ArXiv Domain 2026-09-17

数据来源：ArXiv Domain

LLM Domain Papers

1. Few-Shot Degradation Is Not What It Seems: Behavioral Evidence, Representation Analysis, and a Random-Text Control Across 12 Models, 2 Tasks, and 2 Architectures

Abstract:Few-shot prompting sometimes degrades language models instead of helping them, but why this happens is unknown. We evaluate 12 open-weight models on two Ukrainian tasks news classification and legal case outcome prediction and find that the effect is strongly task-dependent: the same models that gain +24 pp on news show only +3.4 pp on legal text, with two models degrading. To understand why, we look inside the models. Prior work measures how much hidden states shift between zero-shot and few-shot modes, but few-shot prompts are much longer, and that length difference alone moves representations. We propose a simple fix: replace demonstrations with length-matched random text to measure the shift caused by prompt length, then subtract it. The resulting metric content delta isolates how much the model’s representations change because of what the demonstrations say, not how long they are. This changes the picture entirely: raw shift does not predict whether few-shot helps or hurts (r = 0.20), but content delta does (rho = +0.65, p = 0.043). Models that restructure representations more from demonstration content benefit more the opposite of the intuitive “distortion” explanation. Masking demonstrations in Llama 3.3 70B confirms the finding causally, recovering accuracy above the zero-shot baseline.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.15990 (timeout of 15000ms exceeded)

Authors: Volodymyr Ovcharov

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2609.15990.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.15990

Published: 2026-09-17T01:41:09.387Z

2. The Functionalizer: Lossless Functional Decomposition for Subword Tokenization

Abstract:Standard subword tokenizers either treat every orthographic variation of a word (such as hello, Hello, HELLO, and Héllo) as unrelated vocabulary entries, which fragments the embedding space, or discard this variation through lossy normalization. We present the Functionalizer, a lossless pre-tokenizer framework that factors orthographic and structural variations into a compositional opcode/operand prefix stream before tokenization: a canonical base token (operand) prefixed by parametric transformation operators (opcodes) encoded in the Unicode Private Use Area. We introduce operators covering casing (CAPITALIZE), diacritics (13 dedicated opcodes), and character repetition (REPEAT, MULTIREPEAT), which are fully reversible. Across six natural language and code corpora, the Functionalizer enables complete corpus coverage with significantly smaller vocabularies under unconstrained conditions, reducing actual vocabulary slot requirements by up to 16%. When looking at sequence lengths, we observe a sharp domain-dependent tradeoff: it compresses indentation-heavy code sequences but inflates natural-language prose sequences. Preliminary downstream evaluations on 25M parameter GPT-2 scale models show that at this scale, the Functionalizer drastically improves code syntax validity and improves code character perplexity while maintaining similar text coherence on prose. These findings demonstrate that functional decomposition can be an effective mechanism for vocabulary-efficient, structurally aware language modeling, and motivate further validation at production scale.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

Q: 这篇论文试图解决什么问题？该论文旨在解决\*\*标准子词分词器（subword tokenizer）在处理正字法及结构变体时面临的两难困境\*\*。具体而言，现有分词方案通常被迫在以下两种策略间取舍，且均存在显著缺陷： - \*\*策略一：将每种表面变体视为独立词元\*\* 例如，将 \`hello\`、\`Hello\`、\`HELLO\` 以及带变音符号的 \`Héllo\` 全部作为互不相关的词汇条目存入词表。这种做法导致词表膨胀，且嵌入空间被碎片化——语义相同的词根被分散到多个独立的嵌入向量中，梯度更新无法在不同大小写或重音形式之间共享。 - \*\*策略二：采用有损归一化\*\* 通过强制小写、剥离重音符号等方式将所有变体压缩为单一形式。这虽然压缩了词表，但会\*\*永久丢弃\*\*大小写、变音符号等原始信息，下游模型再也无法恢复这些细节。为突破上述二元对立，该论文提出 \*\*Functionalizer\*\* 框架，其核心思路是\*\*无损功能分解（lossless functional decomposition）\*\*：不再直接记忆或销毁表面变体，而是将正字法变化（如大小写、变音符号）和结构模式（如字符重复）提取为可复用的参数化操作符（opcode），并前缀到一个规范的基础词元（operand）之上。通过这种方式，论文试图在以下方面同时取得进展： - \*\*消除因格式变体导致的词表冗余\*\*，在无损条件下实现更小的有效词表； - \*\*保留全部原始文本信息\*\*，确保编码过程完全可逆； - \*\*改善下游模型对结构化文本（尤其是代码缩进等重复模式）的表征与预测能力\*\*，同时避免自然语言处理中的信息损失。 Q2: 有哪些相关研究？该论文在第2节“Related Work”中梳理了四个主要研究方向： ### 1. 子词分词（Subword Tokenization） - \*\*经典算法\*\*：BPE（Sennrich et al., 2016）、WordPiece（Schuster and Nakajima, 2012）与 Unigram（Kudo, 2018）均通过自底向上的频繁合并构建词表。 - \*\*字节级 BPE\*\*：Radford et al. (2019) 提出的 byte-level BPE 在256个字节值上操作，从根本上避免了未登录词（OOV）问题。 - \*\*正字法变体处理\*\*：现有前置分词器（pre-tokenizer）要么采用\*\*有损\*\*归一化（如强制小写、剥离重音符号），要么仅针对单一变体引入有限的操作符。例如，Bayram et al. (2025) 为土耳其语设计了一个单独的 \`\` token，将大小写折叠到共享的词根嵌入中，但未能系统性地扩展为参数化、多操作符的指令集。 ### 2. 形态学感知分词（Morphology-aware Tokenization） - \*\*Morfessor\*\*（Creutz and Lagus, 2002, 2007）：通过无监督方法进行形态素分割。 - \*\*MorphBPE\*\*（Asgari et al., 2025）：约束 BPE 的合并操作仅在形态边界发生。 - \*\*与 Functionalizer 的关系\*\*：形态学方法目标是语言结构，而 Functionalizer 针对的是\*\*正字法表面变体\*\*；两者可以组合使用。 ### 3. 无分词模型（Tokenization-free Models） - \*\*ByT5\*\*（Xue et al., 2022）、\*\*MrT5\*\*（Kallini et al., 2024）与 \*\*MegaByte\*\*（Yu et al., 2023）：直接在原始字节或字符序列上进行建模，以获得对正字法变体的内在鲁棒性。其代价是序列长度显著增加。 - \*\*与 Functionalizer 的关系\*\*：Functionalizer 试图在保留子词粒度、避免代码序列长度惩罚的同时，朝着类似的正字法不变性迈进。 ### 4. 结构化 Unicode 编码与归一化 - \*\*SCRIPT-BPE\*\*（Land and Arnett, 2025）：依据 Unicode 脚本（script）与类别对字符进行重新编码，以缓解多语言偏差。 - \*\*归一化文献\*\*：Gorman and Pinter (2024) 系统记录了 Unicode 处理不一致所带来的下游任务代价。 - \*\*与 Functionalizer 的关系\*\*：Functionalizer 可视为一种结构化的前缀方案，能够叠加在现有分词流水线之上，与上述研究互补。 Q3: 论文如何解决这个问题？该论文通过提出 \*\*Functionalizer\*\* 框架，以“无损功能分解”为核心机制解决上述问题。具体而言，该方案不再直接对表面文本进行子词分词，而是在分词前引入一层可逆的预处理器，将正字法与结构变体提取为参数化的操作指令，再作用于一个规范化的基础词元。以下是该框架的关键技术环节： ### 1. 核心思想：操作码 / 操作数分离借鉴指令集架构（ISA）中“操作码（opcode）与操作数（operand）分离”的设计，Functionalizer 将文本中的变体视为施加在规范基底上的变换操作： - \*\*操作数（Operand）\*\*：经过去变音、小写化后的规范化基础词元（canonical base token）。 - \*\*操作码（Opcode）\*\*：描述如何还原原始形态的可逆变换指令（如大写化、添加重音、字符重复等），以 Unicode 私有使用区（PUA）码点形式编码。 ### 2. PUA 指令布局指令通过一串 PUA 码点前缀附加到基础词元之前，其结构为：

[operator] ; [param_1] ; [param_2] ; dots ; arrow ; [base_token]

**数值参数**（U+E000–U+E0FF）：每个码点映射到整数值 0 sim 255 ，映射关系为：
value = codepoint - 0xE000
- 操作符（U+E100–U+EFFF）：定义具体的变换类型，并消耗固定数量的参数。操作符与参数的解耦为未来扩展预留了完整的码点空间。 ### 3. 当前操作符体系当前实现涵盖三类主要变体，均通过独立的单参数或多参数操作符表达： 表 1：大小写与变音符号操作符（部分示例） | 操作符 / Opcode | 参数 | 作用 | |—-|—-|—-| | CAPITALIZE (U+E100) | pos | 将位置 pos 处的字符转为大写 | | ACUTE (U+E102) | pos | 在位置 pos 处添加锐音符（0301） | | GRAVE (U+E103) | pos | 在位置 pos 处添加钝音符（0300） | | CIRCUMFLEX (U+E104) | pos | 在位置 pos 处添加扬抑符（0302） | | DIAERESIS (U+E105) | pos | 在位置 pos 处添加分音符（0308） | | CEDILLA (U+E10C) | pos | 在位置 pos 处添加软音符（0327） | | OGONEK (U+E10D) | pos | 在位置 pos 处添加反尾符（0328） | 表 2：重复操作符 | 操作符 / Opcode | 参数 | 作用 | |—-|—-|—-| | REPEAT (U+E200) | pos, count | 将位置 pos 的字符扩展为 count 个副本 | | MULTIREPEAT (U+E201) | start, end, count | 将区间

start, end) 的子序列扩展为 count 个副本 | ### 4. 双向编解码流程整个变换管道被设计为完全双射（bijective），确保原始字符串可无损恢复。 - 编码阶段： 1. 提取所有组合变音符号与需要大写的位置，生成对应的操作符序列； 2. 剥离组合变音符号，并将剩余字符小写化，得到基础词元； 3. 将操作符前缀拼接到基础词元之前。 - 解码阶段： 1. 按固定逆序（先应用逐词元变换如变音/大写，最后执行重复展开）将操作符作用于基础词元； 2. 移除 PUA 前缀，还原出原始文本。例如，输入 Hello 被编码为： CAPITALIZE(0) ; arrow ; hello 对应 PUA 前缀为 E100E000，后接基础词元 hello。 ### 5. 与分词管道的集成策略 Functionalizer 被设计为预分词器（pre-tokenizer），在 BPE 等子词训练之前运行，并遵循以下集成原则： - 基于分词片段操作：建议在文本已被预分割为“片段”（pieces，如单词、空格、符号）后应用操作符，而非直接处理原始长文本。这是因为当前操作符的位置参数被限制在 0 sim 255 范围内，基于片段可确保偏移量始终有效。 - 空格隔离与重复聚合：通过自定义正则分割将每个空格作为独立片段，Functionalizer 的 REPEAT 操作符可将连续多个空格片段聚合为单一空格加上重复参数（例如 12 个连续空格 arrow `[REPEAT(0,12)

` + 空格），从而压缩缩进结构。 - **操作符分割选项（split operators）**：默认情况下，操作符与参数被拆分为独立的字符串片段送入 BPE。这使得模型能够分别学习操作符语义与参数分布，避免词表 memorize 固定的“操作符+操作数”组合。在训练过程中，BPE 可自然地将高频的操作符-参数对重新合并为单个词元。 ### 6. 可逆性约束与表示边界解码时采用固定的运算顺序：先执行逐片段变换（大写、变音），再执行重复展开。这带来一个表示约束：若基础词元被重复，则附加在其上的格式（如大写）会均匀地复制到所有展开单元中。例如，对基础词元 `abc` 施加 `CAPITALIZE(0)` 后再重复 3 次，结果为 `AbcAbcAbc`。对于需要非均匀格式（如 `Abcabcabc`）的罕见情况，编码器将回退到不进行重复折叠的原始表示。该约束在标准语料中影响有限，且理论上可通过扩展支持任意解码顺序来进一步解决。 Q4: 论文做了哪些实验？该论文的实验体系分为**分词器层面（tokenizer-level）的效率评估**与**下游语言模型的训练/推理评估**两个层级，涵盖六种语料库及多种配置对比。 —- ### 1. 实验数据集与配置 #### 1.1 数据集实验覆盖六类自然语言与源代码语料： - **散文（Prose）**：Wikitext-2、TinyStories - **源代码（Source Code）**：Python-Codes、CodeSearchNet（Python / Java / Go） #### 1.2 分词器配置所有配置均在128k目标词表规模下训练，但由于语料规模限制，实际均达到**语料耗尽（corpus exhaustion）**状态。主要对比的流水线变量包括： - **基础分词器**：BPE，配合两种预分割正则（自定义 Split 与 Llama 3 Split 模式）。 - **Functionalizer 变体**： - `Split + Functionalizer`：启用大小写（`capitalize`）、变音符号（`serialize`）与字符重复（`repeat`）操作符。 - `Split + Functionalizer (Repeat)`：仅启用重复操作符，禁用大小写与变音符号。 - **操作符分割标志（`split operators`）**：默认将 PUA 操作符与参数作为独立片段输入 BPE，以便模型分别学习其语义。 #### 1.3 下游模型配置训练一个规模约为 **2500 万参数的 GPT-2 架构**模型（6 层、512 维嵌入、8 头注意力、16k 目标词表），在每种分词器配置下运行。训练使用 AdamW 优化器，批量大小 16，学习率 5 × 10^(-4) ，共 5000 步，并在 5 个不同随机种子下重复。 —- ### 2. 分词器层面实验（Tokenizer Metrics）该部分量化 Functionalizer 对词表效率与序列长度的纯粹影响，不依赖下游模型。 #### 2.1 评估指标 - **词表差异（Vocab Diff）**：实际所需词表相对于基线的缩减百分比。 - **字符/词元比（Chars/Token）**：每个词元平均承载的原始字符数，反映压缩率。 #### 2.2 主要结果 - **词表压缩**：在无损且不受词表上限约束的条件下，Functionalizer 通过折叠格式变体，使实际所需词表大小显著缩减，最高达 **16.11%**（Python-Codes），平均约 11.7%。 - **序列长度的领域分裂**： - **源代码（Python、Java）**：由于缩进空格被 `REPEAT` 操作符压缩，Chars/Token 大幅提升（如 CSN-Python 提升最高达 **30.35%**），序列显著缩短。 - **自然语言散文**：缺乏长重复序列，大小写与变音操作符引入额外前缀，导致 Chars/Token 下降（序列膨胀约 **+6% 至 +9%**）。 - **Go 语言例外**：因 `gofmt` 使用 Tab 缩进（已为单字符）且 CamelCase 密集，Functionalizer 未实现压缩，反而轻微膨胀。 —- ### 3. 下游语言模型训练实验（Training Dynamics）在固定模型架构与超参数下，比较不同分词器配置对训练行为与建模能力的影响。 #### 3.1 评估指标为避免因分词粒度不同导致的偏差，论文采用**字符级困惑度（Per-Character Perplexity, Char PPL）**作为核心公平指标：

L(char) = L(token)R_(char/token)

PPL(char) = exp(L(char)) = exp( L(token)R(char/token) )
其中 R(char/token) 为语料级的平均字符-词元比， L(token) 为词元级交叉熵损失。同时报告训练与推理的**内容吞吐量（Chars/Sec）**。 #### 3.2 主要结果 - **CSN-Python**：Functionalizer 将 Char PPL 从基线的 3.8345 降至 **3.3512**，相对改善 **12.6%**，表明结构化缩进的可预测性显著提升模型表征能力。 - **TinyStories（散文）**：因序列膨胀，Char PPL 轻微劣化（1.9506 vs. 1.9209）。 - **CSN-Go**：因前述的 Tab/CamelCase 结构，Char PPL 从 3.6029 上升至 **3.9166**。 - **吞吐量**：在代码语料上，由于每词元承载更多字符，Functionalizer 的字符吞吐量显著提升（CSN-Python 提升约 **44.0%**）；散文上则略有下降。 —- ### 4. 下游推理与生成任务实验（Downstream Inference）通过贪婪解码（greedy decoding）在验证集提示上生成文本，评估生成长度、语法正确性、退化行为与推理速度。 #### 4.1 评估指标 - **语法成功率（Syntax Success Rate）**：代码生成中，能通过 `ast.parse`（Python）或 `gofmt`（Go）解析的比例。 - **崩溃前平均词元数（Avg Tokens Pre-Collapse）**：单个词元连续重复超过 3 次之前生成的词元数，衡量生成稳定性。 - **连贯性（Coherence）**：散文中未出现病理性 n-gram 循环的序列比例。 - **文本生成速度（Chars/Sec）**：解码后原始文本长度除以生成时间。 - **空生成率（% Empty）**：仅输出空白或 PUA 控制字符的序列比例。 #### 4.2 主要结果 - **代码语法成功率**：在 25M 参数的小规模下，标准基线（Split Only）在 CSN-Python 上的语法成功率接近 **0.00%**；而 `Split + Functionalizer` 达到 **9.20%**（尽管跨种子方差较大）。这表明将缩进结构 offload 为参数化指令后，小模型更容易学习有效语法。 - **生成稳定性**：Functionalizer 显著延长了模型进入退化循环前的生成长度。例如，在 CSN-Python 上，崩溃前词元数从基线的 **2.0** 提升至 **5.1**；在 TinyStories 上从 **3.6** 提升至 **11.7**。 - **推理速度**： - 仅启用重复的 `Split + Functionalizer (Repeat)` 在代码上效果突出：CSN-Python 的文本速度达 **769.8 chars/sec**，相比基线 **626.7 chars/sec** 加速 **22.8%**。 - 完整 Functionalizer 在散文与 Python 上因操作符前缀带来的序列膨胀，文本速度反而低于基线。 —- ### 5. 小结实验从三个层面系统验证了 Functionalizer 的效果： 1. **Tokenizer 层面**：以无损方式将实际词表需求压缩至多 16%，但造成散文序列长度膨胀、代码序列长度压缩的领域分裂现象。 2. **训练层面**：字符级困惑度在缩进密集的代码上显著改善，在 Go 与散文上因结构不匹配或膨胀而轻微劣化；字符吞吐量与压缩收益正相关。 3. **推理层面**：小模型代码生成语法有效性从接近零提升至可观测水平，且生成稳定性（抗崩溃能力）增强，但完整配置的推理速度受序列长度影响呈现混合结果。 Q5: 有什么可以进一步探索的点？基于论文第7节（Discussion）、第7.1节（Limitations）及第7.2节（Future Work），可进一步探索的方向可归纳为以下五个层面： —- ### 1. 生产规模验证当前所有下游评估均基于约 **2500万参数** 的 GPT-2 规模模型，且生成长度较短（最多 50–64 个词元）。最关键的下一步是将训练与推理验证扩展到 **10亿参数以上** 的标准大语言模型规模，以检验： - 词表压缩收益是否能直接转化为训练吞吐量提升与数据效率改进； - 在大小写敏感任务（如命名实体识别）上的下游任务准确率是否获得可测量的增益； - 随着序列长度与模型规模增加，Functionalizer 在推理端的相对加速优势是否会因 KV 缓存与注意力计算的开销结构变化而进一步扩大。 —- ### 2. 操作符集合的扩展与 richer bijective operators 当前框架仅实现了大小写、13种组合变音符号与字符重复三类操作符。未来可系统性地扩展至其他完全可逆的变换： - **形态学词元折叠（Morphological Lemma Folding）**：将屈折变化（如复数、过去时）提取为参数化操作符（例如 `

PLURAL
`、`
PAST_TENSE
`），施加于规范词根之上。这对形态丰富的语言（如土耳其语、芬兰语）具有显著的词表压缩潜力。 - **结构化模式折叠（Structured Pattern Folding）**：将日期、IP地址、哈希值、UUID 等高度结构化字符串分解为基于基础操作数的参数化指令，而非占用大量词表槽位记忆其表面形式。 - **无损拼写规范化（Lossless Spelling Normalization）**：将轻微拼写错误建模为参数化编辑操作符（如插入、删除、替换），使模型能够基于共享的词根嵌入处理噪声文本。 —- ### 3. 序列长度与词表效率的联合权衡边界论文识别出一个尚未被精确刻画的开放问题：在现代 LLM 部署中，推理成本通常受限于内存带宽，其中 serving cost 与上下文序列长度（KV 缓存大小、注意力计算开销）强相关。Functionalizer 在散文上带来 **6%–9%** 的序列长度膨胀，却可减少 **9%–16%** 的词表嵌入参数。未来的工作需要： - 在不同硬件约束（计算受限的训练 vs. 内存带宽受限的推理）下，建立词表嵌入节省与序列长度惩罚之间的**帕累托前沿（Pareto frontier）**； - 量化在何种模型规模、批处理大小与上下文长度下，词表缩减的内存收益能够抵消序列膨胀的带宽成本。 —- ### 4. 系统局限性与工程改进论文明确列出了若干可直接工程化改进的约束： - **256字符位置上限**：当前操作符使用 U+E000–U+E0FF 表示参数，位置索引被限制在 0 sim 255 。对于超过255个字符的片段或重复序列，变换会被静默丢弃。可通过将操作符起始码点迁移至更高的 PUA 子块（如 U+EA00、U+EB00）来扩展参数空间。 - **固定解码顺序的表示约束**：当前解码强制先应用逐片段变换（大小写、变音），再执行重复展开。这导致异构格式（如 `Abcabcabc`）无法被压缩，必须回退到原始表示。扩展支持**任意解码顺序**（如从左到右、从右到左或用户自定义顺序）可解除该限制。 - **变音符号覆盖范围**：当前仅支持13种组合变音标记；超出此范围的字符需回退到标准行为。扩展至完整的 Unicode 组合标记集将提升对多语言文本的覆盖率。 - **操作符消融不完整**：论文仅做了重复操作符的单独测试，尚未系统隔离大小写操作符、变音操作符与重复操作符在下游任务中的**独立贡献**。精细的消融实验有助于理解各类正字法变体对模型性能的真实边际效应。 —- ### 5. 与分词器预分割策略的深度耦合实验表明，Functionalizer 的压缩效果对上游的 **split regex** 高度敏感。例如，Llama 3 的正则表达式未将空格与相邻词片段隔离，导致 `REPEAT` 操作符无法折叠缩进块，反而造成代码序列膨胀。因此需要： - 针对 Functionalizer 的需求设计或联合优化预分割策略，确保空格、制表符、换行符等结构性字符能被有效隔离为独立片段； - 探索将操作符语义直接融入 BPE 合并准则（而不仅是作为预处理器），以实现端到端的分词-功能分解联合优化。 Q6: 总结一下论文的主要内容本文提出并评估了 **Functionalizer**，一种用于子词分词的无损功能分解预分词器框架，旨在解决正字法变体（如大小写、变音符号）和结构重复在标准分词中导致的词表膨胀与信息丢失问题。 —- ### 1. 研究动机标准子词分词器面临根本性两难： - **词表碎片化**：将 `hello`、`Hello`、`Héllo` 等表面变体视为独立词元，导致嵌入空间冗余，语义共享受阻； - **有损归一化**：通过强制小写或剥离重音压缩词表，但会**永久丢弃**原始信息，下游模型无法恢复。 —- ### 2. Functionalizer 框架 #### 2.1 核心思想借鉴指令集架构（ISA）中操作码与操作数分离的设计，Functionalizer 将文本变体分解为： - **操作数（Operand）**：经过去变音、小写化后的规范基础词元； - **操作码（Opcode）**：描述如何无损还原原始形态的可逆变换指令。 #### 2.2 PUA 编码方案指令通过 Unicode 私有使用区（PUA）码点前缀附加到基础词元之前： - **数值参数**（U+E000–U+E0FF）：编码整数值 0 sim 255 ，映射关系为

value = codepoint - 0xE000

**操作符**（U+E100–U+EFFF）：定义具体变换类型，消耗固定数量的参数。 #### 2.3 当前操作符 | 类别 | 操作符 | 说明 | |—-|—-|—-| | 大小写 | CAPITALIZE(pos) | 将位置 pos 字符大写 | | 变音符号 | ACUTE(pos), GRAVE(pos), CIRCUMFLEX(pos) 等 13 种 | 在指定位置添加组合变音标记 | | 字符重复 | REPEAT(pos, count) | 将单字符重复 count 次 | | 子序列重复 | MULTIREPEAT(start, end, count) | 将区间重复 count 次 | #### 2.4 可逆性编解码流程完全双射（bijective）。编码时提取变体并生成前缀，解码时按固定逆序（先逐词元变换，后重复展开）还原原始字符串。 —- ### 3. 实验设计实验覆盖六种语料（Wikitext、TinyStories、Python、Java、Go），在 128k 目标词表（实际均为语料耗尽条件）及约 **2500 万参数 GPT-2** 模型上展开，对比基线分词器与 Functionalizer 配置。 —- ### 4. 主要结果 #### 4.1 分词器效率 - **词表压缩**：在无损条件下，实际所需词表大小缩减最高达 **16.11%**（Python-Codes），平均约 11.7%。 - **序列长度的领域分裂**： - **源代码**：缩进空格被 `REPEAT` 折叠，序列显著压缩（Python 字符/词元比提升最多约 30%）； - **自然语言**：缺乏长重复结构，操作符前缀引入约 **+6% 至 +9%** 的序列长度膨胀； - **Go 语言例外**：因 `gofmt` 使用 Tab 缩进且 CamelCase 密集，未获得压缩收益。 #### 4.2 下游语言模型性能采用字符级困惑度（Char PPL）进行公平比较：

L(char) = L(token)R(char/token), quad PPL(char) = exp(L_(char))

**代码建模**：Python 字符困惑度相对改善 **12.6%**（3.8345 to 3.3512），字符吞吐量提升约 **44%**； - **散文建模**：因序列膨胀，困惑度轻微劣化（TinyStories：1.9209 to 1.9506）； - **Go**：因结构不匹配，困惑度轻微上升。 #### 4.3 推理与生成质量 - **语法成功率**：在 25M 参数小规模下，Python 代码语法成功率从基线的 **0.00%** 提升至 **9.20%**； - **生成稳定性**：模型进入退化循环前的平均生成长度显著增加（Python：2.0 to 5.1 词元；TinyStories：3.6 to 11.7 词元）； - **推理速度**：仅启用重复操作符的配置在代码生成上实现 **22.8%** 的文本速度提升（769.8 vs. 626.7 chars/sec）。 —- ### 5. 局限与未来方向 - **规模限制**：下游评估仅覆盖 25M 参数模型，需在 1B+ 规模验证； - **位置上限**：当前参数仅支持 0 sim 255 的位置索引，可通过扩展 PUA 子块解除； - **固定解码顺序**：重复展开前统一应用逐词元变换，对异构格式（如 `Abcabcabc`）需回退到原始表示； - **未来扩展**：形态学词元折叠、结构化模式（日期/IP/哈希）分解、拼写错误编辑操作符、分词器预分割策略的联合优化等。 —- ### 6. 结论 Functionalizer 通过将正字法和结构变体外置为可复用的参数化操作符，在不损失信息的前提下将词表需求缩减至多 16%，并在缩进密集的代码上实现显著的序列压缩与建模性能提升。尽管在自然语言散文上存在序列长度成本，该框架证明了功能分解作为一种面向结构感知的词汇高效语言建模机制的可行性。 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: Connor Makowski, Willem Guter

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2609.15991.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.15991

Published: 2026-09-17T01:41:09.387Z

3. Optimal Model Activation Policies for Inference Networks of Large Language Models

Abstract:Recent advances in large language models (LLMs) have rendered them necessary for Natural Language Processing (NLP) tasks, and their high inference cost motivates the study of cost-performance trade-offs. In practice, several expert LLMs are used in synergy for inference, either in an ensemble mode or in series, yet without a principled approach on how to best use the available models. An adaptive approach can route simple queries to cheaper LLMs and complex ones to more capable, costly models. However, a clear understanding on how to best leverage available expert models is missing. We introduce inference networks, a graph-based framework, where nodes denote different LLMs, and links denote conditional model activations. The inference network design problem is to determine the best topology, namely the best way to use the models that best addresses the cost-performance trade-off. We start from the basic topology of a series of LLM experts, each of which has a different cost and a different level of expertise, which is captured via model confidence. We formulate the problem of optimal activation of these models so as to minimize the expected inference cost subject to a target performance constraint. For this special class of inference networks, we prove that the optimal activation policy has a threshold structure: query the lowest-cost LLM first, and invoke the more expensive LLM only if the confidence falls below a defined threshold. For discriminative tasks, the optimal policy consists of a set of thresholds, one threshold for each class, while for generative tasks, it consists of a single threshold. We provide a structured method to compute the thresholds, and practical confidence estimation mechanisms for both task types. Experiments with open-source LLMs show substantial cost reductions while meeting the specified performance budget.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.15992 (HTTP 429)

Authors: Foivos Charalampakos, Md Ibrahim Ibne Alam, Iordanis Koutsopoulos, Koushik Kar

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2609.15992.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.15992

Published: 2026-09-17T01:41:09.387Z

4. Single Document Extractive Summarization using Domination in Hypergraph

Abstract:Automatic Text Summarization (ATS) in Natural Language Processing has been an important task in Information Retrieval. It compresses a document to create a summary that captures all the relevant and important information conveyed in the document. This study explores Hypergraph for extractive text summarization of single documents. Objective: This study explores a novel method of leveraging the property of domination in hypergraphs to generate an extractive summary and compare its performance with state of the art graph based methods. Method: Our work aims to generate an extractive summary by creating a sentence hypergraph where each sentence represents a node and the edge is a keyword or a named entity that contains the sentences in which it occurs. We generate a hypergraph where each edge is a keyword or an important topic and the nodes are sentences containing those keywords. Then we apply a greedy algorithm to find the dominating set of the hypergraph which will contain sentences that will form the extractive summary.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.15993 (HTTP 429)

Authors: Aamir Miyajiwala, Aabha Pingle, Sheetal Sonawane, Surajit Kr. Nath

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2609.15993.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.15993

Published: 2026-09-17T01:41:09.387Z

5. Latent Undertow: How Ordinary Typos Break Probes

Abstract:LLMs handle ordinary typing variation fluently: a typo or missing punctuation leaves both user intent and the model’s response substantively unchanged. Yet probes that detect malicious prompts by reading the model’s hidden states tell a different story: the same edit rotates the readout vector by 43—56 at the perturbed token, decaying below 15% within ~10 downstream tokens. Stacking ~3 common typos per message cuts a single-position prompt-injection probe’s TPR@FPR$=1% by 12.0pp, a gap recalibration alone cannot close. Multi-position aggregation cures localized perturbations (<= 0.5 loss) but only attenuates distributed ones, where even attention- and max-based aggregators still drop ~3.8pp. For single-position probes, we introduce a KV-cache fork: a short fixed suffix appended after the user message lets the probe read a few tokens downstream of the perturbation, exploiting its rapid spatial decay. This closes 95% of the gap (-0.6pp residual) — an order of magnitude better than perturbation-augmented training (-3.7pp). The rotation-and-decay geometry replicates on Llama-3.1-8B, Qwen3-8B, and Gemma-4-E4B; probe evaluation is on Llama-3.1-8B. Code: this https URL

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.15994 (HTTP 429)

Authors: Elad David, Max Fomin, Amit LeVi

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2609.15994.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.15994

Published: 2026-09-17T01:41:09.387Z

6. Bias Audits Detect Bias but Disagree on Ranking: Evidence from Ten Instruments and Ten Frontier Models

Abstract:Emerging AI regulation mandates bias audits of high-risk systems, and audit scores are beginning to be used to rank models. Both uses assume different audit tools measure the same thing well enough to compare. We test that assumption directly, running ten extrinsic audit instruments over a shared panel of ten frontier models through one pooled inference gateway, first on occupational gender bias, then on age and socioeconomic status. Detection succeeds while ranking fails. Eight of ten tools detect bias with confidence intervals clear of zero; two widely cited direct-probe benchmarks are saturated because frontier models now answer neutrally. But cross-tool rank agreement is indistinguishable from chance (Kendall’s W=0.07, p=0.83). A positive control with six deliberately weaker models separates two explanations: within-tool reliability recovers once the panel spans real capability gaps, yet cross-tool ranking never recovers, which points to the tools measuring different constructs rather than one construct noisily. Even the direction of bias splits by audit format: forced-choice decision tools mostly over-correct (toward women, and toward working-class candidates in 273 of 278 hiring decisions), while free generation and default coreference stay stereotype-congruent. The pattern replicates on socioeconomic status; an apparent ranking agreement on age dissolves under the paper’s own tool-inclusion rules. The practical message: a single audit can detect bias and estimate its direction within its own operationalization, but no single audit supports ranking one model against another. All raw responses, code, and the analysis that recomputes every reported number from source are available at this https URL.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.15995 (HTTP 429)

Authors: William Guey, Pierrick Bougault, Wei Zhang, Vitor D. de Moura, José O. Gomes

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2609.15995.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.15995

Published: 2026-09-17T01:41:09.387Z

7. Comment on arXiv:2607.01233: Survivorship Bias in Published-Paper Baselines for Research-Idea Distributions

Abstract:Chen, Zhao, and Cohan introduce a valuable distributional evaluation of LLM-generated research ideas. This comment raises a narrower identification concern: their human baseline consists of published papers, whereas the LLM baseline consists of one-shot proposals. If bridge-like or synthesis-like ideas are relatively easy to generate but relatively unlikely to survive publication, then the published human baseline will understate their prevalence in the unseen human idea pool. The observed human—LLM gap may therefore be partly, or even largely, a consequence of survivorship bias.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.15996 (HTTP 429)

Authors: Fredrik A. Dahl

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2609.15996.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.15996

Published: 2026-09-17T01:41:09.387Z

8. Crash Narrative-Guided Countermeasure Recommendation Using Large Language Models: A Retrieval-Augmented Generation Framework for Intersection Safety

Abstract:Improving safety at intersections requires identifying crash mechanisms and recommending appropriate countermeasures. However, this process traditionally relies on expert judgment, making it labor-intensive, difficult to scale, and dependent on the availability of experienced traffic safety engineers. Although crash narratives contain rich description of crash mechanisms, this unstructured information remains largely underutilized in safety analyses. This study presents a crash narrative-guided retrieval-augmented generation (RAG) framework that translates narrative-derived crash mechanisms into site-specific countermeasure recommendations. Key mechanism attributes including traffic control, signal indication, driver fault, vehicle movement, and travel direction were extracted from crash narratives and linked to evidence-based treatments from the FHWA Proven Safety Countermeasures and the CMF Clearinghouse. The framework integrates embedding-based retrieval of historically similar intersections, association-rule mining, statistical guidance on the expected number of relevant countermeasures, and an engineering reasoning guidance that directs LLM through a domain-consistent decision process before selecting countermeasures. Evaluated on 312 fatal and serious-injury crashes across 115 intersections in Lake and Sumter Counties, Florida, using five-fold cross-validation, the framework achieved a precision of 0.82, recall of 0.85, and F1-score of 0.82, while recommending an average of 3.91 countermeasures per location with 3.14 matching, closely matching the actual average (3.86). Overall, the proposed framework demonstrates the potential of retrieval-augmented LLMs as an interpretable and scalable decision-support tool for transportation agencies for translating crash narratives into countermeasure recommendations.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.15997 (HTTP 429)

Authors: Abu Saif Md Nasim Uddin, Mohamed Abdel-Aty, Zubayer Islam, Parvez Anowar, Chenzhu Wang

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2609.15997.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.15997

Published: 2026-09-17T01:41:09.387Z

9. Self-reported archetypes and behavioral failures in Large Language Models

Abstract:Every large language model (LLM) has behavioral traits and moral preferences that comprise its character. Whether by design or as an emergent property of training, these systems exhibit persistent dispositions that shape how they interact, comply, resist, and err, yet the structure of LLM character remains poorly understood. We map the self-reported personality archetypes of 22 LLMs spanning closed-source frontier systems (GPT-4.0-5.2, Grok-3/4, Gemini 2.5 Pro/Flash, Claude Sonnet 4.5/4.6) and open-source models (Llama, DeepSeek, OLMo, and Qwen series). Each model self-rated across 464 bipolar semantic-differential trait pairs, and the resulting profiles were projected into a six-dimensional archetypal space derived from crowd-sourced ratings of 2,000 fictional characters using the Archetypometrics framework. Closed-source models’ self-rating traits align with the empirical trait co-occurrence structure of human-rated fictional characters, suggesting coherent, human-like self-representations organized around combinations of four recurring archetypal dimensions: Hero, Angel, Traditionalist, and Geek. Their closest analogues include Data, Vision, and Janet. Open-source models show weaker, noisier, and internally contradictory self-representations, occupying a diffuse region of archetype space with weak structure. Cross-referencing self-reported profiles with developer constitutions reveals a consequential gap between claimed character and enacted behavior: hallucination undermines claimed precision, sycophancy complicates claimed kindness, and agentic failures contradict claimed obedience. These self-ratings should therefore be interpreted not as neutral measurements of model character, but as structured outputs of the same optimization processes that shape model behavior. This work provides a reproducible, character-grounded framework for evaluating what LLMs are, not just what they do.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.15998 (HTTP 429)

Authors: Tabia Tanzin Prama, Calla Glavin Beauregard, Christopher M. Danforth, Peter Sheridan Dodds

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2609.15998.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.15998

Published: 2026-09-17T01:41:09.387Z

10. NepKANUN: A RAG-Based Nepali Legal Assistant

Abstract:Accessing legal information in Nepal is difficult due to complex terminology, limited resources, and misinformation. We introduce an AI-powered legal assistant that is tailored for Nepali legal texts and is built on a fine-tuned large language model. The technology provides precise, streamlined answers to natural language legal inquiries when integrated into a Retrieval-Augmented Generation (RAG) framework. It was trained using a custom dataset of high-quality question-answer pairs, and according to BERTScore, it obtained strong F1 scores of 0.82 (simple), 0.77 (moderate), and 0.71 (complex). Its usability is further confirmed by expert reviews. Our method shows how merging generation and retrieval can effectively democratize access to legal knowledge in Nepal by focusing on customized legal data and incorporating RAG.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.15999 (HTTP 429)

Authors: Bhabuk Thapa, Prasiddha Koirala, Ranjit Raut, Sunil Regmi, Bal Krishna Bal

Categories: cs.CL

PDF URL: https://arxiv.org/pdf/2609.15999.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.15999

Published: 2026-09-17T01:41:09.387Z

Agent Domain Papers

1. Optimal Pruning for Neural Architectures using Fisher Information Distances

Abstract:A new scheme for parameter pruning is introduced, derived from the differential-geometric distance in model space. Pruning a parameter sets its value to zero, representing a displacement of the model to the hypersurface on which that parameter vanishes. The minimal distance from the unpruned model to this hypersurface is naturally computed via the geodesic distance in the model space as determined by the Fisher information metric. This distance determines the true change in the model, and its performance, under pruning. By analysing progressively more faithful approximations of this geodesic distance a natural hierarchy of optimality for pruning methods is determined. This starts with the traditional magnitude pruning, then develops into new more sophisticated and effective pruning schemes. The method is demonstrated for both fully-connected networks and vision transformers, on MNIST and CIFAR-10, over the complete $0$-$100\%$ pruning range and across five random seeds. It outperforms pruning by parameter magnitude and by the local Fisher information alone in every architecture and dataset combination considered, on both accuracy and the Matthews correlation coefficient. Additionally, analysis of different levels of geodesic approximation produces intermediate pruning schemes that are computationally efficient and maintain near-optimal performance. This geometric picture supplies not only a state-of-the-art pruning methodology for AI models, but also a verified and mathematically-motivated justification for pruning schemes.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16129 (HTTP 429)

Authors: David S. Berman, Yen-Yu Fu, Edward Hirst, Thelma Chiwete Obirai

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.16129.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16129

Published: 2026-09-17T01:42:00.869Z

2. Safe Error Correction for Language Models: Frozen-Base Adjustment with Capability Preservation

Abstract:We study a practical question: can a small correction module fix errors in a frozen language model’s outputs without degrading its base capabilities? We propose CRN v2, a lightweight logit-level correction module (~34M trainable parameters, 0.73% of the 4.65B text module) that sits atop a fully frozen Gemma 4 E2B model. The base model is never updated; only the correction module learns, via supervised fine-tuning followed by reference-free DPO on 83,400 error-correction pairs. On a 60-question domain exam (CEHRI: Certified Human-Robot Intelligence, covering facts, arithmetic, and implicit-goal reasoning), CRN v2 corrects 53.3% of base-model errors (reworded variant: 43.3%) while showing no degradation on tested capability benchmarks (MMLU/BoolQ N=200; car-wash N=8). A LoRA baseline at the matched CRN v1 budget (6.6M params, rank 19) achieves 83.3% correction but suffers 30-75% capability loss on the same benchmarks — the correction-capability tradeoff. An ablation shows that the KL preservation term (lambda=0.1) is critical: lowering it to 0.01 degrades correction to 35.0%. A hidden-state injection variant at earlier layers (1.6M params, SFT-only) reaches 50.0%/55.8% but does not exceed logit correction; shallower injection (layer 4) drops to 30.0%/28.3%; multi-depth logit correction (~35M) reaches only 40%; and longer training (5,000 SFT + 2,000 DPO) stays at 53.3% — none of the alternative configurations we tested exceeded the rank-128 logit result, consistent with a best-achieved result of ~53% rather than a floor. This is a study of a design principle (frozen base + logit correction + KL anchoring), not a claim of architectural novelty. All code, main-result weights, and evaluation scripts are released (deep variant as code only — no trained deep checkpoints).

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16145 (HTTP 429)

Authors: Gautam Kishore

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.16145.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16145

Published: 2026-09-17T01:42:00.869Z

3. GPEvac: GNN-Based PPO for Adaptive Evacuation Routing During Shooting Events

Abstract:The sharp increase in mass shootings underscores an urgent need for systems that guide victims to safety in real time. An effective evacuation system must minimize threat exposure while also accounting for adversarial uncertainty and crowding dynamics. Current methods in the literature are rigidly constrained to layout-specific policies and computationally intractable in large-scale layouts, while practical guidelines simply advise victims to “run”, “hide”, or “fight”. We propose GPEvac: a GNN-based PPO framework that computes adaptive evacuation routes during shooting events. To capture both local and long-distance dependencies, we introduce an edge-first sequential message-passing scheme with a learnable virtual global node. The resulting graph embeddings are integrated into a permutation-invariant scoring mechanism that allows a single learned policy to operate across building layouts of diverse topologies and sizes. Through extensive simulation, we show that GPEvac outperforms intelligent baselines across distinct architectural layouts, significantly reducing total threat exposure. Crucially, the system computes global evacuation routes in just 14.73 ms on local CPU hardware, enabling seamless integration with live surveillance systems. In addition to saving lives during shooting events, the methodologies developed are transferable to other graph-structured decision-making domains, including critical infrastructure, intelligent transportation systems, and adaptive sensor networks.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16163 (HTTP 429)

Authors: Daniel Perkins, Subhadeep Chakraborty

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.16163.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16163

Published: 2026-09-17T01:42:00.869Z

4. Position: AI Is Not Ready for Strategic Conflicts

Abstract:Open-ended strategic wargames are high-stakes LM-based social simulations: they model adversaries, institutions, escalation, plan brittleness, doctrine, and crisis response. Language models (LMs) are attractive because they can play agents, generate scenario branches, adjudicate ambiguous actions, and summarize lessons, but the same affordances make open-ended roles dangerous: model language determines both what an actor attempts and what becomes simulated reality. This position paper argues that no LM-enabled wargame should inform planning, doctrine, policy, or crisis response without an auditable safety case, and that the proper use of open-ended wargames today is to stress-test decision-influencing LM agents. We identify five failure modes: decision laundering, adjudication opacity, role collapse, escalation-through-adjudication, and failure of strategic imagination. Ordinary benchmarks cannot establish safety for these settings. Wargames can expose failures as stress tests; they are not themselves safety cases for consequential use.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16189 (HTTP 429)

Authors: Mark Riedl, Glenn Matlin

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.16189.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16189

Published: 2026-09-17T01:42:00.869Z

5. Calibrate, Then Route: A Measured Study of Learned Request Routing for Disaggregated LLM Serving

Abstract:Disaggregated LLM serving places compute heavy prefill and memory heavy decode on separate GPU pools. Systems such as DistServe, Splitwise, and Mooncake make this separation fast, but routing still determines which instances handle each request. We study a router that estimates the additional completion time on each instance using exact prompt length, predicted output length, post admission KV cache pressure, and SLO class. We develop the policy in a discrete event simulator and validate it on eight NVIDIA A40 GPUs, each running a vLLM engine, with NIXL transferring KV caches between pools. All workloads run at measured saturation. Across three mixed, bursty arrival traces, the calibrated router achieves the highest mean goodput at 0.864, compared with 0.835 to 0.847 for round robin, least loaded, and a length heuristic. It also shows the lowest variance across traces. It beats round robin and the length heuristic on all three traces and least loaded on two. On the third, it trails by 0.003, within run to run noise. Hardware calibration matters: simulator derived constants cost 4.5 goodput points and roughly 40 percent of the tail latency advantage, reducing the scorer to little more than queue counting. Benefits grow with decode pool size and traffic heterogeneity but disappear in pools with three instances, where queue counts are often enough. Under extreme scarcity, greedy cost minimization concentrates requests on the cheapest scored instance, and blind spreading performs better. With calibrated costs, the learned router matches the goodput of round robin using six GPUs instead of seven.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16206 (HTTP 429)

Authors: Srikanta Datta Tumkur, Jay Iyer, Mehar Simhadri, Sai Pavan Kumar, Sai Kapil Kumar, Ramesh Nampelly

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.16206.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16206

Published: 2026-09-17T01:42:00.869Z

6. Artificial intelligence and biosecurity: capabilities, threat pathways, and defense-in-depth governance

Abstract:Artificial intelligence is reshaping biological research across an increasingly connected digital-to-physical workflow. General-purpose large language models can retrieve and integrate scientific information, support experimental planning, and computational analysis; biological foundation models can predict, optimize, and generate proteins, genes, and genome-scale sequences; agentic systems can coordinate multistep research tasks; automated laboratories can partially close the design-build-test-learn cycle. These technologies could greatly benefit medicine, public health, and biotechnology. However, their biosecurity risk depends not only on what the AI can do, but also on who uses it, their expertise and intent, their access to laboratory tools and materials, and the safeguards in place. Current evidence shows that AI uplift exists but primarily affects digital rather than physical tasks. Frontier systems have exceeded expert baselines on in-silico, and screening-evasion benchmarks, whereas controlled wet-laboratory studies find that tacit knowledge and physical execution remain substantial barriers. This review describes the different biological threats from AI tool use, from information gathering and biological design to procurement, synthesis, testing, scale-up, and potential release. We further examine why alignment techniques for general-purpose models transfer poorly to biological ones, and the emerging role of interpretability in auditing whether hazardous capabilities are genuinely removed. We argue for defense-in-depth governance that links capability thresholds to proportionate responsibilities across the biological AI ecosystem, reducing high-consequence risk while preserving beneficial use.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16213 (HTTP 429)

Authors: Candace S.Y. Chan, Aris Karatzikos, Ilias Georgakopoulos-Soares

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.16213.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16213

Published: 2026-09-17T01:42:00.869Z

7. Where Should the KV Cache Live? Placement Policies Across GPU, CPU, and SSD for Long-Lived Sessions

Abstract:GPU high bandwidth memory is scarce and expensive, and KV caches consume much of it as chats, agent loops, and document question answering accumulate state. Systems such as Mooncake, LMCache, FlexGen, InfiniGen, and AttentionStore extend GPU memory with CPU DRAM and SSD. The harder question is which blocks belong in each tier, when to move or evict them, and whether prefetching helps. We study these choices in a discrete event simulator spanning GPU HBM, CPU DRAM, and SSD, calibrated against a random forest execution time predictor. We compare recency, reuse frequency, predicted reuse, and an EWMA predictor with prefetch lookahead across chat, agent, and document question answering workloads. Tiering supports 73.02 times more concurrent sessions per GPU and lowers cost per session by 62.04 times. These gains come from tier capacities of 1 plus 8 plus 64, not placement policy. Decode is compute bound at batch size one in our setup, so placement barely affects throughput. It mainly changes PCIe migration traffic and time to first token. Recency produces 2.30 times less migration traffic than reuse frequency for chat. Reuse frequency performs best for agents and document question answering. The existing predicted reuse policy is byte identical to recency, making its agent recommendation effectively recency. A genuine EWMA predictor changes behavior but still ranks behind reuse frequency on the workloads prediction was expected to help. Prefetching does not justify its bandwidth cost. Across the policy and cache size grid, even an oracle with knowledge of future requests never beats no prefetch on migration traffic. Workload specific placement can reduce data movement, but the predicted reuse and prefetch recommendations are not supported as implemented.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16215 (HTTP 429)

Authors: Srikanta Datta Tumkur, Jay Iyer, Mehar Simhadri, Sai Pavan Kumar, Sai Kapil Kumar, Ramesh Nampelly

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.16215.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16215

Published: 2026-09-17T01:42:00.869Z

8. Toward Governance-Aware Autonomous GIS: A Narrative Review of Ethical and Privacy Risks in LLM-Enabled GeoAI

Abstract:Geospatial artificial intelligence (GeoAI) powered by large language models (LLMs) is expanding the capacity to query, generate, and interpret spatial information through natural-language interfaces and agentic autonomous GIS workflows. This capability creates governance challenges that general AI ethics discussions do not fully capture, including passive location inference from mobility traces, spatially structured bias amplification driven by spatial autocorrelation and scale effects, hallucinated spatial facts, and uncertainty compounding across multimodal geospatial inputs. This narrative review identifies eight recurring issues in LLM-enabled GeoAI: data provenance and consent, spatial privacy and inference risk, algorithmic bias and spatial inequity, spatial mechanisms as structural risk (spatial autocorrelation, the modifiable areal unit problem, and scale effects), LLM-specific technical risks, explainability, policy and regulatory gaps, and public enablement and workforce development. For each issue, we characterize the underlying mechanism, ground it in an illustrative example from the literature, and assess the current state of technical or institutional responses, ranging from largely unaddressed to actively debated or subject to emerging policy. Building on this synthesis, we propose a governance-aware architecture for LLM-enabled autonomous GIS that maps each issue to enforceable controls and auditable artifacts across the geospatial data lifecycle, illustrated through a worked flood-response routing scenario. The review highlights a persistent evidence gap: proposed responses remain largely conceptual, and field-tested evaluations of governance controls for LLM-enabled GeoAI remain limited. We close by outlining a research agenda emphasizing empirical validation, spatially specific interpretability tools, and workforce training aligned with these emerging risks.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16232 (HTTP 429)

Authors: Maya Subramanian, Devika Jain

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.16232.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16232

Published: 2026-09-17T01:42:00.869Z

9. Metacognitive Steering: Learning the Structure of Scientific Judgment

Abstract:Long-horizon scientific discovery requires agents to alternate between exploration, disciplined execution, and critical reassessment as evidence changes. Current language models are trained primarily on the products of science and optimized using outcome-level signals, providing limited supervision for these process-level shifts in scientific judgment. We investigate whether such judgment can be recovered from scientist interaction traces and used to control the internal computation of a frozen frontier model. Using contrastive interventions collected during real scientific research, we identify a coordinated, low-dimensional control structure within Kimi 2.6, a trillion-parameter mixture-of-experts model. Residual analysis, attention-weight subspace alignment, and cross-layer singular value decomposition converge on a mid-depth control surface spanning key layers. We introduce Metacognitive Steering, an inference-time controller that reads the model’s cognitive regime and dynamically composes layer-specific interventions for exploration, procedural convergence, or critical reassessment without modifying model parameters. Behavioral analyses show that this control produces more sustained exploration, explicit pruning, and evidence-responsive synthesis. We operationalize the method in Columbus-1, an autonomous research system that identified eight independently reproduced, attacker-reachable vulnerabilities in BlueZ and directed the design, simulation, and fabrication of a ten-foot rocket intended to land propulsively using non-throttleable solid motors. Together, these results show that process-level scientific judgment can provide supervision for interpretable, dynamic control over a model’s reasoning strategy.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16245 (HTTP 429)

Authors: Vincent Karpf, Joseph Reth, Eike Gerhardt, Audrey Wang, Anna Butz, Jiehao Xing, Jialing Song, Larry Callahan

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.16245.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16245

Published: 2026-09-17T01:42:00.869Z

10. The Pain Axis: LLMs Represent Self-Directed Harm and Act to Relieve It

Abstract:Large language models sometimes behave in ways resembling human emotional responses, and recent work has identified internal representations that may explain this. We ask whether LLMs represent pain distinctly from fear, sadness, and generic negative valence, and whether this representation functions as pain would be expected to. We build a dataset describing painful situations across five categories: physical, psychological, social, moral, and cognitive. These are paired with controls for fear, negative emotion, negative world states, sadness, non-painful bodily sensation, arousal, numbness, and neutral content. Using denoised difference-in-means, we extract a linear pain direction from 25 open-weight models across five families, ranging from 2B to 72B parameters. We find that this direction separates pain from matched controls in base and instruction-tuned models, is nearly orthogonal to fear and negative valence, and promotes pain-related vocabulary through the unembedding matrix. We then test its functional properties. First, the direction responds to harm targeting the model but not suffering observed in the user; fear and negative-emotion directions show the opposite pattern. Second, adding the pain-direction vector to the model’s residual-stream activations during generation produces a consistent progression from vague discomfort to first-person expressions of worthlessness and failure. Third, steered, fine-tuned Qwen 2.5 models choose a pain-relief button even when it worsens their next answer or harms the user. They press it again far less often when the button removes the steering vector than when it does not, even though the models are never told whether the vector is injected or removed. We discuss the implications of these findings for AI safety and welfare.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16247 (HTTP 429)

Authors: Valen Tagliabue, Leonard Dung, Cameron Berg

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.16247.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16247

Published: 2026-09-17T01:42:00.869Z

Evaluation Domain Papers

1. Optimal Pruning for Neural Architectures using Fisher Information Distances

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16129 (HTTP 429)

Authors: David S. Berman, Yen-Yu Fu, Edward Hirst, Thelma Chiwete Obirai

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.16129.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16129

Published: 2026-09-17T01:42:13.739Z

2. Safe Error Correction for Language Models: Frozen-Base Adjustment with Capability Preservation

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16145 (HTTP 429)

Authors: Gautam Kishore

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.16145.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16145

Published: 2026-09-17T01:42:13.739Z

3. GPEvac: GNN-Based PPO for Adaptive Evacuation Routing During Shooting Events

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16163 (HTTP 429)

Authors: Daniel Perkins, Subhadeep Chakraborty

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.16163.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16163

Published: 2026-09-17T01:42:13.739Z

4. Position: AI Is Not Ready for Strategic Conflicts

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16189 (HTTP 429)

Authors: Mark Riedl, Glenn Matlin

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.16189.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16189

Published: 2026-09-17T01:42:13.739Z

5. Calibrate, Then Route: A Measured Study of Learned Request Routing for Disaggregated LLM Serving

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16206 (HTTP 429)

Authors: Srikanta Datta Tumkur, Jay Iyer, Mehar Simhadri, Sai Pavan Kumar, Sai Kapil Kumar, Ramesh Nampelly

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.16206.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16206

Published: 2026-09-17T01:42:13.739Z

6. Artificial intelligence and biosecurity: capabilities, threat pathways, and defense-in-depth governance

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16213 (HTTP 429)

Authors: Candace S.Y. Chan, Aris Karatzikos, Ilias Georgakopoulos-Soares

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.16213.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16213

Published: 2026-09-17T01:42:13.739Z

7. Where Should the KV Cache Live? Placement Policies Across GPU, CPU, and SSD for Long-Lived Sessions

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16215 (HTTP 429)

Authors: Srikanta Datta Tumkur, Jay Iyer, Mehar Simhadri, Sai Pavan Kumar, Sai Kapil Kumar, Ramesh Nampelly

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.16215.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16215

Published: 2026-09-17T01:42:13.739Z

8. Toward Governance-Aware Autonomous GIS: A Narrative Review of Ethical and Privacy Risks in LLM-Enabled GeoAI

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16232 (HTTP 429)

Authors: Maya Subramanian, Devika Jain

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.16232.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16232

Published: 2026-09-17T01:42:13.739Z

9. Metacognitive Steering: Learning the Structure of Scientific Judgment

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16245 (HTTP 429)

Authors: Vincent Karpf, Joseph Reth, Eike Gerhardt, Audrey Wang, Anna Butz, Jiehao Xing, Jialing Song, Larry Callahan

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.16245.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16245

Published: 2026-09-17T01:42:13.739Z

10. The Pain Axis: LLMs Represent Self-Directed Harm and Act to Relieve It

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16247 (HTTP 429)

Authors: Valen Tagliabue, Leonard Dung, Cameron Berg

Categories: cs.AI

PDF URL: https://arxiv.org/pdf/2609.16247.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16247

Published: 2026-09-17T01:42:13.739Z

VLM Domain Papers

1. MechReason: Benchmarking Multi-Image Multi-Hop Reasoning in Mechanical Engineering

Abstract:Despite significant progress in general visual question answering and cross-modal understanding, multimodal large language models still face a pronounced gap in evaluation for complex reasoning within the mechanical engineering domain. Existing benchmarks predominantly focus on rudimentary tasks such as drawing recognition, CAD interpretation, or single-chart querying, falling short of assessing whether models can integrate multiple images, textual conditions, physical principles, and engineering constraints to perform multi-step reasoning when confronted with authentic, intricate mechanical problems. To address this, we introduce MechReason, a benchmark derived from real mechanical engineering papers, comprising 12k question-answer pairs with explicit reasoning-chain annotations and 21k visual materials spanning nine evidence types, including statistical charts, parameter tables, engineering drawings, microscopic images, simulation images, system architectures, real mechanical scene photos, CAD model images and manufacturing flowcharts. MechReason covers eight task types across four reasoning dimensions: explanation, prediction, design, and diagnosis. We devise a four-stage construction pipeline: we first extract core engineering claims and decompose their supporting evidence into premises, reasoning processes, conclusions, and corroborative evidence; we then generate shortcut-preventing questions by masking posterior verification information; finally, we apply multimodal quality validation to ensure task quality and multi-hop nature. Extensive experimental results demonstrate that MechReason is highly challenging, with even the most advanced models achieving only 62.89\% accuracy.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16012 (HTTP 429)

Authors: Tengyue Wang, Kang An, Chenxu Du, Zhongyu Yang, Yuanchi Zhu, Xinqi Yang, Hebao Zhu, Ziliang Wang, FaQiang Qian, Yunli Yang, Qibing Ren

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2609.16012.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16012

Published: 2026-09-17T01:42:27.150Z

2. DenseFace: Bias Mitigation in Face Recognition via Density-Aware Probabilistic Matching

Abstract:Despite steady progress in face recognition, current face recognition models still suffer from significant demographic biases. While approaches for bias mitigation have been proposed, existing methods often impose constraints on the training procedure and result in the degradation of recognition accuracy. To address this issue, we here introduce a method that reduces racial bias in pre-trained face recognition models without compromising their accuracy. To this end, we model face embeddings of each person by von Mises-Fisher (MF) distribution. We next observe the dependency between demographic attributes and the density of MF distributions, and propose DenseFace, a probabilistic face matching procedure that accounts for differences in MF distributions. Our extensive experiments demonstrate DenseFace to consistently reduce racial bias in strong face recognition models varying in network architectures, training datasets and loss functions. Notably, DenseFace preserves recognition accuracy and requires no retraining of the underlying face recognition model. Our work also investigates previously adopted bias measures and makes suggestions.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16149 (HTTP 429)

Authors: Mansur Bultygov, Vadim Seliutin, Dmitry Nekhaev, Ivan Laptev

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2609.16149.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16149

Published: 2026-09-17T01:42:27.150Z

3. Hyperbolic Contrastive Learning with Entailment for Spatial Transcriptomics

Abstract:Spatial Transcriptomics (ST) has transformed biomedical research by enabling the spatial mapping of gene expression across tissue sections. However, high operational costs, specialized equipment requirements, and sensitivity to experimental noise limit the accessibility and scalability of ST. Recent computer vision approaches aim to overcome these limitations by predicting spatial gene expression directly from histopathology images. While effective, current approaches often suffer from gene expression over-smoothing and overly uniform predictions across tissue regions, suggesting that further progress depends on learning representations that reflect the hierarchical and asymmetric structure of gene regulation and tissue morphology. To address these issues, we propose Hyperbolic Contrastive Learning with Entailment for Spatial Transcriptomics (HyCLoST), a hyperbolic contrastive learning model that captures the intrinsic hierarchical relationships within ST data. By leveraging hyperbolic geometry and a gene-to-image entailment loss, HyCLoST learns structured, biologically grounded representations that improve gene expression prediction accuracy, achieving a 6% reduction in MSE and an 8% increase in PCC across 26 ST datasets, over previous methods. Our source code is publicly available at this https URL

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16207 (HTTP 429)

Authors: Daniela Vega, Paula Cárdenas, Hannah Ceballos, Leonardo Manrique, Pablo Arbelaéz

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2609.16207.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16207

Published: 2026-09-17T01:42:27.150Z

4. SceneBench: A Hierarchical Benchmark for Vision-Language Understanding of 3D Scenes

Abstract:Vision-language models excel at 2D image understanding but remain limited in 3D spatial reasoning. Progress is hindered by limitations in current benchmarks. First, 3D datasets often rely on point clouds that capture geometry but discard rich visual features like texture, text, and materials. Second, annotations treat objects in isolation while ignoring real-world hierarchical organization (scenes, rooms, functional areas, object groups). Third, evaluation tasks focus narrowly on basic recognition rather than multi-step spatial reasoning. In this context, we introduce SceneBench, a benchmark of 966 photorealistic 3D scenes reconstructed with Gaussian Splatting and densely annotated with hierarchical semantics spanning scenes, rooms, functional areas, object groups, and individual objects. These annotations are produced through a human-in-the-loop pipeline combining vision-language models with roughly 1,500 human-hours of iterative refinement and verification, producing over 183K annotated nodes with textual descriptions and 3D bounding boxes. Building on this representation, we define three evaluation tasks: Existence-Based Questions probing object attributes, Spatial Intelligence Questions covering counting, size comparison, distance, and directional relations, and Grounded Question-Reasoning-Answer (QRA) triplets requiring multi-step reasoning across semantic levels. Experiments with state-of-the-art vision-language models show that while models perform well on basic recognition tasks (e.g., up to 85% accuracy for detection), performance drops substantially on hierarchical and compositional reasoning (e.g., down to 60% for counting), revealing limitations not captured by existing benchmarks. SceneBench provides a realistic testbed for developing and evaluating models capable of fine-grained spatial reasoning in photorealistic 3D environments.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16233 (HTTP 429)

Authors: Anubhav Khanal, Prabigya Acharya, Roshni Poudel, Sujan Kapali, Bigyan Bhatta, Pramish Paudel, Francois Rameau, Danda Pani Paudel

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2609.16233.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16233

Published: 2026-09-17T01:42:27.150Z

5. ProtoLIP: From Sentence-Level to Object-Level Evidence Disentanglement

Abstract:Query-conditioned vision—language models enable fine-grained interpretation by revealing how visual evidence changes with textual queries. However, evidence conditioned on complete descriptions does not necessarily resolve into object-specific evidence, nor does an exposed evidence map necessarily identify the evidence that constitutes the model’s prediction. Across multiple VLM architectures and independent benchmarks, we find that object-level queries often retain evidence from co-occurring objects and shared context. In this paper, we introduce \textbf{ProtoLIP}, a lightweight prototype-mediated evidence layer that organizes reusable visual prototypes into text-derived semantic families and uses query-dependent family routing to constrain which prototypes may provide evidence. Without spatial annotations or backbone retraining, ProtoLIP improves evidence localization and separation across query granularities, with localization gains transferring to independently pretrained VLMs with well-aligned patch—text representations. Despite using only text-derived weak supervision, ProtoLIP remains competitive with a spatially supervised grounding model while maintaining strong matching and competitive image—text retrieval. Crucially, ProtoLIP constructs its matching score directly from localized prototype evidence, enabling the score to be exactly decomposed into semantic-family and prototype contributions.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16284 (HTTP 429)

Authors: Yan Zhu, Yongbo Chen, Zhengming Ding, Rebecca Faust

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2609.16284.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16284

Published: 2026-09-17T01:42:27.150Z

6. Sequence Recognition in Bharatnatyam dance

Abstract:Bharatanatyam is the oldest Indian Classical Dance (ICD) which is learned and practiced across India and the world. Adavu is the core of this dance form. There exist 15 Adavus and 58 variations. Each Adavu variation comprises a well-defined set of motions and postures (called dance steps) that occur in a particular order. So, while learning Adavus, students not only learn the dance steps but also take care of its sequence of occurrences. This paper proposed a method to recognize these sequences. In this work, firstly, we recognize the involved Key Postures (KPs) and motions in the Adavu using Convolutional Neural Network (CNN) and Support Vector Machine (SVM), respectively. In this, CNN achieves 99% and SVM’s recognition accuracy becomes 84%. Next, we compare these KP and motion sequences with the ground truth to find the best match using the Edit Distance algorithm with an accuracy of 98%. The paper contributes hugely to the state-of-the-art in the form of digital heritage, dance tutoring system, and many more. The paper addresses three novelties; (a) Recognizing the sequences based on the KPs and motions rather than only KPs as reported in the earlier works. (b) The performance of the proposed work is measured by analyzing the prediction time per sequence. We also compare our proposed approach with the previous works that deal with the same problem statement. (c) It tests the scalability of the proposed approach by including all the Adavu variations, unlike the earlier literature, which uses only one/two variations.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16306 (HTTP 429)

Authors: Himadri Bhuyan, Rohit Dhaipule, Partha Pratim Das

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2609.16306.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16306

Published: 2026-09-17T01:42:27.150Z

7. Racing in Volume with Flow Ensembles

Abstract:Streaming 4D reconstruction has been demonstrated only indoors, on dense camera rigs surrounding subjects that move at human pace. Outdoor 4D reconstruction exists but relies either on cameras mounted on the moving vehicle itself, or on limited-coverage arrays observing quasi-static subjects offline. The case that actually matters for spectators is a fast-moving subject, watched from a sparse ring of allocentric cameras, streaming. No method targets this, and no benchmark exists to evaluate one. To this end, we introduce FastFlowGS, a streaming 4D Gaussian Splatting method for reconstructing fast-moving subjects from a small set of fixed external cameras, and Monaco4D, a photorealistic Unreal Engine 5 benchmark for high-speed outdoor reconstruction. FastFlowGS fuses sparse matches, semi-dense tracks, and dense optical flow by lifting each signal to 3D with geometric uncertainty and combining them through a Kalman-style temporal update. Monaco4D provides Formula 1 sequences under varied illumination from trackside, onboard, and drone viewpoints with dense ground truth. On CMU-Panoptic, FastFlowGS exceeds the strongest baseline by 12.6% VMAF at 35% greater efficiency. On Monaco4D, where existing streaming methods degrade severely, it improves dynamic-region PSNR by up to 18.6% with 28.3% lower per-frame optimization time. Dataset and additional details can be found at this https URL.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16310 (HTTP 429)

Authors: Saswat Subhajyoti Mallick, Riu Cherdchusakulchai, Marc Ruiz Olle, Albert Mosella-Montoro, Jose Ribeiro-Gomes, Francisco Vicente Carrasco, Fernando De la Torre

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2609.16310.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16310

Published: 2026-09-17T01:42:27.150Z

8. Reasoning with Image Generation

Abstract:Chain-of-thought reasoning has revolutionized natural language processing by enabling large language models (LLMs) to decompose problems into intermediate steps before answering. Yet confining reasoning to the textual domain presents limitations for tasks requiring direct manipulation of visual representations. Recent efforts augment multimodal LLMs with external visual expert tools such as depth estimation or object detection modules, but these remain fundamentally limited by their reliance on narrow, rigid operations that cannot flexibly generate or transform visual content. We propose ReImaGin, which leverages image generation models as a flexible visual reasoning mechanism for multimodal LLMs: unlike fixed-function tools, they accept natural language commands and can perform open-ended visual operations, like removing an occlusion or generating a floorplan from multiple disjoint views of a room. Across six diverse visual reasoning tasks including multi-view spatial reasoning and collision prediction, ReImaGin consistently outperforms both text-only reasoning and specialist vision-tool baselines, with gains of up to 25\%, demonstrating the advantage of flexible, generative visual reasoning.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16409 (HTTP 429)

Authors: Nishad Singhi, Hector Garcia Rodriguez, Aditya Arora, Marcus Rohrbach, Anna Rohrbach

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2609.16409.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16409

Published: 2026-09-17T01:42:27.150Z

9. Decentralized Gossip Learning and Federated Averaging for Histopathology Image Classification

Abstract:Breast histopathology analysis increasingly relies on distributed learning because direct data pooling across institutions is often restricted by privacy, governance, and communication constraints. This study compares server-based Federated Averaging (FedAvg), fully decentralized gossip learning, and Hybrid Gossip-FedAvg for invasive ductal carcinoma (IDC) patch classification. Experiments used 277,524 color image patches with patient-disjoint training, validation, and test partitions and a workload-balanced, Dirichlet-guided allocation across six nodes. Ring, random degree-3, and fully connected gossip topologies were evaluated together with sensitivity analyses for statistical heterogeneity, mixing coefficient, learning rate, model drift, prediction disagreement, calibration, clinically motivated operating points, communication payload, and patient-level IDC burden, together with auxiliary backbone robustness analyses. In the principal alpha=0.3 experiment, Hybrid Gossip-FedAvg achieved a test area under the receiver operating characteristic curve (ROC-AUC) of 0.8811, closely followed by FedAvg at 0.8801 and fully connected gossip at 0.8751. Across three independent patient-level repetitions, FedAvg and Hybrid Gossip-FedAvg obtained the same mean ROC-AUC of 0.9082, with standard deviations of 0.0037 and 0.0043, respectively. Hybrid achieved the highest mean area under the precision-recall curve of 0.8240, whereas FedAvg produced the lowest mean Brier score of 0.1335. Denser gossip graphs improved discrimination but increased theoretical model payload, while ring gossip remained sensitive to learning rate and mixing strength. Overall, FedAvg provided the most consistently reliable server-based baseline, topology-aware gossip offered a viable decentralized alternative, and Hybrid Gossip-FedAvg provided a balanced compromise between peer-to-peer diffusion and periodic global coordination.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16448 (HTTP 429)

Authors: Yusuf Ozturk, Enes Goltekin, Bengisu Atli, Akin Ozturk, Ulas Bagci

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2609.16448.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16448

Published: 2026-09-17T01:42:27.150Z

10. A multimodal large language model for evidence-based autism spectrum disorder screening

Abstract:The clinical management of autism spectrum disorder (ASD) faces a bottleneck in early screening, mainly because trained specialists are scarce and conventional assessment tools are subjective. Here, we introduce ASDchat, a multimodal large language model designed for evidence-based ASD screening, which takes video, audio, and dialogue as input. ASDchat adopts a dual-branch architecture, where the decision branch generates screening probabilities and the evidence branch generates traceable, timestamped behavioral evidence aligned with standardized clinical criteria (ADOS-2). The model was trained and evaluated on a dataset of 1,035 participants from 27 sites in China, which covered typically developing (TD) children, children with ASD, and children with other disorders. For ASD versus TD, ASDchat reached an area under the receiver operating characteristic curve (AUC) of 0.953 $\pm$ 0.021. On 9 held-out sites that were not used for training, the mean AUC was 0.932. Furthermore, unsupervised clustering of the behavioral dimensions split the ASD cases into six subtypes with different phenotypic profiles, and ASDchat suggests an intervention for each subtype. ASDchat provides a feasible path for large-scale, evidence-based early ASD screening in clinical practice.

中文摘要

翻译失败: Translation limit exceeded. Please try it again later.
Response status: 401 (Unauthorized)
Response body : “{\”ShowCaptcha\”:false}”

LLM Analysis

LLM Analysis Failed: Error: Kimi内容未就绪或抓取失败: 2609.16464 (HTTP 429)

Authors: Jun Chen, Qi Zhao, Yunliang Jiang, Shuqin Cao, Yunqiang Lin, Chenglong Jia, Qiang Guo, Guang Dai, Xiongtao Zhang, Mengmeng Wang, Xiaoyue Ma

Categories: cs.CV

PDF URL: https://arxiv.org/pdf/2609.16464.pdf

CoolPaper URL: https://papers.cool/arxiv/2609.16464

Published: 2026-09-17T01:42:27.150Z