层交易逻辑(如信号组合、仓位调整规则),工具调用协议与操作细节保持固定。 - **EvolveTrade (Ours)**:策略代理全面更新交易代理的完整策略,包括工具调用协议与操作规程。 ## 3. 评测指标 - 年化夏普比率(SR↑) - 累计收益率(CR%↑) - 最大回撤(MDD%↓) - 胜率(WR%↑) - 日波动率(Vol%↓) ## 4. 主实验结果 表1报告了各代理在六种市场机制下的平均表现(LLM 方法均运行3次)。核心发现包括: - **EvolveTrade 在多数设置中取得最佳风险收益表现**。对于 GPT-5-mini,其在2025年1月与9月实现 LLM 方法中最高的 SR 与 CR;对于 Gemini-2.5-Flash,其在2025年11月取得最佳,在2026年2月取得次佳。 - **策略进化带来的收益不仅限于相对其他 LLM 基线**。例如,GPT-5-mini 在2025年9月达到 CR 6.84%,超过该期所有规则基线;Gemini-2.5-Flash 在2026年2月达到 SR 2.75 与 CR 2.92%,亦持续优于全部规则基线。 ## 5. 行为与机制分析 ### 5.1 分析指标激活模式 图3对比了 Static TC Agent 与 EvolveTrade 在代码解释器中实际调用的分析指标。结果显示: - Static TC Agent 在所有机制中均锁定于同一套狭窄指标(如 20日收益、年化波动率、夏普比率)。 - EvolveTrade 在策略进化后激活了基线从未调用的机制相关指标:如在2025年4月急跌中启用 **VaR** 与**信号归一化**;在2025年9月上行趋势中启用 **EMA、RSI、SMA** 等趋势跟踪指标。 ### 5.2 工具调用频率变化 图4量化了每日平均工具调用次数: - 价格检索工具( t(price) )在两组代理中均维持约1次/日。 - **代码解释器( t(code) )调用显著上升**:从 Static TC Agent 的约 1.0 次/日提升至 EvolveTrade 的 2.6–4.5 次/日。 - 新闻搜索工具( t_(news) )未出现显著增长。 这表明策略进化将代理行为从外部文本搜集**重定向为可执行的数量化分析**。 ### 5.3 案例级归因分析 图5给出了2025年1月24日 NVDA 急跌前的持仓对比: - Static TC Agent 持有 NVDA 10.7%,依据短期收益与正面新闻。 - EvolveTrade 经策略精炼后仅持有 2.9%,因其策略优先要求更长周期的风险调整证据,并对高波动仓位执行组合改进与换手率检查。 - 次日 NVDA 下跌 17.0%,EvolveTrade 当日回报为 -0.03%,而 Static TC Agent 为 -1.11%。 附录E进一步提供了2025年9月10日的单日归因(图7),显示 EvolveTrade 通过将代码输出转化为可审计的目标权重,在 JPM、CAT、WMT、TSLA 上获得了次日超额收益。 ### 5.4 长期演化稳定性 为验证持续更新是否导致策略文本膨胀与性能衰减,论文在 GPT-5-mini 上执行了 **50个交易日** 的长期实验(表2、图6): - EvolveTrade 在 SR 与 CR 上仍优于所有基线。 - 策略文本长度呈现**非单调变化**:随新指导的加入而增长,但周期性地收缩,表现为冗余指令的整合而非无限累积。 ### 5.5 更新间隔敏感度 表3考察了策略更新频率( N = 1, 3, 5, 7 天)对 GPT-5-mini 平均表现的影响: - N=5 天取得最高的平均 SR 与 CR。 - N=1 天(每日更新)表现最差,提示过度频繁的修订可能过拟合短期噪声。 - 该实验表明策略进化需要在**适应性**与**反馈累积稳定性**之间取得平衡。 ## 6. 交易成本与稳健性检验 附录D(表6)在 10 bps 的比例交易成本下重新计算了 SR 与 CR,并报告了组合换手率(TO): - EvolveTrade 在考虑摩擦后仍保持其在1月、9月(GPT-5-mini)及11月(Gemini-2.5-Flash)的领先优势。 - 其换手率在所有6个评测窗口中均低于 Static TC Agent 与 EvolveStrategy,说明收益提升并非源于激进再平衡。 附录C(表4、表5)报告了3次独立运行的均值与标准差,显示 EvolveTrade 的增益具有跨运行一致性,而非单次幸运结果。 ## 7. 失效模式分析 附录F(表7)分析了 EvolveTrade 在两个4月窗口中表现弱于最强静态基线的原因: - 在这两个时期,EvolveTrade 的平均现金仓位分别比基线高出 10.1 与 35.9 个百分点。 - 过高的现金储备使其在2025年4月反弹与2026年4月牛市中错失了股权收益,揭示了策略进化在特定机制下可能产生过度防御性配置。 Q5: 有什么可以进一步探索的点? 基于论文的局限性与实验洞察,以下方向值得进一步探索: ### 1. 自适应策略更新调度 当前实现采用固定的批次长度 N (如每 5 个交易日更新一次)。表 3 的敏感度分析表明,过于频繁的更新( N=1 )会因过拟合短期噪声而损害表现,而较长的间隔又可能延迟对机制突变的响应。未来可探索**基于市场条件触发的动态更新策略**,例如根据已实现波动率、组合回撤幅度或策略文本的预测不确定性来自适应调整更新频率:

Nt = f(σ(t)^(realized), MDD_t, H_t)

2. 更真实的执行摩擦建模 论文在附录 D 中仅以 10 bps 的比例交易成本近似真实摩擦,未纳入**滑点(slippage)、市场冲击(market impact)与流动性约束**。对于日度再平衡且持仓可能集中于少数资产的 LLM 代理,这些摩擦在回测与实盘之间的差异尤为显著。后续研究可整合限价单簿(LOB)模拟或成交量加权平均价格(VWAP)执行模型,检验策略进化是否仍能提升净夏普比率。 ### 3. 策略文本的压缩与长期记忆机制 图 6 显示策略长度在 50 个交易日内呈非单调增长,但长期运行下仍存在文本膨胀风险。未来可引入**显式的策略摘要与遗忘机制**:在每次精炼时,要求 Policy Agent 不仅追加新规则,还需识别并删除冗余或已失效的指令,将策略长度约束为与当前市场机制相关的“工作集”,从而避免上下文窗口溢出与指令冲突。 ### 4. 跨机制迁移与元学习 实验表明,EvolveTrade 在不同市场机制(震荡、急跌、上行)中激活了不同的分析指标集合。一个开放问题是:进化后的策略是否包含**可迁移的元策略**?例如,在 2025 年 4 月急跌期间习得的尾部风险控制规则,能否在 2026 年 2 月的另一场下行中被快速唤起,而非重新从零进化?引入基于机制识别的策略路由或元学习初始化,可能显著加速新环境下的适应。 ### 5. 反事实归因与噪声鲁棒性 金融市场反馈具有高度随机性,单日盈亏可能掩盖策略的真实质量。当前框架直接利用实现回报 gt 作为反馈信号,未来可探索**反事实(counterfactual)评估**:在策略更新前,要求 Policy Agent 评估“若昨日采取另一套规则,组合表现将如何变化”,或构建基于合成控制方法的因果归因,以降低将运气误读为能力(spurious policy improvement)的风险。 ### 6. 安全护栏与价值对齐 附录 F 揭示了策略进化可能产生过度防御性配置(如在 2026 年 4 月牛市中持有 40.1% 现金,错失上涨收益)。这提示需要**硬约束与软偏好的分层治理**:在策略文本之外,维护一组不可由进化过程覆盖的安全规则(如最大现金上限、单一资产集中度下限),或通过基于人类偏好的强化学习(RLHF)对 Policy Agent 的修订方向进行对齐,防止自我进化走向过度保守或过度激进。 ### 7. 工具空间的动态扩展 当前工具集 T = t(price), t(news), t(code) 是预定义且固定的。未来可研究**工具发现(tool discovery)**:允许策略代理在精炼过程中提出新的 API 调用需求(如接入期权隐含波动率、供应链图数据库或另类数据),并由外部系统评估其可行性与增量价值,从而突破现有分析词汇的天花板。 ### 8. 可解释性辅助的人机协同 EvolveTrade 生成的策略文本具有高度可读性(如图 12–16 所示)。未来可设计**交互式审查界面**,使人类风控专家能够在每个批次结束时审阅 Policy Agent 提出的策略变更,对“激活 VaR 计算”或“提高现金仓位”等关键修改进行批准、驳回或修正,将 LLM 的自我进化能力与人类领域知识相结合,形成混合增强型交易系统。 Q6: 总结一下论文的主要内容 该论文针对大型语言模型(LLM)金融交易代理中**工具使用策略静态固化**的核心瓶颈,提出了一种在线自我进化框架 **EvolveTrade**,并通过多市场机制、多模型主干的实验验证其有效性。 ### 1. 研究背景与动机 现有 LLM 交易代理虽能整合市场数据、新闻与代码工具进行灵活推理,但其信息获取、证据验证与资产配置的**操作规程**通常以固定系统提示的形式在部署前人工设定,此后不再改变。论文指出,这种静态策略存在三重局限: - **非平稳市场适应性差**:金融市场机制持续变化(震荡、急跌、上行),固定策略无法动态调整分析重心; - **经验反馈无法沉淀**:代理每日产生大量决策轨迹与组合回报,但固定提示无法吸收这些信号来改进可复用的决策流程; - **分析词汇僵化**:实证显示,静态代理被锁定在狭窄的分析指标集合中(如 20 日收益、年化波动率),极少激活与当前机制真正相关的指标(如 VaR、EMA、RSI)。 ### 2. 核心方法:EvolveTrade 论文将系统提示重新概念化为一个**文本参数化的时变策略** πt 。在每个交易批次结束后,一个独立的 **Policy Agent** 读取该批次的交易经验,将策略从 π_t 精炼为 π(t+1) ,而底层 LLM 参数与工具接口始终保持冻结。 **形式化框架**: - 第 t 日代理状态: St = (π_t, t, P(t-1), T) ,其中 P(t-1) 为前期组合状态, T 为工具集; - 决策轨迹: h_t = (w_t, d_t^(dec)) ,包含资产配置与每资产推理依据; - 事后反馈: g_t ,包含实现收益、资产收益与仓位变化; - 精炼记录: R_t = (h_t, g_t) ; - 批次记录(每 N 日): B_k = R((k-1)N+1), dots, R_(kN) ; - 策略更新:

π(kN+1) = f(update)(π(kN), B_k) = LLM(I(update), π(kN), B_k)
Policy Agent 的更新指令 I
(update) 要求将具体观察到的行为与提示文本中的特定条款关联,执行有依据的编辑(grounded edits),可修改工具选择、信号交叉验证规则、风险控制与输出结构等。 ### 3. 实验设计 实验基于日收盘回测环境,投资范围为 15 只美股蓝筹股加现金,评测窗口覆盖六种后知识截断的市场机制: - **GPT-5-mini**:2025 年 1 月(震荡)、4 月(急跌反弹)、9 月(上行趋势) - **Gemini-2.5-Flash**:2025 年 11 月(看跌)、2026 年 2 月(震荡)、2026 年 4 月(看涨) 对比基线包括规则策略(SPY、B&H、MACD 等)与 LLM 策略(Static Base、Static Tool-Calling、EvolveBase、EvolveStrategy)。默认每 N=5 个交易日执行一次策略更新。核心评测指标为年化夏普比率(SR)、累计收益率(CR)、最大回撤(MDD)、胜率(WR)与日波动率(Vol)。 ### 4. 主要结果与发现 - **风险收益表现提升**:EvolveTrade 在多数模型-机制组合中取得最优或次优的 SR 与 CR。例如,GPT-5-mini 在 2025 年 1 月与 9 月分别实现 SR 5.12 与 8.43;Gemini-2.5-Flash 在 2025 年 11 月表现最佳,在 2026 年 2 月优于全部规则基线。 - **机制相关指标的激活**:行为分析显示,静态代理在三种机制中调用几乎相同的指标;而 EvolveTrade 在 4 月急跌中激活了 **VaR** 与**信号归一化**,在 9 月上行中激活了 **EMA、RSI、SMA** 等趋势跟踪指标,扩展了分析词汇。 - **工具使用模式转变**:EvolveTrade 的代码解释器调用频率从约 1.0 次/日提升至 2.6–4.5 次/日,而新闻检索未显著增加,表明策略进化将重心从文本搜集转向**可执行的数量化分析**。 - **案例级归因**:2025 年 1 月 24 日 NVDA 大跌前,EvolveTrade 经策略精炼将 NVDA 仓位从 10.7% 降至 2.9%,次日 NVDA 下跌 17.0%,EvolveTrade 实现 -0.03% 而静态代理为 -1.11%,展示了策略-仓位的可追踪因果链。 - **长期稳定性与成本稳健性**:在 50 个交易日长期实验中,EvolveTrade 持续领先;引入 10 bps 交易成本后,其收益优势仍然保持,且换手率低于多数 LLM 基线。策略文本长度呈非单调变化,可周期性地整合冗余指令。 - **更新频率敏感度**:每日更新( N=1 )因过拟合短期噪声而表现最差, N=5 达到最佳平衡,提示策略进化需要适度的反馈累积周期。 ### 5. 局限与未来方向 论文指出当前框架采用**固定更新间隔**,未根据市场变化动态调度;交易成本估计未纳入滑点与流动性约束;且在个别窗口(如 2025 年 4 月与 2026 年 4 月)出现过高的防御性现金配置,错失上涨收益。 ### 6. 结论 EvolveTrade 表明,LLM 交易代理的核心优化对象不应仅是单次交易决策,而是**可复用的信息获取与行动程序本身**。通过将系统提示视为可从实现经验中在线精炼的策略文本,代理能够在不重新训练模型、不修改工具的前提下,持续提升风险调整收益并激活机制相关的分析能力。这为构建适应非平稳金融市场的稳健 LLM 代理提供了关键方向。 Q7: 想要进一步了解论文 以上只是了解一篇论文的几个基本FAQ。如果你还想与Kimi进一步讨论该论文,请点击 这里 为你跳转Kimi AI网页版,并启动一个与该论文相关的新会话。

Authors: Sehee Kim, Yumin Choi, Minki Kang, Sung Ju Hwang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.17632.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17632

Published: 2026-09-18T01:28:53.527Z


3. One Color Preprocessing Improves DSATUR

Abstract:The Graph Coloring Problem (GCP) is NP-hard and DSATUR stands as one of the fastest heuristics for it despite producing colorings that typically use more colors than state-of-the-art coloring algorithms. We propose SSLD (Semidefinite Spectral Learning with DSATUR), which improves DSATUR by preprocessing a first good color class before letting DSATUR complete coloring the rest of the given graph. We obtain this color class from a Semidefinite Programming (SDP), similar to an SDP used to compute the Lovász theta number. To the best of our knowledge, SSLD is the first approach to improve DSATUR by preprocessing through fixed color classes. We evaluate SSLD against DSATUR and against a naive 1-color-class preprocessing algorithm on DIMACS instances, random graphs (Erdős—Rényi, Watts-Strogatz, Barabási—Albert), Frequency Assignment and Job Shop Scheduling instances. SSLD matches or beats DSATUR in almost every case across over 1600 benchmark instances, and out performs the naive GISD baseline, allows us to confirm the value brought by the SDP-guided choice of the first color class. This quality comes at a runtime cost of roughly 195 times slower that DSATUR, but demonstrating that SDP-guided preprocessing of a first color class is a direction for future improvements.

中文摘要

摘要:图着色问题(GCP)是NP难问题,尽管DSATUR通常生成的着色使用的颜色比最先进的着色算法更多,但它仍然是其中最快的启发式算法之一。我们提出了SSLD(带DSATUR的半正定谱学习),通过在让DSATUR完成对给定图的其余部分的着色之前,预处理出第一个好的颜色类,从而改进了DSATUR。我们通过半正定规划(SDP)得到这个颜色类,其方法类似于用于计算洛伐兹θ数的SDP。据我们所知,SSLD是第一个通过固定颜色类预处理来改进DSATUR的方法。我们在DIMACS实例、随机图(Erdős—Rényi、Watts-Strogatz、Barabási—Albert)、频率分配和作业车间调度实例上,将SSLD与DSATUR以及一个简单的单颜色类预处理算法进行了评估。在超过1600个基准实例中,SSLD几乎在每种情况下都能匹配或优于DSATUR,且优于简单的GISD基线,这使我们能够确认SDP指导的首个颜色类选择所带来的价值。这种优势以大约比DSATUR慢195倍的运行时间为代价,但证明了通过SDP指导预处理首个颜色类是未来改进的一个方向。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17633 (HTTP 429)

Authors: Adam Nouira, Lucas Isenmann

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.17633.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17633

Published: 2026-09-18T01:28:53.527Z


4. Physics-Constrained Digital Twins for Sensor Integrity in Urban Pedestrian Flow: Detecting Stealthy False Data Injection with Conformal Guarantees

Abstract:City pedestrian counting systems now feed economic indicators, planning decisions and safety operations, yet the twins built on top of them treat the incoming stream as ground truth. We study what happens when it is not. We formalise stealthy false data injection for city-scale pedestrian sensing, where the map from latent flow to observation is far more rank deficient than in the power and water networks for which stealth has been characterised. Our twin estimates directed flows on the pedestrian street graph, assimilates counts through a learned graph-localised gain, and is trained against a flow conservation residual that couples metered and unmetered segments. Detection combines the innovation with that residual, and the alarm threshold is set by adaptive conformal calibration rather than by hand. To measure what the physics buys, we define the attack margin, the relative reduction in worst-case corruption of the estimated flow field, achieved against a white-box adversary that optimises directly through the twin. On six years of Melbourne data the margin reaches 0.54 against a single compromised device and falls to 0.19 when a third of the fleet is compromised, on a network where only 1.18 per cent of walkable segments are metered. Replacing the street graph by a distance graph collapses it to 0.09, which shows that the gain comes from the conservation law rather than from locality.

中文摘要

摘要:城市行人计数系统现在为经济指标、规划决策和安全操作提供数据,但构建在其上的数字孪生将输入数据流视为真实情况。我们研究在输入数据流不真实时会发生什么。我们形式化了针对城市规模行人感知的隐蔽性虚假数据注入,其中从潜在流量到观测的映射比在已表征隐蔽性的电力和水网络中更具秩缺陷。我们的孪生系统在行人街道图上估计定向流量,通过学习到的图局部增益来同化计数,并通过耦合计量和未计量段的流量守恒残差进行训练。检测将创新量与该残差结合,并通过自适应拟合校准设置报警阈值,而非手动设置。为了衡量物理保护的作用,我们定义了攻击边界,即在面对直接通过孪生系统进行优化的白盒攻击者时,估计流场最坏情况破坏的相对降低。在六年的墨尔本数据上,对单个受损设备攻击时,攻击边界达到0.54;当三分之一的设备受损时,下降至0.19,而该网络中仅有1.18%的可步行路段被计量。如果将街道图替换为距离图,攻击边界会降至0.09,这表明增益来自守恒定律而非局部性。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17635 (HTTP 429)

Authors: Oscar Mogollon Gutierrez, Fatemeh Ghasemi, Mohammadhossein Homaei, Andres Caro, Mar Avila

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.17635.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17635

Published: 2026-09-18T01:28:53.527Z


5. What You Can’t See Is Still What You Learn: A Preregistered Sixty-Society Confirmation That Evidence Masking Drives Compositional Generalization

Abstract:Restricting what a module can read may improve what a system learns to compute. We test this in a preregistered confirmation with sixty four-cell systems sharing a frozen language-model backbone and communicating through learned continuous packets. Five conditions vary evidence masking, ownership markers, and replacement of foreign evidence with neutral filler, across six initialization clusters, each with two data orders, on one fresh task world. With markers available in both regimes, masking improved accuracy on held-out two- and three-operation compositions by median paired differences of 0.846 and 0.859; all twelve pairs cleared the required margins, and the full preregistered behavioral criterion passed. The unmarked replication also passed. No globally visible system passed the marker-following check, so the effect of usable role information remains unresolved. The filler condition yielded seven full generalizers, but its decomposition criteria were inconclusive. Packet interventions in all eighteen audited masked systems followed the predicted intermediate-value changes on eligible cases; these finite, success-conditioned audits do not establish mediation. The results confirm a large advantage of the tested masking regime, while leaving its finer attribution and generality open. Protocols, results, and checkpoints are public.

中文摘要

摘要:限制模块可以读取的内容可能会提高系统学习计算的能力。我们在一项预注册确认实验中测试了这一点,该实验使用六十四个单元的系统,这些系统共享冻结的语言模型骨干,并通过学习到的连续数据包进行通信。五种情况分别改变了证据屏蔽、所有权标记以及用中性填充物替换外来证据,在六个初始化集群中进行,每个集群有两种数据顺序,在一个新的任务世界中进行测试。当标记在两种机制中都可用时,屏蔽提高了对保留的二操作和三操作组合的准确率,中位成对差异分别为0.846和0.859;所有十二对都超过了所需的界限,并且完整的预注册行为标准通过了。未标记的重复实验也通过了。没有全球可见的系统通过标记遵循检查,因此可用角色信息的作用仍未解决。填充物条件产生了七个完全泛化器,但其分解标准尚不明确。在所有十八个经过审计的屏蔽系统中,数据包干预都在符合条件的实例上遵循预测的中间值变化;这些有限的、受成功条件限制的审计并未建立中介作用。结果确认了所测试的屏蔽机制具有显著优势,同时其更细致的归因和普遍性仍未确定。协议、结果和检查点均已公开。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17637 (HTTP 429)

Authors: Narcis Marincat

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.17637.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17637

Published: 2026-09-18T01:28:53.527Z


6. CapMem: A Benchmark for Caption-Based Episodic Memory in Egocentric Video

Abstract:Wearable assistants require episodic memory over egocentric video, yet current vision-language models face bounded frame budgets, growing visual-token costs, and long-context retrieval failures. Under these practical constraints, we study whether textual captions can serve as reusable episodic memory. We define the Episodic Memory Video Caption QA task and introduce CapMem, a human-annotated benchmark with 75 videos totaling 33.7 hours, and 1,000 multiple-choice questions across 16 scenarios. On long videos (>20 min), full-coverage CaptionQA with 30s and 60s caption windows outperforms direct VideoQA for 10/12 and 8/12 models, respectively. On the same video subset, a matched-frame control across six Qwen models retains mean accuracy gains of 3.22 and 2.55 points, respectively. Our caption-guided retrieve-and-verify harness further improves accuracy by up to 5.3 points. These results support the effectiveness of caption memory for episodic reasoning over long egocentric video.

中文摘要

摘要:可穿戴助手在以自我为中心的视频上需要情节记忆,但现有的视觉-语言模型面临帧数限制、不断增加的视觉代币成本以及长上下文检索失败的问题。在这些实际约束条件下,我们研究文本字幕是否可以作为可重用的情节记忆。我们定义了情节记忆视频字幕问答(Episodic Memory Video Caption QA)任务,并介绍了 CapMem,这是一个人类标注的基准数据集,包含 75 个视频,总时长 33.7 小时,以及覆盖 16 种场景的 1,000 道多项选择题。在长视频(>20 分钟)上,使用 30 秒和 60 秒字幕窗口的全覆盖 CaptionQA 在 12 个模型中分别优于直接 VideoQA 的模型数量为 10/12 和 8/12。在相同的视频子集上,六个 Qwen 模型的匹配帧控制分别保持平均准确率提升 3.22 和 2.55 个百分点。我们的字幕引导检索和验证机制进一步将准确率提高最多 5.3 个百分点。这些结果支持了字幕记忆在长自我中心视频上的情节推理有效性。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17688 (HTTP 429)

Authors: Dingli Liang, Yiqiao Xie, Yukai Huang, Zhaokai Wang, Weitong Cai, Guangwen Feng, Jifei Song, Zhensong Zhang, Hang Zhang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.17688.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17688

Published: 2026-09-18T01:28:53.527Z


7. GraphEcho: Structural Redundancy and Evidence Provenance in LLM Graph Agents

Abstract:A large language model (LLM) agent can follow more graph paths without acquiring more independent evidence. GraphEcho tests whether agents mistake these repeated encounters for additional corroboration. The benchmark varies path counts and evidential origins while holding evidence content fixed, and evaluates both judgments and active exploration. Controlled synthetic experiments reveal model-dependent judgment shifts, but redundant supporting paths increase the share of repeated walks across all evaluated frozen agents. Provenance-aware post-training (PAPT) reduces revisits and improves synthetic accuracy, yet covers fewer distinct sources. On scientific claims, it continues to reduce repetition while accuracy declines. These findings expose a gap between efficient exploration and effective evidence use: an agent can learn to stop repeating itself while overlooking information it needs. GraphEcho provides a controlled way to evaluate both what graph agents conclude and whether their exploration reaches distinct evidential sources.

中文摘要

摘要:一个大型语言模型(LLM)代理可以在不获取更多独立证据的情况下,沿更多图路径进行推演。GraphEcho 测试代理是否会将这些重复的遭遇误认为是额外的证据支持。该基准在保持证据内容不变的情况下,变化路径数量和证据来源,并评估判断和主动探索两方面。受控的合成实验显示了模型依赖的判断变化,但冗余的支持路径增加了所有被评估冻结代理重复遍历的比例。源感知的后训练(PAPT)减少了重复访问并提高了合成准确性,但覆盖的不同来源较少。在科学声明上,它仍然减少重复,但准确性下降。这些发现揭示了高效探索与有效利用证据之间的差距:代理可以学会停止重复自身,同时忽略所需信息。GraphEcho 提供了一种受控方法来评估图代理的结论以及其探索是否达到了不同的证据来源。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17695 (HTTP 429)

Authors: Sikun Wang, Yixi Zhou, Lei Fan, Fan Zhang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.17695.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17695

Published: 2026-09-18T01:28:53.527Z


8. GVD: Governed Versioning and Deduplication for Document Repositories

Abstract:Document repositories evolve continuously. Guidelines and policies are revised, superseded, and re-uploaded, so the same content recurs in different wording and newer versions refine or contradict earlier ones. These inconsistencies belong to the growing collection rather than to any single document, yet existing work treats versioning, duplicate detection, and contradiction detection as isolated pairwise tasks and stops once a pair is labeled. We present GVD (Governed Versioning and Deduplication), a framework that unifies cross-document version linking with rule-level conflict resolution under an auditable update policy. Incoming documents are assigned to version families through bidirectional rule alignment, and their rules are compared against the family memory to identify duplicates, contradictions, asymmetric refinements, and new knowledge, with Counterfactual Span Probing (CSP) resolving related pairs that inference misclassifies as neutral. Relation-specific policies suppress duplicates and escalate only consequential changes for review, retaining version lineage as an audit trail. The pipeline runs fully locally, with no large language model. On 120 enterprise documents processed as 140 ingestions across 59 version families, GVD reaches an F1 of 0.97 for version-family construction and 0.94 for rule-level consistency, with CSP raising rule consistency from 0.90 to 0.94.

中文摘要

摘要:文档库在不断演变。指南和政策会被修改、替换并重新上传,因此相同的内容会以不同的措辞重复出现,新版本会完善或与早期版本产生矛盾。这些不一致属于不断增长的文档集合,而不是某个单一文档。然而,现有的工作将版本管理、重复检测和矛盾检测视为孤立的成对任务,并在标注一对文档后就停止。我们提出了GVD(Governed Versioning and Deduplication,受控版本管理与重复清理)框架,该框架在可审计的更新策略下,将跨文档的版本关联与规则级冲突解决统一起来。新进文档通过双向规则对齐被分配到版本族中,并将其规则与版本族记忆进行比较,以识别重复、矛盾、不对称细化以及新知识,其中反事实跨度探查(CSP)解决推理误判为中性相关对的问题。针对关系的策略抑制重复,仅将重要的变更升级以供审查,同时保留版本沿革作为审计轨迹。该流程完全在本地运行,不依赖大型语言模型。在处理59个版本族中的120份企业文档,共140次导入的实验中,GVD在版本族构建上的F1达到0.97,在规则级一致性方面达到0.94,其中CSP将规则一致性从0.90提升至0.94。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17696 (HTTP 429)

Authors: Mohammadreza Sediqin, Shivali Dalmia, Sumukha Thoppanahalli, Abhishek Mukherji

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.17696.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17696

Published: 2026-09-18T01:28:53.527Z


9. NeMo Data Designer: An Extensible Framework for Multimodal Synthetic Data Generation

Abstract:We present NeMo Data Designer (NDD), an open-source, general-purpose framework for multi-modal synthetic data generation (SDG). Designed to be intuitive to use, NDD provides a declarative configuration format in which human and/or agent users define each dataset column, with column types spanning text, code, structured outputs, images, embeddings, and statistical samplers that are explicitly configured to steer dataset diversity. Additional column types and functionality can be introduced using the framework’s flexible plugin system. NDD’s configuration is an inspectable artifact, supporting workflow sharing and reproducibility. SDG is an inherently iterative process. NDD therefore builds a preview-and-revision loop into its core workflow, allowing users to generate and inspect a small number of records, refine the specification, and rerun generation at full scale. At runtime, NDD resolves dependencies, schedules calls to user-provided model endpoints, and retries failed requests. We describe NDD’s architecture and programming model and present case studies spanning structured, agentic, multimodal, and domain-specialized tasks, including datasets used in Nemotron model development and in production enterprise deployments.

中文摘要

摘要:我们提出了 NeMo 数据设计器(NeMo Data Designer,NDD),这是一个开源的、通用的多模态合成数据生成(SDG)框架。NDD 旨在易于使用,提供了一种声明式配置格式,用户(人类和/或代理)可以在其中定义每个数据集列,列类型涵盖文本、代码、结构化输出、图像、嵌入以及可显式配置以引导数据集多样性的统计采样器。通过框架灵活的插件系统,还可以引入其他列类型和功能。NDD 的配置本身是可检查的工件,支持工作流共享和可重现性。SDG 本质上是一个迭代过程。因此,NDD 在其核心工作流中构建了预览与修订循环,允许用户生成并查看少量记录,完善规范,并重新运行全量生成。在运行时,NDD 会解析依赖关系、调度对用户提供的模型端点的调用,并重试失败的请求。我们描述了 NDD 的架构和编程模型,并展示了涵盖结构化、代理、多模态及领域专用任务的案例研究,包括用于 Nemotron 模型开发和生产企业部署的数据集。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17699 (HTTP 429)

Authors: Johnny Greco, Nabin Mulepati, Andre Manoel, Eric Tramel, Kirit Thadaka, Mike Knepper, Dhruv Nathawani, Dane Corneil, Yev Meyer, Alex Watson, Maarten Van Segbroeck

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.17699.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17699

Published: 2026-09-18T01:28:53.527Z


10. A Systematic Evaluation of the COTQ Provincial Land Cover Product: Structural Consistency, Spectral Separability, and Relative Positioning Against ESA, ESRI, and Google Products

Abstract:High-resolution land use and land cover (LULC) products derived from Sentinel-2 imagery are widely used for environmental monitoring and land management, yet their performance can vary across regions with complex ecological gradients and heterogeneous surface conditions. In Quebec, these limitations motivated the development of a provincial 10-m land-cover product, the COTQ, designed to support annual monitoring of land occupation and soil artificialisation. This study presents a systematic evaluation of the COTQ product relative to three global 10-m LULC datasets: ESA WorldCover, ESRI LandCover, and Google DynamicWorld. This paper does not introduce a new mapping methodology but focuses on analysing the behaviour and consistency of the COTQ using complementary evaluation approaches. All products are harmonized under a common legend and compared using structural indicators (object-size distributions, shape complexity, Adjusted Rand Index, and Intersection over Union), spectral separability metrics derived from Sentinel-2 reflectance data, and a targeted photo-interpretation of disagreement areas. The analysis is conducted over eight Sentinel-2 tiles selected to represent the main bioclimatic domains of Québec, from temperate and boreal forests to northern tundra environments. The results show that the COTQ exhibits structural and spectral characteristics most similar to ESA WorldCover among the reference global products, while revealing systematic differences linked to class definitions and thematic priorities, particularly for urban areas, wetlands, and rocky or cryptogamic surfaces. This multi-criteria evaluation provides an objective characterization of the COTQ product and clarifies its relative positioning with respect to existing global land-cover datasets for operational land monitoring in Québec.

中文摘要

摘要:基于Sentinel-2影像的高分辨率土地利用和土地覆盖(LULC)产品被广泛用于环境监测和土地管理,但其性能在生态梯度复杂、地表条件异质的地区可能存在差异。在魁北克,这些局限性促使开发了省级10米土地覆盖产品COTQ,旨在支持土地占用和土壤人工化的年度监测。本研究系统评估了COTQ产品相对于三种全球10米LULC数据集的表现:ESA WorldCover、ESRI LandCover和Google DynamicWorld。本文未引入新的制图方法,而是重点分析COTQ在不同评估方法下的表现和一致性。所有产品均在统一的图例下进行统一,并通过结构指标(目标大小分布、形状复杂性、调整兰德指数和交并比)、从Sentinel-2反射率数据衍生的光谱可分性指标,以及针对不一致区域的定向影像解译进行比较。分析覆盖了八个选定的Sentinel-2切片,以代表魁北克的主要生物气候区,从温带和北方针叶林到北部苔原环境。结果显示,在参考的全球产品中,COTQ在结构和光谱特征上最接近ESA WorldCover,同时也揭示了与类别定义和专题重点相关的系统性差异,尤其是在城市地区、湿地以及岩石或隐花植物表面。本多标准评估提供了COTQ产品的客观特征描述,并阐明了其相对于现有全球土地覆盖数据集的相对定位,为魁北克运营性土地监测提供参考。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17731 (HTTP 429)

Authors: Étienne Clabaut, Samuel Foucher, Yacine Bouroubi

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.17731.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17731

Published: 2026-09-18T01:28:53.527Z


Evaluation Domain Papers

1. Making AI-Assisted Claims Independently Challengeable: Publication Authority and a Protocol for Falsifiable Publication Records

Abstract:AI-assisted claims can appear authoritative when evidence, analysis, human authorization, presentation, and correction history refer to different states. Provenance, attestation, and transparency expose history but alone do not specify the publication transition examined here. We develop Publication Authority as an exact-state, non-transferable, single-use publication capability and instantiate it in PAC-2026 (Publication-Accountability Calculus), a machine-readable AIJIM Protocol candidate. We evaluate its fourth bounded semantic freeze (SF-4), a fixed-profile specification designed for replaceable bindings. Six obligations govern evidence, runs and artifacts, measurement disclosure, authorization, surface correspondence, and lifecycle continuity. Each yields a target-bound witness, localized counterexample, or localized unverifiability; none can compensate for another. Only a fresh, complete all-pass record derives the permit consumed by one atomic publication transition. We use identity vectors, adversarial cases, finite models, and historical implementations. Ten models explored 110,764 safe reachable states; 76 unsafe configurations produced the expected violation or observer countermodel. A reader surface passing its correspondence check cannot authorize publication unless the accepted record admits that surface. SF-4 separates evidence horizon from verification time and rejects an authentic but causally invalid authorization. A historical predecessor path reproduced 17 frozen authorization-successor outcomes. A later in-house, instance-blind test of known case classes matched all 183 scored expectations; same-host package execution reproduced its 240 archived observations. Results support internal coherence, bounded safety, fault sensitivity, and limited constructibility, but not factual truth, general refinement, blind interoperability, field efficacy, or standards status.

中文摘要

摘要:当证据、分析、人工授权、呈现和修正历史涉及不同状态时,人工智能辅助的声明可能看起来具有权威性。来源、证明和透明度揭示了历史,但单独不足以指定此处所研究的出版转换。我们提出了出版权威(Publication Authority),作为一种精确状态、不可转让、一次性使用的出版能力,并在 PAC-2026(出版责任演算,Publication-Accountability Calculus)中实现,它是机器可读的 AIJIM 协议候选方案。我们评估了其第四个有界语义冻结(SF-4),这是一种为可替换绑定设计的固定配置规范。六项义务管理证据、运行和工件、测量披露、授权、表面对应和生命周期连续性。每一项都产生目标绑定的见证、本地化反例或本地化不可验证性;没有一项可以替代另一项。只有一个新的、完整的全通过记录,才能派生出由一次原子出版转换消耗的许可。我们使用身份向量、对抗性案例、有限模型和历史实现。十个模型探索了 110,764 个安全可达状态;76 个不安全配置产生了预期的违规或观察者反模型。通过其对应检验的读者表面无法授权出版,除非被接受的记录承认该表面。SF-4 将证据视域与验证时间分离,并拒绝真实但因果上无效的授权。历史前置路径重现了 17 个冻结的授权-后继结果。稍后的内部、实例盲测试对已知案例类别匹配了全部 183 个评分期望值;同一主机软件包执行重现了其 240 个归档观察结果。结果支持内部一致性、有界安全性、故障敏感性和有限可构造性,但不支持事实真实性、一般完善性、盲互操作性、现场有效性或标准状态。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17631 (HTTP 429)

Authors: Torsten Olivi Tiltack, Yifei Dong, Kun Yu, Xu Wang, Wei Liu, Jianlong Zhou, Ren Ping Liu, Fang Chen

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.17631.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17631

Published: 2026-09-18T01:29:19.418Z


2. EvolveTrade: Experience-Driven Policy Refinement for Self-Evolving LLM Trading Agents

Abstract:Large language model (LLM) trading agents can combine market data, news, and executable analysis, but their behavior is often controlled by static hand-written tool-use policies that are fixed before deployment. This limits their ability to adapt how they gather evidence, invoke tools, verify signals, and manage risk under changing market regimes. We introduce EvolveTrade, a self-evolving framework that treats the system prompt of a tool-using trading agent as a text-parameterized policy. After each update interval, a Policy Agent revises this policy using accumulated decision traces and realized portfolio feedback, while keeping the backbone LLM fixed. The updated policy is then used for the next batch of trading decisions, enabling the agent to refine its information-acquisition and portfolio-construction procedure over time. Experiments across multiple market regimes and two LLM backbones show that EvolveTrade often improves Sharpe Ratio and Cumulative Return over fixed-policy LLM baselines, achieving the improved SR and CR in most evaluated settings. Behavioral analyses further show that self-evolved policies increase code-mediated analysis and activate regime-relevant computations; case-level policy-to-return attributions trace how policy-induced allocation changes contribute to realized return differences. These results suggest that adapting the reusable procedure governing tool use is a key direction for building more robust LLM trading agents.

中文摘要

摘要:大型语言模型(LLM)交易代理可以结合市场数据、新闻和可执行分析,但它们的行为通常由在部署前固定的静态手写工具使用策略所控制。这限制了它们在不断变化的市场环境下调整证据收集、调用工具、验证信号和管理风险的能力。我们提出了EvolveTrade,这是一种自我进化框架,将使用工具的交易代理的系统提示视为文本参数化策略。在每个更新间隔之后,策略代理使用累积的决策轨迹和实现的投资组合反馈来修改该策略,同时保持核心LLM固定。更新后的策略随后用于下一批交易决策,使代理能够随着时间的推移优化其信息获取和投资组合构建过程。在多个市场环境和两个LLM骨干模型上的实验表明,EvolveTrade通常在夏普比率和累计收益上优于固定策略LLM基线,在大多数评估设置中实现了改善的SR和CR。行为分析进一步显示,自我进化策略增加了基于代码的分析,并激活了与市场环境相关的计算;个案级策略到收益的归因跟踪了策略引起的配置变化如何对实现的收益差异产生贡献。这些结果表明,调整支配工具使用的可重复程序是构建更稳健LLM交易代理的关键方向。

LLM Analysis

Q: 这篇论文试图解决什么问题? 该论文旨在解决\*\*大型语言模型(LLM)交易代理中工具使用策略静态固化\*\*的核心问题。具体而言,现有LLM交易代理虽然具备整合市场数据、新闻与可执行分析的能力,但其决策程序(即“策略”)通常在部署前由人工预定义并以固定系统提示的形式冻结,无法根据实际交易经验与市场环境的变化进行自我调整。 该问题可进一步分解为以下几个层面: - \*\*静态策略对非平稳市场的适应性不足\*\* 金融市场具有显著的\*\*非平稳性\*\*(non-stationarity),不同市场机制(如震荡、急跌反弹、稳步上行)要求差异化的信息获取、信号验证与风险管理方式。然而,静态策略迫使代理在截然不同的市场环境中重复使用相同的分析范式与工具调用模式,导致其在变化的市场条件下难以维持稳健表现。 - \*\*经验反馈无法沉淀为可复用的决策程序\*\* 现有方法即使允许代理调用外部工具(如价格检索、新闻搜索、代码解释器),其“何时、为何、如何”调用这些资源的操作规程在部署后保持不变。代理每日产生的大量决策轨迹(包括资产级推理与投资组合反馈)未能被系统性利用来修订其可复用的工具使用流程。 - \*\*分析词汇与风险控制的僵化\*\* 如论文所示,固定策略导致代理被锁定在一套固定的分析指标中(例如始终计算20日收益、年化波动率等),而极少激活与当前市场机制真正相关的计算(如下跌行情中的风险价值(VaR)与信号归一化,或上行趋势中的指数移动平均(EMA)与相对强弱指标(RSI))。代理缺乏机制去扩展或调整其分析工具箱。 为应对上述挑战,论文提出 \*\*EvolveTrade\*\* 框架,其核心思想是: > 将交易代理的系统提示视为一个\*\*文本参数化的策略\*\* π_t ,并引入一个独立的策略代理(Policy Agent),在每个交易批次结束后,基于累积的决策轨迹 h_t 与实现的投资组合反馈 g_t 在线修订该策略: >


π(kN+1) = f(update)(π_(kN), B_k)

其中 B_k 为第 k 个批次的历史记录集合。通过这种方式,代理能够在保持底层LLM参数与工具接口完全冻结的前提下,使其信息获取、证据验证与投资组合构建程序随经验持续演化。 Q2: 有哪些相关研究? 根据论文第2节,相关研究主要分布于以下两个方向: ### 1. LLM 金融交易代理(LLM Agents for Financial Trading) 该方向探索如何为 LLM 配备记忆、工具、多模态信号与多智能体协作能力,以支持金融决策。 - **记忆与角色增强系统** FinMem(Yu et al., 2025b)引入分层记忆以跨会话保留市场观察;TradingAgents、StockAgent、QuantAgent 与 TradExpert(Xiao et al., 2025; Zhang et al., 2024a; Xiong et al., 2025a; Ding et al., 2024)则通过专门化的代理角色、模拟交易环境或固定分析管道来组织决策。 - **多模态与实时数据平台** FinAgent 及相关金融代理平台(Zhang et al., 2024b; Han et al., 2025; Yang et al., 2024b)整合多模态输入、实时数据、领域工具与反思式决策机制。 - **基准评测** AI-Trader、LiveTradeBench 与 FinAgentBench(Fan et al., 2025; Yu et al., 2025a; Choi et al., 2025)在更贴近现实的金融信息与市场条件下评估 LLM 代理。 - **可靠性研究** 近期工作开始关注 LLM 交易中的可靠性问题,包括噪声源信任、虚假股票代码记忆与时间泄漏(Li et al., 2026; Jeon and Lee, 2026; Benhenda, 2026)。 **与 EvolveTrade 的区别**:上述系统虽然证明了 LLM 可整合异构金融证据,但其观察通道、提示模板、角色结构或工具工作流通常在部署前即被固定, governing 信息获取、证据验证与风险控制的**操作规程**基本保持静态。 ### 2. 自我进化 LLM 代理(Self-evolving LLM Agents) 该方向研究 LLM 代理如何基于自身的执行轨迹、反馈或任务结果进行部署后改进。 - **工具使用基础** ReAct(Yao et al., 2023)与 Toolformer(Schick et al., 2023)等早期工作实现了推理与外部动作的交错,并探索了 API 调用时机。 - **提示与管道优化** 一类通用方法通过验证信号、类梯度反馈或进化搜索来修订指令或 LM 程序(Pryzant et al., 2023; Yang et al., 2024a; Khattab et al., 2024; Yuksekgonul et al., 2025; Agrawal et al., 2026; Choi et al., 2026; Zhang et al., 2025)。 - **代理级自我进化** 更进一步的代理级方法将成功与失败转化为反思记忆、可复用工作流或推理策略(Shinn et al., 2023; Wang et al., 2024, 2025; Ouyang et al., 2026; Pan et al., 2026)。 - **交易领域的策略进化** ATLAS 与 SHARP(Papadakis et al., 2026; Chen et al., 2026)尝试从市场反馈中调整提示或结构化策略;AlphaQuanter 与 FLAG-Trader(Deng et al., 2026; Xiong et al., 2025b)则表明学习的工具编排或策略优化可改善固定多代理基线。 **与 EvolveTrade 的区别**:上述方法或优化推理指令,或维护经验记忆,或离线学习工具使用策略,而非**持续地在线精炼**已部署代理在信息收集、验证与行动过程中使用的**操作规程**。EvolveTrade 的独特之处在于将系统提示视为**文本参数化的工具使用策略** π_t ,并基于代理自身的工具调用轨迹、交易决策与投资组合反馈进行在线更新:

π(kN+1) = f(update)(π_(kN), B_k)
Q3: 论文如何解决这个问题? 论文通过提出 **EvolveTrade** 框架,将 LLM 交易代理的系统提示视为一个**文本参数化的可进化策略**,并设计了一套基于实现交易经验的在线策略精炼机制。具体解决方案包含以下层面: ### 1. 将系统提示形式化为时变策略 论文首先将交易代理的决策环境形式化。在第 t 个交易日开始时,代理的状态为:

St = (π_t, t, P(t-1), T)
其中 πt 是当日生效的工具使用策略(以自然语言系统提示实现), P(t-1) 为先前投资组合状态, T 为可用工具集(价格检索、新闻搜索、Python 代码解释器)。与静态基线不同,此处 π_t 被显式建模为随时间变化的变量,允许在不同交易日持有不同的策略文本。 ### 2. 构建可审计的策略反馈记录 为消除金融市场中延迟且嘈杂的反馈信号带来的歧义,论文将决策轨迹与事后反馈配对,形成精炼记录:

h_t = (w_t, d_t^(dec))

R_t = (h_t, g_t)

  • h_t 包含代理提交的资产配置 w_t 及每支资产的推理依据 d_t^(dec) (即代理为设定该权重所使用的证据与论证) - g_t 为事后反馈,记录当日投资组合收益、各资产收益及仓位变化 该记录 R_t 使得策略修订能够基于**代理具体做了什么**以及**市场随后如何回应**进行归因,而非仅依赖孤立的盈亏数字。 ### 3. 在线策略自我进化 EvolveTrade 以固定的 N 个交易日为批次进行策略更新。设第 k 个批次覆盖交易日 (k-1)N+1, dots, kN ,则该批次内的策略保持恒定:

π((k-1)N+1) = ·s = π(kN)
批次结束时,系统收集该批次的全部记录:
Bk = R((k-1)N+1), dots, R(kN)
随后,一个独立的 **Policy Agent**(由相同或另一 LLM 实现)执行基于语言的精炼步骤:
π
(kN+1) = f(update)(π(kN), B_k)

f(update)(π, B) = LLM(I(update), π, B)
其中 I(update) 为固定的更新指令,要求 Policy Agent: - 分析 B_k 中的记录,识别当前提示中哪些具体条款导致了良好或糟糕的表现 - 将观察到的行为与提示中的具体行项关联,执行**有依据的编辑**(grounded edits),而非泛泛重写 - 输出完整的更新后策略文本 ### 4. 策略内容的可进化维度 策略精炼不限于高层交易逻辑,而是涵盖工具使用的完整操作规程。一次精炼可能改变: - **工具选择与优先级**:例如增加代码解释器调用频率,减少重复性新闻检索 - **查询 formulation**:如调整价格回溯窗口长度或新闻检索关键词 - **信号解释与交叉验证**:例如引入”新闻信号需经价格确认后方可增仓”的规则 - **风险控制与仓位调整**:如在不同市场机制下激活风险价值(VaR)、信号归一化或趋势跟踪指标(EMA、RSI、SMA) - **输出结构**:要求代理在最终输出中包含可审计的目标权重与再平衡动作 ### 5. 保持基础能力冻结以实现模块化进化 在整个过程中,底层 LLM 参数 pθ 与工具接口 T 始终保持冻结。唯一被学习的对象是**策略文本本身**。这种设计使得框架可应用于任何已部署的 LLM 交易代理,无需重新训练模型或修改工具实现,从而将”市场经验的沉淀”与”基础推理能力”解耦。 通过上述机制,EvolveTrade 使交易代理能够在不同市场机制(震荡、急跌反弹、上行趋势)中逐步激活此前被静态策略抑制的分析工具与风险控制程序,实现从固定分析词汇到**机制相关计算**的自适应扩展。 Q4: 论文做了哪些实验? 论文在第6节及附录中设计并执行了多组实验,涵盖不同市场机制、模型主干与消融条件,具体可分为以下几个层面: ## 1. 实验设置与评测环境 - **交易环境**:基于 LiveTradeBench 构建的日收盘回测模拟。每个交易日收盘后,代理提交目标权重,系统按收盘价再平衡组合。 - **投资范围**:15只美股蓝筹股(如 AAPL、MSFT、NVDA、JPM 等)及现金。 - **评测窗口**:每月为一个在线评测周期,代理每日观察市场、提交配置、接收组合反馈,并按固定间隔更新策略。 - **模型主干**:分别使用 **GPT-5-mini** 与 **Gemini-2.5-Flash** 作为交易代理与策略代理的底层模型。 - **策略更新间隔**:默认每 N=5 个交易日执行一次策略精炼。 - **市场机制**:为检验跨机制泛化性,实验覆盖了六种不同的后知识截断市场机制: - GPT-5-mini:2025年1月(震荡)、2025年4月(急跌反弹)、2025年9月(上行趋势) - Gemini-2.5-Flash:2025年11月(看跌)、2026年2月(震荡)、2026年4月(看涨) ## 2. 基线方法 实验对比了规则基线与多组 LLM 基线,以隔离工具访问与策略进化的效应: **规则基线** - SPY、Buy and Hold (B&H)、MACD、KDJ&RSI、ZMR、SMA **LLM 基线** - **Static Base Agent**:固定观察窗口(价格+新闻),无工具调用。 - **Static Tool-Calling (TC) Agent**:具备完整工具集(价格、新闻、代码解释器),但策略文本固定。 - **EvolveBase**:无工具,但允许策略自我进化,以隔离“纯策略进化”在无工具场景下的效应。 - **EvolveStrategy**:策略代理仅更新高层交易逻辑(如信号组合、仓位调整规则),工具调用协议与操作细节保持固定。 - **EvolveTrade (Ours)**:策略代理全面更新交易代理的完整策略,包括工具调用协议与操作规程。 ## 3. 评测指标 - 年化夏普比率(SR↑) - 累计收益率(CR%↑) - 最大回撤(MDD%↓) - 胜率(WR%↑) - 日波动率(Vol%↓) ## 4. 主实验结果 表1报告了各代理在六种市场机制下的平均表现(LLM 方法均运行3次)。核心发现包括: - **EvolveTrade 在多数设置中取得最佳风险收益表现**。对于 GPT-5-mini,其在2025年1月与9月实现 LLM 方法中最高的 SR 与 CR;对于 Gemini-2.5-Flash,其在2025年11月取得最佳,在2026年2月取得次佳。 - **策略进化带来的收益不仅限于相对其他 LLM 基线**。例如,GPT-5-mini 在2025年9月达到 CR 6.84%,超过该期所有规则基线;Gemini-2.5-Flash 在2026年2月达到 SR 2.75 与 CR 2.92%,亦持续优于全部规则基线。 ## 5. 行为与机制分析 ### 5.1 分析指标激活模式 图3对比了 Static TC Agent 与 EvolveTrade 在代码解释器中实际调用的分析指标。结果显示: - Static TC Agent 在所有机制中均锁定于同一套狭窄指标(如 20日收益、年化波动率、夏普比率)。 - EvolveTrade 在策略进化后激活了基线从未调用的机制相关指标:如在2025年4月急跌中启用 **VaR** 与**信号归一化**;在2025年9月上行趋势中启用 **EMA、RSI、SMA** 等趋势跟踪指标。 ### 5.2 工具调用频率变化 图4量化了每日平均工具调用次数: - 价格检索工具( t(price) )在两组代理中均维持约1次/日。 - **代码解释器( t(code) )调用显著上升**:从 Static TC Agent 的约 1.0 次/日提升至 EvolveTrade 的 2.6–4.5 次/日。 - 新闻搜索工具( t_(news) )未出现显著增长。 这表明策略进化将代理行为从外部文本搜集**重定向为可执行的数量化分析**。 ### 5.3 案例级归因分析 图5给出了2025年1月24日 NVDA 急跌前的持仓对比: - Static TC Agent 持有 NVDA 10.7%,依据短期收益与正面新闻。 - EvolveTrade 经策略精炼后仅持有 2.9%,因其策略优先要求更长周期的风险调整证据,并对高波动仓位执行组合改进与换手率检查。 - 次日 NVDA 下跌 17.0%,EvolveTrade 当日回报为 -0.03%,而 Static TC Agent 为 -1.11%。 附录E进一步提供了2025年9月10日的单日归因(图7),显示 EvolveTrade 通过将代码输出转化为可审计的目标权重,在 JPM、CAT、WMT、TSLA 上获得了次日超额收益。 ### 5.4 长期演化稳定性 为验证持续更新是否导致策略文本膨胀与性能衰减,论文在 GPT-5-mini 上执行了 **50个交易日** 的长期实验(表2、图6): - EvolveTrade 在 SR 与 CR 上仍优于所有基线。 - 策略文本长度呈现**非单调变化**:随新指导的加入而增长,但周期性地收缩,表现为冗余指令的整合而非无限累积。 ### 5.5 更新间隔敏感度 表3考察了策略更新频率( N = 1, 3, 5, 7 天)对 GPT-5-mini 平均表现的影响: - N=5 天取得最高的平均 SR 与 CR。 - N=1 天(每日更新)表现最差,提示过度频繁的修订可能过拟合短期噪声。 - 该实验表明策略进化需要在**适应性**与**反馈累积稳定性**之间取得平衡。 ## 6. 交易成本与稳健性检验 附录D(表6)在 10 bps 的比例交易成本下重新计算了 SR 与 CR,并报告了组合换手率(TO): - EvolveTrade 在考虑摩擦后仍保持其在1月、9月(GPT-5-mini)及11月(Gemini-2.5-Flash)的领先优势。 - 其换手率在所有6个评测窗口中均低于 Static TC Agent 与 EvolveStrategy,说明收益提升并非源于激进再平衡。 附录C(表4、表5)报告了3次独立运行的均值与标准差,显示 EvolveTrade 的增益具有跨运行一致性,而非单次幸运结果。 ## 7. 失效模式分析 附录F(表7)分析了 EvolveTrade 在两个4月窗口中表现弱于最强静态基线的原因: - 在这两个时期,EvolveTrade 的平均现金仓位分别比基线高出 10.1 与 35.9 个百分点。 - 过高的现金储备使其在2025年4月反弹与2026年4月牛市中错失了股权收益,揭示了策略进化在特定机制下可能产生过度防御性配置。 Q5: 有什么可以进一步探索的点? 基于论文的局限性与实验洞察,以下方向值得进一步探索: ### 1. 自适应策略更新调度 当前实现采用固定的批次长度 N (如每 5 个交易日更新一次)。表 3 的敏感度分析表明,过于频繁的更新( N=1 )会因过拟合短期噪声而损害表现,而较长的间隔又可能延迟对机制突变的响应。未来可探索**基于市场条件触发的动态更新策略**,例如根据已实现波动率、组合回撤幅度或策略文本的预测不确定性来自适应调整更新频率:

Nt = f(σ(t)^(realized), MDD_t, H_t)

2. 更真实的执行摩擦建模 论文在附录 D 中仅以 10 bps 的比例交易成本近似真实摩擦,未纳入**滑点(slippage)、市场冲击(market impact)与流动性约束**。对于日度再平衡且持仓可能集中于少数资产的 LLM 代理,这些摩擦在回测与实盘之间的差异尤为显著。后续研究可整合限价单簿(LOB)模拟或成交量加权平均价格(VWAP)执行模型,检验策略进化是否仍能提升净夏普比率。 ### 3. 策略文本的压缩与长期记忆机制 图 6 显示策略长度在 50 个交易日内呈非单调增长,但长期运行下仍存在文本膨胀风险。未来可引入**显式的策略摘要与遗忘机制**:在每次精炼时,要求 Policy Agent 不仅追加新规则,还需识别并删除冗余或已失效的指令,将策略长度约束为与当前市场机制相关的“工作集”,从而避免上下文窗口溢出与指令冲突。 ### 4. 跨机制迁移与元学习 实验表明,EvolveTrade 在不同市场机制(震荡、急跌、上行)中激活了不同的分析指标集合。一个开放问题是:进化后的策略是否包含**可迁移的元策略**?例如,在 2025 年 4 月急跌期间习得的尾部风险控制规则,能否在 2026 年 2 月的另一场下行中被快速唤起,而非重新从零进化?引入基于机制识别的策略路由或元学习初始化,可能显著加速新环境下的适应。 ### 5. 反事实归因与噪声鲁棒性 金融市场反馈具有高度随机性,单日盈亏可能掩盖策略的真实质量。当前框架直接利用实现回报 gt 作为反馈信号,未来可探索**反事实(counterfactual)评估**:在策略更新前,要求 Policy Agent 评估“若昨日采取另一套规则,组合表现将如何变化”,或构建基于合成控制方法的因果归因,以降低将运气误读为能力(spurious policy improvement)的风险。 ### 6. 安全护栏与价值对齐 附录 F 揭示了策略进化可能产生过度防御性配置(如在 2026 年 4 月牛市中持有 40.1% 现金,错失上涨收益)。这提示需要**硬约束与软偏好的分层治理**:在策略文本之外,维护一组不可由进化过程覆盖的安全规则(如最大现金上限、单一资产集中度下限),或通过基于人类偏好的强化学习(RLHF)对 Policy Agent 的修订方向进行对齐,防止自我进化走向过度保守或过度激进。 ### 7. 工具空间的动态扩展 当前工具集 T = t(price), t(news), t(code) 是预定义且固定的。未来可研究**工具发现(tool discovery)**:允许策略代理在精炼过程中提出新的 API 调用需求(如接入期权隐含波动率、供应链图数据库或另类数据),并由外部系统评估其可行性与增量价值,从而突破现有分析词汇的天花板。 ### 8. 可解释性辅助的人机协同 EvolveTrade 生成的策略文本具有高度可读性(如图 12–16 所示)。未来可设计**交互式审查界面**,使人类风控专家能够在每个批次结束时审阅 Policy Agent 提出的策略变更,对“激活 VaR 计算”或“提高现金仓位”等关键修改进行批准、驳回或修正,将 LLM 的自我进化能力与人类领域知识相结合,形成混合增强型交易系统。 Q6: 总结一下论文的主要内容 该论文针对大型语言模型(LLM)金融交易代理中**工具使用策略静态固化**的核心瓶颈,提出了一种在线自我进化框架 **EvolveTrade**,并通过多市场机制、多模型主干的实验验证其有效性。 ### 1. 研究背景与动机 现有 LLM 交易代理虽能整合市场数据、新闻与代码工具进行灵活推理,但其信息获取、证据验证与资产配置的**操作规程**通常以固定系统提示的形式在部署前人工设定,此后不再改变。论文指出,这种静态策略存在三重局限: - **非平稳市场适应性差**:金融市场机制持续变化(震荡、急跌、上行),固定策略无法动态调整分析重心; - **经验反馈无法沉淀**:代理每日产生大量决策轨迹与组合回报,但固定提示无法吸收这些信号来改进可复用的决策流程; - **分析词汇僵化**:实证显示,静态代理被锁定在狭窄的分析指标集合中(如 20 日收益、年化波动率),极少激活与当前机制真正相关的指标(如 VaR、EMA、RSI)。 ### 2. 核心方法:EvolveTrade 论文将系统提示重新概念化为一个**文本参数化的时变策略** πt 。在每个交易批次结束后,一个独立的 **Policy Agent** 读取该批次的交易经验,将策略从 π_t 精炼为 π(t+1) ,而底层 LLM 参数与工具接口始终保持冻结。 **形式化框架**: - 第 t 日代理状态: St = (π_t, t, P(t-1), T) ,其中 P(t-1) 为前期组合状态, T 为工具集; - 决策轨迹: h_t = (w_t, d_t^(dec)) ,包含资产配置与每资产推理依据; - 事后反馈: g_t ,包含实现收益、资产收益与仓位变化; - 精炼记录: R_t = (h_t, g_t) ; - 批次记录(每 N 日): B_k = R((k-1)N+1), dots, R_(kN) ; - 策略更新:

π(kN+1) = f(update)(π(kN), B_k) = LLM(I(update), π(kN), B_k)
Policy Agent 的更新指令 I
(update) 要求将具体观察到的行为与提示文本中的特定条款关联,执行有依据的编辑(grounded edits),可修改工具选择、信号交叉验证规则、风险控制与输出结构等。 ### 3. 实验设计 实验基于日收盘回测环境,投资范围为 15 只美股蓝筹股加现金,评测窗口覆盖六种后知识截断的市场机制: - **GPT-5-mini**:2025 年 1 月(震荡)、4 月(急跌反弹)、9 月(上行趋势) - **Gemini-2.5-Flash**:2025 年 11 月(看跌)、2026 年 2 月(震荡)、2026 年 4 月(看涨) 对比基线包括规则策略(SPY、B&H、MACD 等)与 LLM 策略(Static Base、Static Tool-Calling、EvolveBase、EvolveStrategy)。默认每 N=5 个交易日执行一次策略更新。核心评测指标为年化夏普比率(SR)、累计收益率(CR)、最大回撤(MDD)、胜率(WR)与日波动率(Vol)。 ### 4. 主要结果与发现 - **风险收益表现提升**:EvolveTrade 在多数模型-机制组合中取得最优或次优的 SR 与 CR。例如,GPT-5-mini 在 2025 年 1 月与 9 月分别实现 SR 5.12 与 8.43;Gemini-2.5-Flash 在 2025 年 11 月表现最佳,在 2026 年 2 月优于全部规则基线。 - **机制相关指标的激活**:行为分析显示,静态代理在三种机制中调用几乎相同的指标;而 EvolveTrade 在 4 月急跌中激活了 **VaR** 与**信号归一化**,在 9 月上行中激活了 **EMA、RSI、SMA** 等趋势跟踪指标,扩展了分析词汇。 - **工具使用模式转变**:EvolveTrade 的代码解释器调用频率从约 1.0 次/日提升至 2.6–4.5 次/日,而新闻检索未显著增加,表明策略进化将重心从文本搜集转向**可执行的数量化分析**。 - **案例级归因**:2025 年 1 月 24 日 NVDA 大跌前,EvolveTrade 经策略精炼将 NVDA 仓位从 10.7% 降至 2.9%,次日 NVDA 下跌 17.0%,EvolveTrade 实现 -0.03% 而静态代理为 -1.11%,展示了策略-仓位的可追踪因果链。 - **长期稳定性与成本稳健性**:在 50 个交易日长期实验中,EvolveTrade 持续领先;引入 10 bps 交易成本后,其收益优势仍然保持,且换手率低于多数 LLM 基线。策略文本长度呈非单调变化,可周期性地整合冗余指令。 - **更新频率敏感度**:每日更新( N=1 )因过拟合短期噪声而表现最差, N=5 达到最佳平衡,提示策略进化需要适度的反馈累积周期。 ### 5. 局限与未来方向 论文指出当前框架采用**固定更新间隔**,未根据市场变化动态调度;交易成本估计未纳入滑点与流动性约束;且在个别窗口(如 2025 年 4 月与 2026 年 4 月)出现过高的防御性现金配置,错失上涨收益。 ### 6. 结论 EvolveTrade 表明,LLM 交易代理的核心优化对象不应仅是单次交易决策,而是**可复用的信息获取与行动程序本身**。通过将系统提示视为可从实现经验中在线精炼的策略文本,代理能够在不重新训练模型、不修改工具的前提下,持续提升风险调整收益并激活机制相关的分析能力。这为构建适应非平稳金融市场的稳健 LLM 代理提供了关键方向。 Q7: 想要进一步了解论文 以上只是了解一篇论文的几个基本FAQ。如果你还想与Kimi进一步讨论该论文,请点击 这里 为你跳转Kimi AI网页版,并启动一个与该论文相关的新会话。

Authors: Sehee Kim, Yumin Choi, Minki Kang, Sung Ju Hwang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.17632.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17632

Published: 2026-09-18T01:29:19.418Z


3. One Color Preprocessing Improves DSATUR

Abstract:The Graph Coloring Problem (GCP) is NP-hard and DSATUR stands as one of the fastest heuristics for it despite producing colorings that typically use more colors than state-of-the-art coloring algorithms. We propose SSLD (Semidefinite Spectral Learning with DSATUR), which improves DSATUR by preprocessing a first good color class before letting DSATUR complete coloring the rest of the given graph. We obtain this color class from a Semidefinite Programming (SDP), similar to an SDP used to compute the Lovász theta number. To the best of our knowledge, SSLD is the first approach to improve DSATUR by preprocessing through fixed color classes. We evaluate SSLD against DSATUR and against a naive 1-color-class preprocessing algorithm on DIMACS instances, random graphs (Erdős—Rényi, Watts-Strogatz, Barabási—Albert), Frequency Assignment and Job Shop Scheduling instances. SSLD matches or beats DSATUR in almost every case across over 1600 benchmark instances, and out performs the naive GISD baseline, allows us to confirm the value brought by the SDP-guided choice of the first color class. This quality comes at a runtime cost of roughly 195 times slower that DSATUR, but demonstrating that SDP-guided preprocessing of a first color class is a direction for future improvements.

中文摘要

摘要:图着色问题(GCP)是NP难问题,尽管DSATUR通常生成的着色使用的颜色比最先进的着色算法更多,但它仍然是其中最快的启发式算法之一。我们提出了SSLD(带DSATUR的半正定谱学习),通过在让DSATUR完成对给定图的其余部分的着色之前,预处理出第一个好的颜色类来改进DSATUR。我们从半正定规划(SDP)中获得这个颜色类,该SDP类似于用于计算洛伐兹θ数的SDP。据我们所知,SSLD是第一个通过固定颜色类预处理来改进DSATUR的方法。我们在DIMACS实例、随机图(Erdős—Rényi、Watts-Strogatz、Barabási—Albert)、频率分配和作业车间调度实例上,将SSLD与DSATUR以及一个简单的单颜色类预处理算法进行了评估。在超过1600个基准实例中,SSLD几乎在每种情况下都能匹配或优于DSATUR,且优于简单的GISD基线,这使我们能够确认SDP指导的第一个颜色类选择所带来的价值。这种优势以大约比DSATUR慢195倍的运行时间为代价,但证明了通过SDP指导预处理首个颜色类是未来改进的一个方向。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17633 (HTTP 429)

Authors: Adam Nouira, Lucas Isenmann

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.17633.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17633

Published: 2026-09-18T01:29:19.418Z


4. Physics-Constrained Digital Twins for Sensor Integrity in Urban Pedestrian Flow: Detecting Stealthy False Data Injection with Conformal Guarantees

Abstract:City pedestrian counting systems now feed economic indicators, planning decisions and safety operations, yet the twins built on top of them treat the incoming stream as ground truth. We study what happens when it is not. We formalise stealthy false data injection for city-scale pedestrian sensing, where the map from latent flow to observation is far more rank deficient than in the power and water networks for which stealth has been characterised. Our twin estimates directed flows on the pedestrian street graph, assimilates counts through a learned graph-localised gain, and is trained against a flow conservation residual that couples metered and unmetered segments. Detection combines the innovation with that residual, and the alarm threshold is set by adaptive conformal calibration rather than by hand. To measure what the physics buys, we define the attack margin, the relative reduction in worst-case corruption of the estimated flow field, achieved against a white-box adversary that optimises directly through the twin. On six years of Melbourne data the margin reaches 0.54 against a single compromised device and falls to 0.19 when a third of the fleet is compromised, on a network where only 1.18 per cent of walkable segments are metered. Replacing the street graph by a distance graph collapses it to 0.09, which shows that the gain comes from the conservation law rather than from locality.

中文摘要

摘要:城市行人计数系统现在为经济指标、规划决策和安全操作提供数据,但构建在其上的数字孪生将输入数据流视为真实情况。我们研究在输入数据不为真实情况时会发生什么。我们形式化了针对城市规模行人感知的隐蔽性虚假数据注入,其中从潜在流量到观测的映射比已表征隐蔽性的电力和水网络在秩上要缺失更多。我们的孪生模型在行人街道图上估计有向流量,通过学习的图局部增益同化计数,并通过流量守恒残差进行训练,该残差将被计量和未计量的路段耦合。检测将创新量与该残差结合,报警阈值通过自适应符合校准设定,而非手动设定。为了衡量物理规律带来的好处,我们定义了攻击余量,即在面对通过孪生模型直接优化的白箱攻击者时,对估计流场最坏情况破坏的相对降低。在墨尔本六年的数据上,面对单个受损设备时,余量达到0.54,而当三分之一的车队受损时跌至0.19,而网络中只有1.18%的可步行路段被计量。将街道图替换为距离图,会将其降至0.09,这表明增益来自守恒定律,而非局部性。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17635 (HTTP 429)

Authors: Oscar Mogollon Gutierrez, Fatemeh Ghasemi, Mohammadhossein Homaei, Andres Caro, Mar Avila

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.17635.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17635

Published: 2026-09-18T01:29:19.418Z


5. What You Can’t See Is Still What You Learn: A Preregistered Sixty-Society Confirmation That Evidence Masking Drives Compositional Generalization

Abstract:Restricting what a module can read may improve what a system learns to compute. We test this in a preregistered confirmation with sixty four-cell systems sharing a frozen language-model backbone and communicating through learned continuous packets. Five conditions vary evidence masking, ownership markers, and replacement of foreign evidence with neutral filler, across six initialization clusters, each with two data orders, on one fresh task world. With markers available in both regimes, masking improved accuracy on held-out two- and three-operation compositions by median paired differences of 0.846 and 0.859; all twelve pairs cleared the required margins, and the full preregistered behavioral criterion passed. The unmarked replication also passed. No globally visible system passed the marker-following check, so the effect of usable role information remains unresolved. The filler condition yielded seven full generalizers, but its decomposition criteria were inconclusive. Packet interventions in all eighteen audited masked systems followed the predicted intermediate-value changes on eligible cases; these finite, success-conditioned audits do not establish mediation. The results confirm a large advantage of the tested masking regime, while leaving its finer attribution and generality open. Protocols, results, and checkpoints are public.

中文摘要

摘要:限制模块可以读取的内容可能会提高系统学习计算的能力。我们在一项预注册确认实验中验证了这一点,该实验使用六十四个单元的系统,这些系统共享冻结的语言模型骨干,并通过学习的连续数据包进行通信。实验设置了五种条件,分别变化了证据屏蔽、所有权标记,以及用中性填充物替换外来证据,在六个初始化集群中,每个集群有两种数据顺序,在一个新的任务世界中进行测试。在两种制度下都提供标记的情况下,屏蔽在保留的两步和三步操作组合上的准确率有所提高,中位配对差异分别为0.846和0.859;所有十二对都超过了所需的标准,且完全满足预注册的行为准则。无标记的重复实验也通过了检验。没有任何全局可见的系统通过了标记跟随检查,因此可用角色信息的效果仍未解决。填充条件产生了七个完全泛化者,但其分解标准不确定。对所有十八个审核过的屏蔽系统进行的数据包干预遵循了预测的中间值变化;这些有限的、以成功为条件的审核不能确立中介关系。结果确认了所测试的屏蔽机制具有显著优势,同时其更精细的归因和普适性仍待探讨。协议、结果和检查点均已公开。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17637 (HTTP 429)

Authors: Narcis Marincat

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.17637.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17637

Published: 2026-09-18T01:29:19.418Z


6. CapMem: A Benchmark for Caption-Based Episodic Memory in Egocentric Video

Abstract:Wearable assistants require episodic memory over egocentric video, yet current vision-language models face bounded frame budgets, growing visual-token costs, and long-context retrieval failures. Under these practical constraints, we study whether textual captions can serve as reusable episodic memory. We define the Episodic Memory Video Caption QA task and introduce CapMem, a human-annotated benchmark with 75 videos totaling 33.7 hours, and 1,000 multiple-choice questions across 16 scenarios. On long videos (>20 min), full-coverage CaptionQA with 30s and 60s caption windows outperforms direct VideoQA for 10/12 and 8/12 models, respectively. On the same video subset, a matched-frame control across six Qwen models retains mean accuracy gains of 3.22 and 2.55 points, respectively. Our caption-guided retrieve-and-verify harness further improves accuracy by up to 5.3 points. These results support the effectiveness of caption memory for episodic reasoning over long egocentric video.

中文摘要

摘要:可穿戴助手在以自我为中心的视频上需要情节记忆,但现有的视觉-语言模型面临帧数限制、不断增加的视觉代币成本以及长上下文检索失败的问题。在这些实际约束条件下,我们研究文本字幕是否可以作为可重用的情节记忆。我们定义了情节记忆视频字幕问答(Episodic Memory Video Caption QA)任务,并介绍了 CapMem,这是一个人类标注的基准数据集,包含 75 个视频,总时长 33.7 小时,以及覆盖 16 种场景的 1,000 道多项选择题。在长视频(>20 分钟)上,使用 30 秒和 60 秒字幕窗口的全覆盖 CaptionQA 在 12 个模型中分别优于直接 VideoQA 的模型数量为 10/12 和 8/12。在相同的视频子集上,六个 Qwen 模型的匹配帧控制分别保持平均准确率提升 3.22 和 2.55 个百分点。我们的字幕引导检索和验证机制进一步将准确率提高最多 5.3 个百分点。这些结果支持了字幕记忆在长自我中心视频上的情节推理中的有效性。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17688 (HTTP 429)

Authors: Dingli Liang, Yiqiao Xie, Yukai Huang, Zhaokai Wang, Weitong Cai, Guangwen Feng, Jifei Song, Zhensong Zhang, Hang Zhang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.17688.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17688

Published: 2026-09-18T01:29:19.418Z


7. GraphEcho: Structural Redundancy and Evidence Provenance in LLM Graph Agents

Abstract:A large language model (LLM) agent can follow more graph paths without acquiring more independent evidence. GraphEcho tests whether agents mistake these repeated encounters for additional corroboration. The benchmark varies path counts and evidential origins while holding evidence content fixed, and evaluates both judgments and active exploration. Controlled synthetic experiments reveal model-dependent judgment shifts, but redundant supporting paths increase the share of repeated walks across all evaluated frozen agents. Provenance-aware post-training (PAPT) reduces revisits and improves synthetic accuracy, yet covers fewer distinct sources. On scientific claims, it continues to reduce repetition while accuracy declines. These findings expose a gap between efficient exploration and effective evidence use: an agent can learn to stop repeating itself while overlooking information it needs. GraphEcho provides a controlled way to evaluate both what graph agents conclude and whether their exploration reaches distinct evidential sources.

中文摘要

摘要:一个大型语言模型(LLM)代理可以在不获取更多独立证据的情况下,沿更多图路径进行推演。GraphEcho 测试代理是否会将这些重复的遭遇误认为是额外的证据支持。该基准在保持证据内容不变的情况下,变化路径数量和证据来源,并评估判断和主动探索两方面。受控的合成实验显示了模型依赖的判断变化,但冗余的支持路径增加了所有被评估冻结代理重复遍历的比例。源感知的后训练(PAPT)减少了重复访问并提高了合成准确性,但覆盖的不同来源较少。在科学声明上,它仍然减少重复访问,但准确性下降。这些发现揭示了高效探索与有效利用证据之间的差距:代理可以学会停止自我重复,但却忽略了所需的信息。GraphEcho 提供了一种受控的方法来评估图代理的结论以及它们的探索是否触及了不同的证据来源。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17695 (HTTP 429)

Authors: Sikun Wang, Yixi Zhou, Lei Fan, Fan Zhang

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.17695.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17695

Published: 2026-09-18T01:29:19.418Z


8. GVD: Governed Versioning and Deduplication for Document Repositories

Abstract:Document repositories evolve continuously. Guidelines and policies are revised, superseded, and re-uploaded, so the same content recurs in different wording and newer versions refine or contradict earlier ones. These inconsistencies belong to the growing collection rather than to any single document, yet existing work treats versioning, duplicate detection, and contradiction detection as isolated pairwise tasks and stops once a pair is labeled. We present GVD (Governed Versioning and Deduplication), a framework that unifies cross-document version linking with rule-level conflict resolution under an auditable update policy. Incoming documents are assigned to version families through bidirectional rule alignment, and their rules are compared against the family memory to identify duplicates, contradictions, asymmetric refinements, and new knowledge, with Counterfactual Span Probing (CSP) resolving related pairs that inference misclassifies as neutral. Relation-specific policies suppress duplicates and escalate only consequential changes for review, retaining version lineage as an audit trail. The pipeline runs fully locally, with no large language model. On 120 enterprise documents processed as 140 ingestions across 59 version families, GVD reaches an F1 of 0.97 for version-family construction and 0.94 for rule-level consistency, with CSP raising rule consistency from 0.90 to 0.94.

中文摘要

摘要:文档库在不断演变。指南和政策会被修改、替换并重新上传,因此相同的内容会以不同的措辞重复出现,新版本会完善或与早期版本产生矛盾。这些不一致属于不断增长的文档集合,而不是某个单一文档。然而,现有的工作将版本管理、重复检测和矛盾检测视为孤立的成对任务,并在标注一对文档后就停止。我们提出了GVD(Governed Versioning and Deduplication,受控版本管理与重复清理)框架,该框架在可审计的更新策略下,将跨文档的版本关联与规则级冲突解决统一起来。新进文档通过双向规则对齐被分配到版本族中,并将其规则与版本族记忆进行比较,以识别重复、矛盾、不对称细化以及新知识,其中反事实跨度探查(CSP)解决推理误判为中性相关对的问题。针对关系的策略抑制重复,仅将重要的变更升级以供审查,同时保留版本沿革作为审计轨迹。该流程完全在本地运行,无需使用大型语言模型。在处理59个版本族中的120份企业文档,共140次导入的实验中,GVD在版本族构建上的F1达到0.97,在规则级一致性检测上的F1达到0.94,而CSP将规则一致性从0.90提升至0.94。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17696 (HTTP 429)

Authors: Mohammadreza Sediqin, Shivali Dalmia, Sumukha Thoppanahalli, Abhishek Mukherji

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.17696.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17696

Published: 2026-09-18T01:29:19.418Z


9. NeMo Data Designer: An Extensible Framework for Multimodal Synthetic Data Generation

Abstract:We present NeMo Data Designer (NDD), an open-source, general-purpose framework for multi-modal synthetic data generation (SDG). Designed to be intuitive to use, NDD provides a declarative configuration format in which human and/or agent users define each dataset column, with column types spanning text, code, structured outputs, images, embeddings, and statistical samplers that are explicitly configured to steer dataset diversity. Additional column types and functionality can be introduced using the framework’s flexible plugin system. NDD’s configuration is an inspectable artifact, supporting workflow sharing and reproducibility. SDG is an inherently iterative process. NDD therefore builds a preview-and-revision loop into its core workflow, allowing users to generate and inspect a small number of records, refine the specification, and rerun generation at full scale. At runtime, NDD resolves dependencies, schedules calls to user-provided model endpoints, and retries failed requests. We describe NDD’s architecture and programming model and present case studies spanning structured, agentic, multimodal, and domain-specialized tasks, including datasets used in Nemotron model development and in production enterprise deployments.

中文摘要

摘要:我们提出了 NeMo 数据设计器(NeMo Data Designer,NDD),这是一个开源的、通用的多模态合成数据生成(SDG)框架。NDD 旨在使用直观,提供了一种声明式配置格式,用户(人类和/或代理)可以在其中定义每个数据集列,列类型涵盖文本、代码、结构化输出、图像、嵌入以及可显式配置以引导数据集多样性的统计采样器。通过框架灵活的插件系统,还可以引入其他列类型和功能。NDD 的配置是一个可检查的工件,支持工作流共享和可重复性。SDG 本质上是一个迭代过程。因此,NDD 在其核心工作流中构建了预览与修订循环,允许用户生成并检查少量记录,完善规范,然后全量重新生成。在运行时,NDD 会解析依赖关系、调度对用户提供的模型端点的调用,并重试失败请求。我们描述了 NDD 的架构和编程模型,并展示了涵盖结构化、代理、多模态和领域专用任务的案例研究,包括在 Nemotron 模型开发和企业生产部署中使用的数据集。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17699 (HTTP 429)

Authors: Johnny Greco, Nabin Mulepati, Andre Manoel, Eric Tramel, Kirit Thadaka, Mike Knepper, Dhruv Nathawani, Dane Corneil, Yev Meyer, Alex Watson, Maarten Van Segbroeck

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.17699.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17699

Published: 2026-09-18T01:29:19.418Z


10. A Systematic Evaluation of the COTQ Provincial Land Cover Product: Structural Consistency, Spectral Separability, and Relative Positioning Against ESA, ESRI, and Google Products

Abstract:High-resolution land use and land cover (LULC) products derived from Sentinel-2 imagery are widely used for environmental monitoring and land management, yet their performance can vary across regions with complex ecological gradients and heterogeneous surface conditions. In Quebec, these limitations motivated the development of a provincial 10-m land-cover product, the COTQ, designed to support annual monitoring of land occupation and soil artificialisation. This study presents a systematic evaluation of the COTQ product relative to three global 10-m LULC datasets: ESA WorldCover, ESRI LandCover, and Google DynamicWorld. This paper does not introduce a new mapping methodology but focuses on analysing the behaviour and consistency of the COTQ using complementary evaluation approaches. All products are harmonized under a common legend and compared using structural indicators (object-size distributions, shape complexity, Adjusted Rand Index, and Intersection over Union), spectral separability metrics derived from Sentinel-2 reflectance data, and a targeted photo-interpretation of disagreement areas. The analysis is conducted over eight Sentinel-2 tiles selected to represent the main bioclimatic domains of Québec, from temperate and boreal forests to northern tundra environments. The results show that the COTQ exhibits structural and spectral characteristics most similar to ESA WorldCover among the reference global products, while revealing systematic differences linked to class definitions and thematic priorities, particularly for urban areas, wetlands, and rocky or cryptogamic surfaces. This multi-criteria evaluation provides an objective characterization of the COTQ product and clarifies its relative positioning with respect to existing global land-cover datasets for operational land monitoring in Québec.

中文摘要

摘要:基于Sentinel-2影像的高分辨率土地利用和土地覆盖(LULC)产品广泛用于环境监测和土地管理,但其性能在具有复杂生态梯度和异质地表条件的地区之间可能存在差异。在魁北克,这些局限促使开发了省级10米土地覆盖产品COTQ,旨在支持对土地占用和土壤人工化的年度监测。本研究系统地评估了COTQ产品相对于三个全球10米LULC数据集:ESA WorldCover、ESRI LandCover和Google DynamicWorld。本文未引入新的制图方法,而是侧重于利用互补评估方法分析COTQ的行为和一致性。所有产品在统一图例下进行协调,并使用结构指标(物体大小分布、形状复杂度、调整后兰德指数和交集与联合)、源自Sentinel-2反射率数据的光谱可分离性指标,以及对分歧区域的有针对性照片解释进行比较。分析在八块Sentinel-2瓷砖上进行,这些瓦片选取代表魁北克主要生物气候区域,从温带和针叶林到北部苔原环境。结果显示,COTQ在参考全球产品中展现出与ESA WorldCover最为相似的结构和光谱特征,同时揭示了与类别定义和主题优先事项相关的系统差异,尤其是在城市区域、湿地以及岩石或隐花地表中。这一多标准评估客观地描述了COTQ产品,并澄清了其相对于魁北克现有全球土地覆盖数据集的相对位置,以便进行操作性土地监测。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17731 (HTTP 429)

Authors: Étienne Clabaut, Samuel Foucher, Yacine Bouroubi

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.17731.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17731

Published: 2026-09-18T01:29:19.418Z


VLM Domain Papers

1. Selective Prediction and Uncertainty-Aware Referral for Pap Smear Classification

Abstract:Deep learning models for cervical cytology are almost always evaluated as if every prediction must be acted upon, yet a screening system deployed alongside a cytopathologist need not classify every slide: it can defer the cases it is least certain about. Evaluating such a system requires asking not only how often it is correct, but whether its confidence ranks its errors to the bottom. This paper studies selective prediction and uncertainty-aware referral on the Herlev Pap smear dataset under a binary Normal-versus-Abnormal formulation. Two lightweight transformer backbones (Swin-Tiny, TinyViT-5M) are fine-tuned on Herlev from ImageNet-pretrained weights with weighted random sampling, calibrated by post-hoc temperature scaling fit on a held-out calibration subset, and compared against a soft-voting ensemble of both models. Discrimination is reported alongside expected calibration error (ECE) and, as the primary endpoint, the area under the risk-coverage curve (AURC). No statistically significant difference was detected between the two configurations in accuracy or macro-F1, yet the ensemble halves AURC (0.0022 vs. 0.0045, a 51.8% reduction, lower in all five folds) and extends the coverage at which zero errors are made from 18.3% to 72.8% of the pooled test predictions. The same ensemble is nonetheless worse calibrated in absolute terms (ECE 0.0339 vs. 0.0247) and produces more false negatives (14 vs. 10). These results separate two properties that are frequently conflated: the ability to rank predictions by trustworthiness, and the accuracy of the confidence values themselves. Ensembling improves the former while degrading the latter, and the former directly governs the observed risk-coverage tradeoff, whereas the latter governs the interpretation of the reported confidence values.

中文摘要

摘要:宫颈细胞学的深度学习模型几乎总是被评估为必须对每个预测采取行动,但与细胞病理学家并用的筛查系统不必对每一张切片进行分类:它可以推迟对最不确定的病例进行分类。评估此类系统不仅需要问其正确频率,还要考虑其置信度是否使误差排名垫底。本文研究了Herlev宫颈涂片数据集中选择性预测和不确定性意识转诊,采用二元正常与异常表述。Herlev上对两个轻量级变压器骨干(Swin-Tiny、TinyViT-5M)进行了微调,基于ImageNet预训练权重加权随机抽样,通过事后温度尺度拟合校准,并与两个模型的软投票集合进行比较。辨别率与预期校准误差(ECE)及主要终点风险覆盖曲线面积(AURC)一同报告。两种配置在准确率或宏观F1上未检测到统计学上显著差异,但集合变量使AURC减半(0.0022对0.0045,减少51.8%,五个方向均更低),并将零误差覆盖率从18.3%扩展到72.8%。同一集合在绝对值上校准较差(ECE 0.0339对0.0247),且产生更多假阴性(14对10)。这些结果区分了两个常被混淆的属性:按可信度排序预测的能力,以及置信度本身的准确性。集成改进前者,同时降低后者,前者直接支配观察到的风险与覆盖权衡,而后者则控制报告置信值的解释。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17545 (HTTP 429)

Authors: Nisreen Albzour, Sarah S. Lam

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2609.17545.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17545

Published: 2026-09-18T01:29:44.334Z


2. A Heisenberg Lift Descriptor for Order Sensitive Online Handwriting Recognition

Abstract:Online handwriting recognition systems typically represent pen trajectories through fixed-length Euclidean shape descriptors that capture the spatial outline of each stroke, but are insensitive to the order in which that outline is produced. Two strokes that trace the same region of the plane in opposite directions are indistinguishable to any such order-blind representation, yet their traversal directions may carry decisive class information in characters where loop orientation and stroke sequencing matter. This paper introduces a Heisenberg-lift framework that addresses this gap through a compact, interpretable, order-sensitive augmentation for online pen-trajectory features. The simplest instance is the terminal signed area, a single parameter-free scalar appended to an existing Euclidean descriptor at negligible computational cost. Evaluated on two standard online handwriting benchmarks, this one-scalar addition, consistently raises classifier accuracy over the Euclidean baseline. On the hardest character pair in our study, the letters o and y, the signed area alone achieves perfect separation while the Euclidean baseline falls short. The advantage grows further under additive coordinate noise, a practically relevant degradation in pen-trajectory data. A richer fifteen-dimensional extension, derived from a noncommutative Heisenberg-group subdivision scheme, provides additional gains in noisy and loop-structured conditions. Dimension-matched statistical controls confirm that all improvements reflect geometric information rather than feature-count inflation. The resulting descriptor is lightweight, closed-form, and directly interpretable, making it a practical augmentation for online handwriting and related document-trajectory classification pipelines in which the direction of stroke execution carries discriminative information.

中文摘要

摘要:在线手写识别系统通常通过固定长度的欧几里得形状描述符来表示笔迹轨迹,这些描述符捕捉每一笔的空间轮廓,但对于生成该轮廓的顺序不敏感。两个以相反方向描绘同一区域的笔画,对于任何这种顺序盲目表示来说都是无法区分的,但在循环方向和笔画顺序重要的字符中,它们的遍历方向可能携带决定性的类别信息。本文提出了一个海森堡升阶框架,通过对在线笔迹特征进行紧凑、可解释、顺序敏感的增强来弥补这一空白。最简单的实例是终端符号面积,它是一个无需参数的标量,可附加在现有欧几里得描述符上,计算成本极低。在两个标准在线手写基准上评估时,这一标量补充始终提高了分类器的准确率,相较于欧几里得基线。在我们研究中最难的字符对——字母o和y,仅符号面积就实现了完美区分,而欧几里得基线则不够。该优势在坐标噪声叠加下进一步增强,这是笔迹轨迹数据中实际相关的退化情况。一个由非交换海森堡群细分方案导出的更丰富的十五维扩展,在噪声和环状结构条件下提供了额外的提升。维度匹配的统计控制确认,所有改进反映的是几何信息而非特征数量增加。最终得到的描述符轻量、闭式且可直接解释,使其成为在线手写及相关文档轨迹分类流程中实用的增强方法,其中笔画执行方向携带判别信息。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17565 (HTTP 429)

Authors: Hassan Ugail, Newton Howard

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2609.17565.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17565

Published: 2026-09-18T01:29:44.334Z


3. Adaptive Interpolatory Curve Subdivision with Learned Local Angles

Abstract:Curve subdivision is pivotal in computer graphics for generating smooth geometric objects from control polygons. Interpolatory subdivision is especially attractive because the refined curve is guaranteed to pass through the designer’s control points. Classical four-point and six-point schemes preserve this property, but their behaviour is governed by a single global tension parameter, limiting their ability to adapt across flat regions, sharp turns and varying local geometries. We introduce an adaptive local-angle formulation that keeps the interpolatory structure intact while learning how each new vertex should be inserted. A compact edge-wise predictor assigns one insertion angle per edge, while the original vertices are copied exactly at every refinement level. Interpolation is therefore a structural property of the operator and does not depend on the trained weights. The same predictor is used with geometry-specific geodesic primitives on the Euclidean plane, the two-sphere and the Poincaré disk. Under a matched-density evaluation protocol, the method reduces nearest-neighbour error by factors of five to seventeen over the best validation-tuned fixed-tension baseline, and by about 1.8 over centripetal Catmull-Rom in the Euclidean case. It also substantially reduces bending energy and tangent roughness, while remaining competitive with separately trained per-geometry models.

中文摘要

摘要:曲线细分在计算机图形学中对于从控制多边形生成平滑几何物体至关重要。插值细分尤其具有吸引力,因为细化后的曲线保证通过设计者的控制点。经典的四点和六点方案保持了这一特性,但它们的行为受单一全局张力参数控制,从而限制了其在平坦区域、急转弯和不同局部几何形状下的自适应能力。我们提出了一种自适应局部角度的形式,它在保持插值结构完整的同时,学习每个新顶点应如何插入。紧凑的边缘预测器为每条边分配一个插入角,而原始顶点在每个细化层都被精确复制。因此,插值是该算子的结构性特征,而不依赖训练权重。同一个预测器在欧几里得平面、二维球面和庞加莱圆盘上结合几何特定的测地线原型使用。在匹配密度的评估协议下,该方法相较经过验证调优的固定张力基线,可将最近邻误差降低五到十七倍,在欧几里得情况下相较向心Catmull-Rom降低约1.8倍。同时,它显著降低了弯曲能量和切线粗糙度,同时在不同几何模型单独训练的情况下仍具有竞争力。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17566 (HTTP 429)

Authors: Hassan Ugail, Newton Howard

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2609.17566.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17566

Published: 2026-09-18T01:29:44.334Z


4. DualCount: Structurally Consistent Density and Point Modeling for Zero-Shot Object Counting

Abstract:Zero-shot object counting aims to estimate the number of objects specified by a text query without category-specific training. Recent approaches primarily rely on density regression or detection-style instance prediction. While effective, density-based models often suffer from spatial ambiguity and background leakage due to weakly regulated mass allocation, leading to fragmented or part-biased representations that increase counting error in complex scenes. In this work, we propose an instance-aware dual-decoder framework that structurally couples density and point representations for zero-shot object counting. Instead of treating density estimation as independent pixel-wise regression, we interpret it as a structured mass allocation problem over a latent set of object instances. Predicted instance centers induce a soft instance-wise decomposition of the density map, upon which we enforce two geometric constraints: (1) per-instance mass conservation, ensuring each object contributes approximately one unit of density mass, and (2) center-of-mass alignment, encouraging each density component to concentrate around its corresponding predicted center. These constraints introduce instance-level geometric consistency and lead to more accurate mass allocation, thereby reducing counting error. Extensive experiments on FSC-147, PUCPR+, and CARPK show that our approach consistently reduces counting error and establishes new state-of-the-art performance in zero-shot object counting.

中文摘要

摘要:零样本对象计数旨在在没有特定类别训练的情况下,根据文本查询估计对象数量。最近的方法主要依赖于密度回归或检测风格的实例预测。虽然有效,但基于密度的模型由于质量分配约束较弱,通常会出现空间模糊和背景泄漏,导致分散或部位偏向的表示,从而在复杂场景中增加计数误差。在本工作中,我们提出了一个面向实例的双解码器框架,将密度表示和点表示在结构上结合,用于零样本对象计数。我们不再将密度估计视为独立的像素级回归,而是将其解释为潜在对象实例集合上的结构化质量分配问题。预测的实例中心引导密度图的软实例级分解,基于此我们施加两个几何约束:(1) 每实例质量守恒,确保每个对象大约贡献一个单位的密度质量;(2) 质心对齐,鼓励每个密度组件集中在其对应的预测中心周围。这些约束引入了实例级别的几何一致性,从而实现更准确的质量分配,降低计数误差。在 FSC-147、PUCPR+ 和 CARPK 上的广泛实验表明,我们的方法在零样本对象计数中始终降低计数误差,并建立了新的最先进性能。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17613 (HTTP 429)

Authors: Xuan Cuong Ngo

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2609.17613.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17613

Published: 2026-09-18T01:29:44.334Z


5. Geometry-Driven Shadow Harmonisation for Composited Faces: A Multiplicative, Albedo-Preserving Relighting Pipeline

Abstract:Face swapping and face compositing pipelines routinely produce a face that is geometrically well aligned but photometrically implausible: the donor face carries flat, near-frontal studio illumination while the host body and background carry directional scene light. Most existing remedies re-synthesise the face through colour transfer, neural relighting, or inverse rendering, and therefore risk altering identity, skin tone, and texture. We present a conservative alternative: geometry-driven form-shadow injection. The pipeline never repaints the face. It estimates a per-pixel gain field $g\in[g{\min},1]$ from a rasterised 3D face proxy and multiplies it channel-uniformly onto linear RGB, so the operator can only darken and cannot shift chromaticity. A dense landmark mesh is rasterised into a depth buffer, from which we derive surface normals, a cavity term, and screen-space cast shadows. Key-light direction is estimated from host-side cues (body, background, hair halo); on-face cues are downweighted because they recover the donor’s lighting. Shadow magnitude is not matched to the host: it is set by a three-parameter transfer $(\tau,\sigma,g{\min})$. The shading field is divided by its 75th percentile over skin, then gated, scaled, clamped, smoothed, and re-clipped inside a feathered, skin-gated face mask. On an analytic face heightfield, the default $(\tau,\sigma,g_{\min})=(0.90,0.45,0.82)$ modifies 56.5% of face pixels with mean gain 0.938 (0.890 on modified pixels) and drives 3.4% of pixels to the floor. Hue invariance is a corollary of the operator. We analyse the transfer in closed form, ablate its parameters, and discuss failure modes of a monotone, darkening-only formulation, including double-shadowing of non-flat donors.

中文摘要

摘要:面部交换和面部合成流程常常产生几何对齐但光度学上不合理的面部:供体面部承载平面、近前方工作室照明,而主体和背景则携带定向场景光。大多数现有方案通过颜色转移、神经重照或反向渲染重新合成面部,因此存在改变身份、肤色和纹理的风险。我们提出了一个保守的替代方案:几何驱动形影注入。该管道从不重新绘制面部。它从光栅化3D面部代理估算每像素增益场$g\in[g{\min},1]$,并均匀地将该值乘以线性RGB,因此操作员只能变暗且无法改变色度。密集的地标网格被光栅化成深度缓冲区,从中我们推导出表面法线、空腔项和屏幕空间投射阴影。主光方向由宿主侧线索(身体、背景、发光)估计;面部线索因恢复供体光照而权重降低。阴影大小不匹配宿主:由三参数转移 $(\tau,\sigma,g{\min})$ 设定。阴影场被除以皮肤上的第75百分位,然后进行门控、缩放、夹持、平滑,并在带有羽毛、皮肤门控的面罩内重新裁剪。在解析面高场上,默认的 $(\tau,\sigma,g_{\min})=(0.90,0.45,0.82)$ 修改了 56.5% 的人脸像素,平均增益为 0.938(修改后像素为 0.890),并将 3.4% 的像素降至最低。色相不变性是算符的一个推论。我们分析闭合形式的转移,消融其参数,并讨论单调、仅加暗的表述的失效模式,包括非平面供体的双影。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17740 (HTTP 429)

Authors: Vijesh KP

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2609.17740.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17740

Published: 2026-09-18T01:29:44.334Z


6. How to make effective use of domain experts for image classification?

Abstract:A lot of expectations have been put for years on integrating domain expert knowledge in image classification models. Several approaches have been explored, Concept Bottleneck Models (CBMs) opened up a new avenue of research leading to many variants, and more recently to Concept-based Embedding Models (CEMs). CBM consider binary encoding of each concept, while CEM expands this idea by embedding each concept through two vectors. However in real-life scenarii, domain experts’ knowledge is usually organized in concepts determined by various attributes, each attribute encoded either with numerical values, or range of values, or binary values, or categorical values. In this work, we first finetune an image feature extractor for classifying attributes representing the downstream object classes, where the class attributes have been specified by experts under various encoding formats. A classification head is then learnt from these various attributes to categorize target objects. We experimentally show that it improves the classification for three datasets: Kaggle fish dataset, AWA2 and a more challenging new wood charcoal dataset. We then propose an automatic selection of potential missclassified data. In this second step, experts are asked for those data to eventually modify the predicted attributes in order to improve the classification.

中文摘要

摘要:多年来,人们对在图像分类模型中整合领域专家知识寄予了很高的期望。已经探索了几种方法,概念瓶颈模型(Concept Bottleneck Models, CBMs)开辟了一条新的研究途径,产生了许多变体,而最近则发展为基于概念的嵌入模型(Concept-based Embedding Models, CEMs)。CBM 考虑对每个概念进行二进制编码,而 CEM 通过两个向量嵌入每个概念来扩展这一思想。然而在实际场景中,领域专家的知识通常是以由各种属性确定的概念组织的,每个属性可以用数值、数值范围、二进制值或类别值进行编码。在本研究中,我们首先微调图像特征提取器,以分类表示下游对象类别的属性,其中类属性已经由专家以各种编码格式指定。随后,从这些不同的属性中学习一个分类头,用于对目标对象进行分类。我们通过实验表明,这可以改善三个数据集的分类效果:Kaggle 鱼类数据集、AWA2 数据集以及一个更具挑战性的全新木炭数据集。随后,我们提出了一种自动选择潜在误分类数据的方法。在第二步中,将邀请专家对这些数据进行审查,以最终修改预测的属性,从而提高分类性能。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17749 (HTTP 429)

Authors: Dieu-Donné Fangnon, Diane Lingrand, Aurélie Liard, Marco Corneli, Antoine Pasqualini, Frédéric Precioso

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2609.17749.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17749

Published: 2026-09-18T01:29:44.334Z


7. Beyond Performance Metrics: Uncertainty Mapping of Label Ambiguity in Fazekas Score Prediction

Abstract:Reference labels used to train medical image classification models are not always as certain as they may appear, and this uncertainty has implications on performance metrics. In this study, we propose a framework to analyze model performance for periventricular Fazekas score prediction that goes beyond conventional metrics. The Fazekas score is an ordinal visual rating scale used to assess the severity of white matter hyperintensities and is known to be affected by inter-rater variability. While the best Fazekas score prediction model achieved a Matthews correlation coefficient (MCC) of 0.70, performance varied across data splits and loss functions, making interpretation of model capabilities difficult. Rather than interpreting epistemic uncertainty of a model’s prediction as an isolated scalar value, our approach of uncertainty mapping relates uncertainty to its position within the learned feature representation. This highlights regions of class-boundary transitions where cases appear more ambiguous and misclassifications are more likely. It also identifies potential label disagreement, including low-uncertainty misclassified cases that expert review found to be inconsistent with the original reference Fazekas score. Therefore, uncertainty mapping allows model behaviour to be examined in relation to class separation and potential model-label disagreement. Loss function choice also influenced the uncertainty profile, with some models showing clearer class separation and more localized uncertainty in ambiguous regions than others. These findings suggest that uncertainty mapping for Fazekas score predictions can support model interpretation and targeted dataset review when reference labels are affected by ambiguity/ inter-rater variability.

中文摘要

摘要:用于训练医学图像分类模型的参考标签并不总是像看起来那么确定,而这种不确定性对性能指标有影响。在本研究中,我们提出了一个框架,用于分析脑室周围 Fazekas 评分预测模型的性能,超越了传统指标。Fazekas 评分是一种序数视觉评分量表,用于评估白质高信号的严重程度,并且已知会受到评分者间差异的影响。虽然最优的 Fazekas 评分预测模型实现了 0.70 的 Matthews 相关系数(MCC),但在不同数据划分和损失函数下性能有所差异,这使得模型能力的解释变得困难。我们的方法不会将模型预测的不确定性解释为孤立的标量值,而是通过不确定性映射将不确定性与其在学习特征表示中的位置关联。这突出了类别边界过渡区域,在这些区域中案例显得更模糊且更容易发生误分类。它还识别了潜在的标签分歧,包括经过专家审查发现与原始参考 Fazekas 评分不一致的低不确定性误分类案例。因此,不确定性映射允许在类别分离和潜在模型-标签分歧的背景下审视模型行为。损失函数的选择也影响了不确定性特征,其中一些模型表现出更清晰的类别分离和在模糊区域中更局部化的不确定性。这些发现表明,在参考标签受模糊或评分者间差异影响时,Fazekas 评分预测的不确定性映射可以支持模型解释和有针对性的数据集审查。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17753 (HTTP 429)

Authors: Susanne Schmid, Johanna Ospel, Richard Frayne, Roberto Souza

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2609.17753.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17753

Published: 2026-09-18T01:29:44.334Z


8. Not All Patches Are Equally Forgettable: Spatially Localized Domain Unlearning in Vision-Language Models

Abstract:Pre-trained vision-language models (VLMs) exhibit strong cross-domain recognition performance even without additional training. However, this robustness can also preserve undesirable domain-specific behavior, as domain-related and semantic information often remain entangled within the learned representation space, making selective domain unlearning challenging. Existing approaches typically address this problem through latent-space disentanglement and prompt- or feature-level interventions, without directly attributing and attenuating individual patch-token contributions. However, here we suggest that rather than uniformly suppressing the full representation, it may be more effective to exploit the spatial structure of vision transformers to localize and suppress patch regions that contribute disproportionately to forget-domain prediction. Patches that strongly influence forget-domain prediction may not be equally important for semantic recognition, suggesting that forgetting should be guided according to the domain contribution of different visual regions. Specifically, we propose a two-stage patch-selective framework that first estimates patch-level domain sensitivity and then selectively attenuates patches whose contribution to forget-domain prediction is stronger than their semantic utility. We evaluate our framework on Office-Home, Mini DomainNet, and DomainNet. Experimental results demonstrate improved forgetting-retention tradeoffs compared to prior methods while improving retained-domain recognition by up to 3.8\%. Additional evaluations under visually overlapping and unseen-domain settings further demonstrate improved robustness under distribution shift.

中文摘要

摘要:预训练的视觉-语言模型(VLMs)即使在没有额外训练的情况下,也表现出强大的跨域识别性能。然而,这种鲁棒性也可能保留不希望存在的域特定行为,因为域相关信息和语义信息通常在学习到的表示空间中纠缠在一起,使得选择性域遗忘具有挑战性。现有方法通常通过潜在空间的解耦以及提示或特征层面的干预来解决这一问题,而不直接归因并减弱单个补丁令牌的贡献。然而,在本文中,我们提出,与其统一抑制整个表示,不如利用视觉变换器的空间结构,定位并抑制对遗忘域预测贡献过大的补丁区域可能更为有效。对遗忘域预测影响较大的补丁不一定对语义识别同样重要,这表明遗忘应根据不同视觉区域的域贡献来引导。具体而言,我们提出了一个两阶段的补丁选择框架:首先估计补丁级别的域敏感性,然后选择性地减弱其对遗忘域预测贡献大于其语义效用的补丁。我们在 Office-Home、Mini DomainNet 和 DomainNet 上评估了我们的框架。实验结果表明,与以往方法相比,在改善遗忘-保留权衡的同时,保留域识别率提高了最多 3.8%。在视觉重叠和未见域设置下的额外评估进一步证明了在分布变化下的鲁棒性改善。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17790 (HTTP 429)

Authors: Akanksha Singh, Vinod K. Kurmi

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2609.17790.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17790

Published: 2026-09-18T01:29:44.334Z


9. AgenTeeth: A Model-Agnostic Framework for Suppressing Hallucination in Frozen Vision-Language Models on Dental X-Rays via Tool Evidence Injection

Abstract:Vision-language models (VLMs) remain largely unreliable on panoramic dental radiographs and can rely on learned anatomical priors rather than evidence in the image. This is particularly problematic for tooth localization and spatial reasoning, and fine-tuned dental VLMs can retain the same spatial biases. We present AgenTeeth, a model-agnostic, tool-augmented framework that grounds frozen VLMs using seven specialized dental vision experts. A question-aware orchestrator selects the relevant tools, whose detections are mapped to FDI tooth numbers or anatomical regions and returned as structured findings together with annotated image overlays. A fresh synthesis call then answers the question using this evidence, without fine-tuning the underlying VLM. On MMOral-OPG-Bench, AgenTeeth improves four backbone VLMs by 12.9-23.0 percentage points over their baselines. Our strongest configuration reaches 65.66% on open-ended VQA, compared with 45.35% for OralGPT-Plus. The advantage also holds at matched scale: a frozen Qwen2.5-VL-7B-Instruct with AgenTeeth reaches 48.11%, exceeding OralGPT-Plus built on the same backbone after supervised fine-tuning and reinforcement learning for tool use. We release the framework, all seven expert models, and a dentist-annotated dataset for alveolar bone-loss detection in panoramic radiographs.

中文摘要

摘要:视觉-语言模型(VLMs)在全景牙科放射片上仍然在很大程度上不可靠,并且可能依赖学习到的解剖先验而非图像中的证据。这在牙齿定位和空间推理方面尤其成问题,即使是经过微调的牙科VLM也可能保留相同的空间偏差。我们提出了AgenTeeth,这是一种模型无关的、工具增强的框架,利用七个专门的牙科视觉专家来定位冻结的VLMs。一个问题感知的协调器选择相关工具,这些工具的检测结果被映射到FDI牙齿编号或解剖区域,并作为结构化发现返回,同时附带注释的图像覆盖层。随后,一个新的综合调用使用这些证据回答问题,而无需对底层VLM进行微调。在MMOral-OPG-Bench上,AgenTeeth使四个主干VLM相比其基线提升了12.9到23.0个百分点。我们最强的配置在开放式VQA上达到65.66%,而OralGPT-Plus为45.35%。该优势在相同规模时也成立:冻结的Qwen2.5-VL-7B-Instruct配合AgenTeeth达到48.11%,超过了在同一主干上通过监督微调和工具使用强化学习构建的OralGPT-Plus。我们发布了该框架、所有七个专家模型,以及用于全景放射片牙槽骨丧失检测的牙医注释数据集。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17800 (HTTP 429)

Authors: Ahmed Rafid, Fariya Ahmed, Rumman Adib, Mehedi Ahamed, Ajwad Abrar, Tareque Mohmud Chowdhury

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2609.17800.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17800

Published: 2026-09-18T01:29:44.334Z


10. Wind on Trees: Testing Physical Grounding in Dynamic 4D Gaussian Splatting

Abstract:Monocular reconstruction of wind-driven vegetation is severely underconstrained: motion along the viewing direction is largely unobservable, a moving canopy offers few reliable correspondences, and nearly the entire scene is dynamic, providing little static reference. Directly-learned deformation fields in 4D Gaussian Splatting therefore optimize photometric consistency rather than recover the motion that produced it. We replace that field with a physically parameterized deformation prior: one damped harmonic oscillator per rigid part, driven by the observed wind and integrated by differentiable RK4, supervised photometrically alone. To test whether such a prior is physically grounded rather than merely well fit, we build a controlled synthetic testbed of three procedurally generated trees spanning an order of magnitude in skeleton complexity, whose per-part natural frequency follows from its own geometry and whose damping ratio is a fixed constant, both held out of training. On it, we measure held-out views, temporal extrapolation, zero-shot transfer to unseen wind speeds, and recovery of the physical parameters themselves. The prior costs appearance fidelity on in-distribution views and extrapolates markedly better outside the training window and the training wind, while parameter recovery is far weaker than it first appears: frequency recovery survives an untrained null control on only the sparsest of the three trees, and damping is not recovered at all.

中文摘要

摘要:风驱动植被的单目重建严重受限:沿视线方向的运动几乎无法观测,移动的树冠提供的可靠对应点很少,并且几乎整个场景都是动态的,几乎没有静态参考。因此,直接学习的四维高斯点云形变场优化的是光度一致性,而不是恢复产生它的运动。我们用一个物理参数化的形变先验来替代该场:每个刚性部分一个阻尼谐振子,由观测到的风驱动,并通过可微分的 RK4 积分,仅通过光度监督。为了测试这种先验是否有物理基础而不仅仅是拟合良好,我们建立了一个受控的合成测试平台,包括三棵程序生成的树,骨架复杂度相差一个数量级,每一部分的固有频率由其自身几何决定,阻尼比为固定常数,这些参数在训练中未被使用。在此平台上,我们测量了保留视点、时间外推、对未见过风速的零-shot 转移以及物理参数的恢复。该先验在分布内视点上牺牲了外观保真度,但在训练窗口和训练风速之外的外推效果明显更好,而参数恢复远比表面看到的要弱:频率恢复在未训练的空控制实验中仅在三棵树中最稀疏的情况下存活,阻尼则完全无法恢复。

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.17810 (HTTP 429)

Authors: Weiying Chen, Edmond Lou

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2609.17810.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.17810

Published: 2026-09-18T01:29:44.334Z