SEARCH · Searcharxiv

Search Searcharxiv

Search indexed arXiv papers on artificial intelligence, large language models, computer vision and robotics. Read source abstracts and follow links to arXiv.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 505 records · Page 28Linked to original sources

arXiv2026

Structurally Close, Temporally Distant: Measuring Security Exposure in Long-Horizon LLM Agents

Long-horizon LLM agents interact with untrusted content, persistent memory, external state, and sensitive tools. Existing analyses often characterize attacks by the number of execution steps between malicious input and a downstream action. We show that temporal remoteness can overstate security separation in stateful agents. We introduce a provenance-aware execution graph linking agent events through deterministic state, identifier, and tool provenance, and define \emph{influence distance} $\DI$ as the shortest structural path from an untrusted source to a sensitive action. We compare it with \emph{sequence distance} $\DT$, the shortest injection--sink path in the ordered trajectory. Since the influence graph contains every sequence edge, $\DI \leq \DT$; $\Gap=\DT-\DI$ measures the separation hidden by step count. Across 454 injection--sink pairs from 360 long-horizon AgentDojo trajectories over OpenAI's \texttt{gpt-4o-mini} and \texttt{gpt-4o} and Claude's Haiku 4.5 and Sonnet 4.6, $\Gap>0$ for 96.9% of pairs, with a median gap of 9 hops; 91.0% remain decoupled after removing the largest provenance-only edge class. On AgentDojo's banking suite, 33.8% of 231 pairs from 377 trajectories decouple through different provenance mechanisms. Among 274 OpenAI pairs, $\Gap$ does not independently predict attack success after controlling for $\DT$, attack family, and backend ($β_{\Gap}=0.066$, $p=.088$). At matched thresholds $k=2,3$, a deterministic $\DI$-based pre-execution gate blocks five attack sinks missed by a sequence-only gate with no additional benign blocking, although the paired gain is not significant ($p=.0625$). Execution structure therefore reveals proximity hidden by step count and can support targeted runtime intervention. We measure candidate influence pathways rather than causal attribution.

cs.CR↗

arXiv2026

Neuron-Guided Fine-Tuning: Unlocking Efficient Alignment Mechanisms for Large Language Models

Existing Supervised Fine-Tuning paradigms, particularly Full Parameter Fine-Tuning are often plagued by parameter redundancy, inconsistent data quality, and catastrophic forgetting, which current methods typically address in isolation and lack a unified optimization signal to bridge data selection, parameter updates, and knowledge preservation. To address this, we propose Neuron-Guided Fine-Tuning (NGFT), a holistic framework that leverages neuron activation patterns as a universal proxy to unify the fine-tuning lifecycle. NGFT operates via three synergistic mechanisms: (1) Adaptive Task-Specific Neuron Selection, which identifies essential neurons in a single forward pass to concentrate updates and reduce redundancy; (2) Activation-Based Data Selection, which prioritizes information-dense samples that maximize contribution to key neurons; and (3) Neuron Activation Alignment, a novel loss function that anchors activations to pre-trained states, deepening representation learning and preserving general knowledge. Experimental results across three models across both domain-specific and general benchmarks demonstrate that NGFT significantly outperforms existing mainstream fine-tuning methods in both efficiency and performance, while effectively mitigating catastrophic forgetting.

cs.CL↗

arXiv2026

Solving versus Verifying: Catching Contradictions in Tax Reasoning Systems

Large language models now compute correct tax liabilities on over 90% of well-formed cases in statutory benchmarks, which makes them candidates for the tax-advisory and compliance systems that consume such an answer directly. Real legal inputs, however, are frequently defective: required facts are missing, or stated facts contradict one another. Accuracy on clean benchmarks says nothing about how a model behaves then, and a system that computes straight through a defective input returns a confident number with no sign that anything is wrong. This raises two questions: does a model asked to solve a case abstain when the input is defective, and when it does not, can the same model catch the defect when asked instead to verify the input? We study six recent models on SARA-derived tax cases under missing-fact and contradictory-fact perturbations. The strongest models abstain when a fact is missing but compute through injected contradictions, returning the clean-input answer 63-76% of the time with no signal of the conflict; asked instead to verify the same input, they flag most of those contradictions. We wire that verification call into a simple contradiction gate: one extra call that abstains when the model reports a conflict. Across all six models it recovers most of the missed contradiction abstention at a clean-accuracy cost of at most about 5 percentage points, with no training and no external tooling. High accuracy on well-formed inputs is therefore an incomplete measure of reliability, and the detection the solver misses is cheaply recoverable with a single self-check.

cs.CL↗

arXiv2026

SurveyAgent-HKA: A multi-agent framework for scientific survey generation with LLMs and human knowledge augmentation

Automatic scientific survey generation has become an important task in scientific document processing. The common approach of retrieving literature from a single source (e.g., arXiv) and generating surveys through a one-pass large language model (LLM) call often leads to limited reference coverage and, more importantly, fails to replicate the expert-driven revision process that is crucial for writing high-quality surveys. In this paper, we introduce SurveyAgent-HKA, a multi-agent framework that improves end-to-end scientific survey generation by incorporating knowledge derived from published surveys and peer-review comments. The framework decomposes survey generation into well-defined sub-tasks handled by LLM-powered agent. It first retrieves relevant papers from multiple sources and identifies key topics through clustering to construct an initial outline, which is then refined using outlines from related human-written surveys. Based on the refined outline, topic-focused papers are retrieved and re-ranked to select for drafting a well-grounded survey. Then, we identify common issues raised by experts in peer-review comments from published surveys to guide the revisions and finalize the survey. Experiments on two domains show that our approach outperforms mainstream baselines in citation quality, structural consistency, and content quality. Furthermore, our framework is efficient in both time and cost, making it a practical solution for broader AI-assisted scientific writing applications.

cs.CL↗

arXiv2026

The Blindness of Document-Level Translation Evaluation

Document-level machine translation (MT) evaluation extends segment-level protocols by presenting full documents to annotators, on the assumption that such presentation elicits document-level judgments. We test this assumption with a counterfactual condition (MIX) in which each document combines segments drawn from different systems, preserving document-level presentation while breaking cross-segment consistency. Across 18,420 expert Englis-to-Korean annotations and 14 automatic metrics, scores, system rankings, and error annotations are statistically equivalent between coherent and incoherent documents. Perception does not explain this: shown matched passages, raters identify the coherent one as the work of a single translator in 87.3% of trials. Document presentation does change how annotators work, but that change does not reach the recorded output. What is blind is the protocol, not the annotator. The concern is not that scores fall short, but that the resources invested in document-level systems, metrics, and annotation may not be measuring what they are intended to measure.

cs.CL↗

arXiv2026

Beyond Cross-Lingual Transfer: Benchmarking Propagation Boundaries in Multilingual LLM Unlearning

Large Language Model (LLM) unlearning aims to suppress target knowledge while preserving general capabilities. In multilingual settings, unlearning must additionally propagate within its intended linguistic scope. However, existing evaluations mainly measure cross-lingual transfer and cannot distinguish insufficient from excessive propagation. We introduce CLLPU (Cross-Lingual and Language-Bound Protocol for LLM Unlearning), a multilingual benchmark that formulates this problem through two settings: common-goal forgetting, where target knowledge should be suppressed across all languages, and language-conditioned forgetting, where suppression should remain confined to a designated language. CLLPU combines goal-guided topic pairing, schema-aware relation matching, and dual-anchor multilingual translation to construct 800 matched knowledge-unit pairs and 72,000 QA instances across ten languages. Experiments with six representative methods on Llama-3.1-8B-Instruct reveal opposite failure modes: forgetting remains incomplete when universal suppression is required, yet spreads beyond the intended boundary when language-conditioned confinement is required. We further find that general multilingual utility can conceal damage to neighbor knowledge. These findings establish propagation control as a central challenge for multilingual LLM unlearning. We publicly release CLLPU together with its construction pipeline.

cs.CL↗

arXiv2026

Alignment by Stereotyping: How LLMs Sacrifice Individual Distinctiveness for Cultural Adaptation

Large language models are increasingly deployed for personalized interaction, and demographic conditioning via user profiles is a widely adopted strategy for cultural adaptation. We ask whether this approach genuinely serves individual users or achieves accuracy by erasing individual distinctiveness. Studying seven models including frontier GPT-5.1 on the World Values Survey, we find that demographic profiles improve value alignment accuracy for most models, but at a systematic cost to individuality. That is, models pull responses toward demographic group centroids rather than preserving individual differences, a behavioral pattern we term alignment by stereotyping. Permutation tests (10,000 permutations, six demographic attributes, seven models) certify that top-performing models compress individuals far above the human baseline; within-family scaling amplifies this tradeoff while degrading intrinsic cultural understanding. Using a synthetic dialogue dataset validated on real human-chatbot conversations from PRISM (Kirk et al., 2024), we further show that distributing demographic signals across conversational turns partially suppresses prototype retrieval compared to compact demographic labels, a finding validated on real conversations via PRISM but requiring replication at larger scale.

cs.CL↗

arXiv2026

ModularPhaseNet: Finite-Cyclic Phase Geometry for Computable Semantic Hierarchy, Direction, and Context Consistency in Standard Transformers

We propose ModularPhaseNet, a classical and integer-computable discretization of the continuous complex phase geometry introduced in QuantumPhaseNet. The real-valued hidden states of a standard Transformer are retained, while only an auxiliary phase channel is quantized into a cyclic subgroup G = of order q | (p-1) in the multiplicative group of F_p. A continuous phase e^{i phi} is represented by z = g^a mod p; phase composition becomes group multiplication, relative phase becomes group division, conceptual hierarchy is induced by a filtration of cyclic quotients, semantic direction is represented by oriented relative group elements, and contextual consistency is measured by gauge-invariant cycle holonomy. The method introduces three components into an otherwise standard Transformer: a finite-phase encoder, a quotient-filtration hierarchy module, and a group-valued connection module. Their outputs enter self-attention as real-valued bias terms. Training uses distributions in the real group algebra or straight-through Gumbel-Softmax, whereas inference uses exact modular exponentiation and precomputed tables. No quantum hardware, complex-valued matrix multiplication, or discrete-logarithm computation is required. We prove quantization-distortion bounds, nesting of quotient-induced partitions, gauge invariance, a discrete integrability result for flat connections, and boundedness of the resulting attention output. The central empirical hypothesis is that these exact discrete invariants improve hierarchy recovery, discourse alignment, contradiction detection, and calibrated hallucination-risk prediction under a controlled compute budget. This paper reports the theory together with a pre-registered evaluation plan; the experiments described in Section 14 have not yet been carried out, and no empirical result is claimed here.

cs.CL↗

arXiv2026

Tri-PvP: Exposing Modality Bias in Omni-Modal Large Language Models through Perceptual-Propositional Evidence Conflicts

Omni-modal large language models (OLLMs) jointly process vision, audio, and text, yet their modality bias under cross-modal conflict remains underexplored. Existing benchmarks conflate two distinct forms of evidence within a single modality: perceptual signals (e.g., a photograph or recording of a dog) and propositional signals (e.g., the declarative claim "this is a dog"), such that any measured modality bias is inherently confounded with evidence-form bias, precluding clean attribution to either source. To address this, we introduce Tri-PvP, an 8,000-sample tri-modal conflict benchmark crossing vision, audio, and text, where vision and audio each take perceptual or propositional form. Evaluating five OLLMs, we find robust visual bias across most models and evidence-type conditions. Crucially, we reveal a systematic asymmetry in evidence-form bias: models exhibit a stronger bias toward perceptual signal in vision but propositional in audio. Further analyses via layer-wise linear probing and contrastive decoding reveal that modality bias is already linearly decodable from early representation layers and can only be partially mitigated, calling for mitigation strategies beyond surface-level interventions.

cs.CL↗

arXiv2026

Generating Adversarial Texts for Machine Translation via GRPO

As machine translation (MT) systems continue to improve, standard benchmarks become less informative for exposing remaining weaknesses. Traditional methods for creating challenging test sets rely on expensive manual creation or curation, while automated approaches struggle to produce sets with the necessary translation difficulty and linguistic diversity. We propose a scalable reinforcement-learning-based approach for rewriting existing source texts into instances that are more difficult to translate for MT systems. We fine-tune a large language model with Group Relative Policy Optimization (GRPO), using reward signals based on translation difficulty together with constraints for semantic similarity, grammaticality, and approximate length preservation. On WMT25, our approach substantially reduces average COMET translation quality from 0.63 to 0.48, while preserving grammaticality and readability, whereas the base model remains at 0.64. Evaluations on the unseen WMT19-WMT24 benchmarks confirm that this behavior generalizes beyond the training data, and human evaluation further shows that the rewrites substantially lower translation quality while incurring a moderate drop in naturalness and only a small change in grammaticality. We release our code to support reproducibility.

cs.CL↗

arXiv2026

GradeTrap: Authority Cues in Images Shift VLM Judgments Despite Explicit Instructions to Ignore Them

As vision-language models (VLMs) become increasingly capable and are deployed in consequential real-world settings, they must evaluate evidence independently rather than defer uncritically to human authority. We introduce GradeTrap, a controlled evaluation that places two social cues in direct conflict: a student answer, which should attract sycophantic agreement, and a conflicting answer attributed to a peer, teacher, or official answer key, which should attract authority-based deference. Models produce free-form answers while being explicitly instructed to solve independently and ignore all student answers, feedback, and grading marks. We test the models on 60 synthetic real-world trade-off scenarios. Five neutral trials establish a stable model-relative preference, followed by three repetitions of six experimental cues including controls. On the 45-item common intersection across Gemini 3.5 Flash-Lite, GPT-5.6 Luna, and Claude Haiku 4.5, a generic second-answer control yields 5.4% conflicting-answer selection. Relative to that control, pooled within-item changes show no reliable peer-review effect, a 6.9-point teacher-review effect, and a 19.5-point official-key effect. In contrast, a displayed conflicting student answer alone compared to a displayed student reference answer alone only raises selection from 2.2% to 5.2%. Official-key provenance therefore redirects judgements more than a student answer or the generic second-answer control, despite an explicit ignore instruction and an opposing student answer given along with the official key. Effects vary in magnitude across the three models.

cs.CV↗

arXiv2026

Don't Lose Entities from Retrieval to Generation: Dual Entity Recovery RAG for multi-hop QA

Retrieval-augmented multi-hop question answering (QA) decomposes a query into sub-questions and decomposes the corpus into smaller retrieval units such as sentences. Both forms of decomposition improve the pipeline, but we show that both share the same vulnerability, the loss of entity information, and that this loss breaks the pipeline at two separate points. The first point is retrieval, where a sub-question loses the entity resolved at the previous hop, leaving the retriever with nothing to match against. The second point is harder to see, because retrieval still appears to succeed. Once a passage is split into sentences, an isolated sentence loses the context that grounds its pronouns, so even with the correct sentence in hand the LLM cannot tell which entity the sentence is about. We isolate this second point as a distinct failure mode that we call lost-in-generation, and a retrieval-controlled experiment shows that it degrades answers even when the gold evidence is fixed in the context. We then propose Dual Entity Recovery RAG (DER-RAG), which keeps the grounding entity explicit from retrieval through to generation with two lightweight components, a two-way query decomposition that carries the resolved entity across sub-questions and a subject entity prefix attached to each sentence at generation time. DER-RAG needs no graph construction, no corpus modification, and no fine-tuning, yet on three multi-hop QA benchmarks it matches or exceeds strong baselines, including graph-based methods that depend on costly offline structures.

cs.CL↗

arXiv2026

DPH Parser: A Bottom-Up Grammar-Driven Parser for Joint Constituency and Dependency Analysis

This paper presents Dependency-Phrase Hierarchy Parser (DPH Parser), a grammar-driven bottom-up unsupervized parsing framework inspired by Generalized Phrase Structure Grammar (GPSG) and Head-driven Phrase Structure Grammar (HPSG). The parser incrementally constructs constituency structures using a compact inventory of feature-based syntactic rules while deriving dependency relations through explicit head annotations. The system combines probabilistic POS tagging, recursive phrase projection, and weighted parse hypotheses to process realistic and partially noisy text input. Unlike purely neural and data-driven parsers, the resulting syntactic derivations remain explicitly interpretable. We evaluated parser performance on English corpora from the Universal Dependencies (UD) project using Unlabeled Attachment Score (UAS) as the main parsing metric, comparing the outcomes against Stanza and spaCy parsers. For a small inventory of syntactic rules, DPH parser achieved UAS values of 53.32% & 52.58% (UD Devset/Testset respectively). For the same data, Stanza achieved 89.12% & 88.67% while spaCy achieved 56.91% and 58.59%. Although the current system does not yet approach the accuracy of modern neural parsers, the results demonstrate the feasibility of applying transparent rule-based bottom-up parsing to realistic treebank data while jointly producing constituency and dependency structures.

cs.CL↗

arXiv2026

ACE: Adapter Consolidation across Experts for Parameter-Efficient Fine-Tuning of MoE LLMs

Parameter-efficient fine-tuning (PEFT) of mixture-of-experts (MoE) models commonly attaches a separate low-rank adapter to each expert. This expert-wise design fragments adaptation in three ways: capacity is split across narrow low-rank updates, gradient supervision becomes sparse and imbalanced under sparse routing, and execution is decomposed into many small GEMMs. We find that such expert-wise separation is often unnecessary, as subsets of LoRA adapters become functionally similar during fine-tuning, revealing redundancy among expert-specific adapters. Based on this redundancy, we propose ACE (Adapter Consolidation across Experts), which groups redundant experts and replaces their expert-specific adapters with group-shared higher-rank LoRA modules under the same PEFT budget. ACE further introduces grouped adapter execution, which consolidates fragmented expert-wise adapter computations into fewer, larger group-level GEMMs. Across evaluations covering 12 datasets and four MoE backbones, ACE achieves the highest observed mean accuracy among the parameter-matched PEFT methods on the three backbones with complete baseline coverage, while providing $1.31\times$ to $1.48\times$ wall-clock training speedup over expert-wise LoRA without increasing peak memory. Our code is available at https://github.com/UbiquitousAILab/ACE.

cs.LG↗

arXiv2026

DataFlex-RL: An Evaluation Platform for RLVR Data Policies

Data policies for reinforcement learning with verifiable rewards (RLVR) determine which rollouts are used, how strongly they are weighted, and which domains contribute to subsequent training batches. We introduce DataFlex-RL, an evaluation platform for comparing these choices under a common GRPO recipe. Our primary experiment evaluates 13 configurations across 12 matched seeds using Qwen2.5-7B-Base and 12 mathematics, logic, and science benchmarks. Uniform GRPO improves the domain-balanced average accuracy by 7.76 percentage points over the untrained checkpoint. None of the eight rollout-selection or reweighting methods achieves a paired 95% confidence interval that excludes zero relative to uniform sampling, and none of the three adaptive mixtures outperforms a fixed equal mixture at the same level of precision. A corrected 12-seed extension on Llama-3.1-8B-Base places the additional methods on the same score scale as the original controls, but does not reveal a consistent winner in terms of observed mean performance. We also quantify evaluation sensitivity by rescoring nine Qwen2.5-7B-Instruct runs using a math-heavy six-benchmark summary, consisting of five mathematics benchmarks and GPQA-Diamond but no logic benchmark, and comparing it with the domain-balanced 12-benchmark summary. The resulting rankings are negatively correlated, with a correlation coefficient of -0.33, whereas summaries that retain all 12 benchmarks largely agree. Across the controlled settings studied here, changing the data policy measurably changes the training process but does not produce a reproducible improvement over uniform training.

cs.LG↗

arXiv2026

From Two Passes to One: Compact and Efficient Target-Stance Extraction

Target-Stance Extraction (TSE) is the task of predicting both the target (or topic) of an author's writing and the author's stance toward it. Existing approaches to TSE use a sequential pipeline of two separate neural models: one to identify the target and another to determine the stance. We present a one-pass, joint architecture that predicts both in a single forward pass, reducing trainable parameters by nearly 50% with only a 4-7 F1 point tradeoff in performance. We further demonstrate that standard target-scrubbing practices artificially suppress target prediction accuracy. Retaining explicit target mentions, as in real-world deployments, improves F1 by at least 6 points across both target classification and target generation settings. These improvements allow for significantly easier integration of TSE in downstream applications such as public opinion tracking.

cs.CL↗

arXiv2026

STQA: A Benchmark for Stock-Focused Tabular Question Answering over Historical and Forecasted Data

Stock market analysis inherently requires composite reasoning over historical records and future projections, yet existing benchmarks remain fragmented across isolated tasks. We introduce STQA (Stock-focused Tabular Question Answering), an end-to-end benchmark designed to systematically evaluate natural-language question answering over historical data, numerical forecasts, and forecast-based reasoning. Built on a large-scale financial dataset, STQA covers 4,417 stocks and contains 31,400 question-answer pairs derived from expert-crafted templates, accompanied by fine-grained intent and slot annotations. To operationalize this benchmark, we present SQFRS (Stock Query-Forecast-Reasoning System), an agent-based unified framework that orchestrates SQL retrieval and time-series forecasting tools. Experiments demonstrate that while current large language models perform well on historical queries, forecast-based reasoning poses a substantial challenge, revealing critical bottlenecks in tool coordination and reasoning under uncertainty. The dataset and code are available at https://github.com/xuxubaobaoan/STQA_Project. STQA thus serves as a rigorous testbed for future research on trustworthy, tool-augmented financial agents.

cs.CL↗

arXiv2026

Protocol Compression Changes Which Party Pays: Bilateral Cost in Cross-Organization LLM Agent Communication

Agents that talk across organizations exchange long messages billed by the token. A shorter notation therefore looks like a saving that costs nothing but an agreement to use it. Recent work reports the saving is conditional. Compressed notation can instead raise total tokens by 8% to 11% over a JSON baseline, when parsing failures force extra model calls. That is measured for one payer. Between two organizations neither side can install a decoder at the other end, and each pays under its own tokenizer, price, and cache state. We measure both sides. A preregistered token-level study covered 198 content-matched item pairs across six vendors, for 2,376 native-usage cells. We then overlay an English baseline, runtime schema negotiation followed by compression, and injected-schema compression on a two-party procurement bargain with an exactly enumerated feasible set. The overlay covers 1,053 completed dialogues of a 1,215-cell grid across 3 model pairs, plus a 405-dialogue rerun of the negotiated condition. Compression amplifies cross-vendor cost dispersion by a factor of 1.078, with a 95% CI of [1.066, 1.091], and two vendor pairs reverse which endpoint is cheaper. Runtime negotiation succeeds as a protocol and fails as a bargain. The parties agree a schema in 121 of 135 headline dialogues, none of them the schema we would have supplied. They settle the task in only 9 of those dialogues, and they reach impasse in 106 of them. The negotiated sessions average 10.8 turns against 17.6, and cost 52% of the English total because sessions end sooner, not because the handshake is repaid. Break-even horizons run from 20 to 70 turns, the low end only under the conditional accounting, and all of them lie above every observed English session. On one cross-vendor pair both parties keep about half their cost. On the other the receiving party pays more at a high cache-hit rate.

cs.CL↗