arXiv subjects

Search papers

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

At least 919 records · Page 51Linked to original sources

arXiv2026

From One Query to Many: A Critical Survey of Privacy Accounting in RAG

Retrieval-augmented generation (RAG) enables large language models to leverage external knowledge but introduces privacy risks when the retrieved corpus contains sensitive information. Differential privacy (DP) offers formal protection, yet interpreting and comparing privacy guarantees across existing DP-RAG methods remains challenging. In particular, a privacy budget reported for a single response does not directly characterize privacy loss over repeated queries, and even identical values of $\varepsilon$ may represent different guarantees depending on the protected data unit, accounting horizon, and observable outputs. This survey critically examines how privacy accounting evolves from single-query generation to multi-query RAG systems. We categorize existing approaches into query-time privatization, multi-query accounting, and one-time corpus privatization. We analyze their underlying assumptions and privacy--utility trade-offs. Rather than ranking methods by $\varepsilon$ alone, we propose a framework that interprets a DP-RAG guarantee along three dimensions: privacy granularity, accounting scope, and release boundaries. We conclude that an $\eps$ is a claim only together with the three slots: the protected unit, the accounting scope, and the release boundary.

cs.CR↗

arXiv2026

FGSS-BO: Feedback-Guided Hypothesis Management for LLM-Based Bayesian Optimization

Large language model (LLM)-guided Bayesian optimization (BO) uses past observations to guide search, but proposed directions and their revision can remain implicit. We propose Feedback-Guided Search State (FGSS), a structured hypothesis-management interface connecting language-based guidance to observed evidence, executable parameter interventions, and comparative support updates. An LLM reflector proposes and revises hypotheses; a deterministic curator validates them and maintains identities and test records. The evolving state guides candidate construction and priorities. Instantiated on LLAMBO, FGSS-BO retains its surrogate and default expected improvement selection, with a bounded testing policy. On two retrospective protocol/media tasks with nearest-neighbor response oracles, it attains higher sample mean final best and search AUC than the matched LLAMBO baseline and evaluated non-LLM baselines. Prose-reflection and EI-only controls show mixed component effects. Six direct-training hyperparameter-optimization tasks show task-dependent transfer, strongest on diabetes SVM; exact-grid and reaction controls examine other settings. FGSS exposes hypotheses, their execution, and cross-round revisions for inspection while supporting search gains in selected settings.

cs.CL↗

arXiv2026

Semisimplicity and Rigidity of the Kontsevich--Zorich Cocycle for Products of Strata

We study the Kontsevich--Zorich cocycle, its fixed tensor constructions, and their finite direct sums over products of strata of Abelian differentials. The acting groups are closed subgroups $G\subset(\mathrm{SL}_2\mathbb R)^n$ that, up to factor permutations and coordinatewise conjugation by rotations, are products of diagonal $\mathrm{SL}_2\mathbb R$ blocks and a finite central group. For ergodic invariant probability measures, we prove Deligne-type semisimplicity: the bundles admit Hodge-compatible decompositions into $G$-invariant pieces $V_i\otimes_{A_i}W_i$, where $A_i$ are division algebras, $V_i$ carry pure variations of Hodge structure, and the multiplicity spaces $W_i$ carry finite direct sums of pure Hodge structures. Every invariant subbundle arises from $A_i$-submodules of the $W_i$. If block marginal measures have smooth local supports and are locally equivalent to smooth volume, Hodge-orthogonal projections onto invariant subbundles are locally polynomial in period coordinates on the normalized support. The proof obtains a local product-of-quadrics model and joint polynomiality from polynomiality along stable and unstable leaves. On the cone obtained by scaling each diagonal block together, projection coefficients admit homogeneous degree-zero expressions with integral powers of reciprocal block areas. For the full product group $(\mathrm{SL}_2\mathbb R)^n$, single-stratum measure classification supplies the support hypothesis, yielding rigidity for every ergodic invariant probability measure. Under the transported support hypothesis, arbitrary coordinatewise $\mathrm{SL}_2\mathbb R$-conjugates also admit polynomial equivariant projectors, though not necessarily Hodge-orthogonal. Genus-two examples show that nonrotational conjugation can destroy Deligne-torus invariance, and independent factor scalings can require square roots of area functions.

math.DS↗

arXiv2026

Beyond Confidence: Rethinking Uncertainty Calibration in Deep Code Models

Deep learning models have shown significant promise in code-related tasks. While confidence calibration techniques have been proposed to improve model reliability, their effectiveness in real-world tasks with strict accuracy requirements remains uncertain. In this paper, we investigate two key questions: first, whether existing measurement metrics, including confidence scores and uncertainty metrics, are effective for guiding selective prediction; and second, whether existing calibration techniques can improve these metrics to better identify cases where model predictions are likely to succeed or fail. We conduct a systematic study on defect prediction and vulnerability detection using seven code models, evaluating thirteen uncertainty signals and three widely used post-hoc calibration methods. Our analysis reveals three major findings: (1) even strong uncertainty signals can remain poorly calibrated and provide limited separation at the high-confidence end, constraining selective prediction in practice; (2) standard post-hoc calibrators often improve calibration scores but do not consistently translate into better selective prediction behavior; and (3) commonly used calibration metrics can disagree with selective prediction performance, underscoring the need to evaluate both jointly. Moreover, our findings motivate a decision-oriented view of reliability in deep code models, where calibrated scores must be evaluated together with selective-prediction quality and retained-region risk before they can support accept--defer deployment.

cs.SE↗

arXiv2026

Characterizing Predictability-Latency Trade-offs of KV-Cache SSD Offloading in LMCache for LLM Serving Systems

KV-cache offload is widely used to stretch GPU memory for LLM serving, but its storage behavior has not been characterized at the block-device level. In this paper, we study LMCache through realworld multi-session workloads that span same/different context x same/different prompt, using over 100 stateless requests per workload. Our comparison is observational rather than factorial: it contrasts two realizable deployment snapshots -- LMCache 0.3.0 with buffered I/O and the kernel page cache, and LMCache 0.3.16 with prefix-aware deduplication and O_DIRECT. Key findings define the paper. First, the legacy stack appears almost read-free at the SSD layer in all four workloads, but a flush_ram experiment shows that this is conditional on page-cache warmth: once the cache is evicted, the same path issues about 1 GB of sequential reads in 683 ms. Second, deduplication delivers a dramatic write reduction only for the exact-repeat workload: one set collapses from 1.62 GB to 11.75 MB of writes, while the other three sets remain in the 2-5 GB range. Third, O_DIRECT converts a hidden and bimodal read cost into an explicit and stable one: on one set, warm TTFT rises from 27 ms to 93 ms because each of the 99 warm requests re-reads the same ~12 MB partial chunk from SSD. The main conclusion is therefore not "newer is better," but a system trade-off: predictability versus average-case latency.

cs.OS↗

arXiv2026

FocusOPD: Structure-Guided On-Policy Distillation for Diffusion Models

Multi-teacher on-policy distillation enables rectified-flow models with diffusion-transformer architectures to combine task-specific capabilities within a single student. However, matching model outputs does not guarantee that their internal representations are aligned, and it remains unclear where and when additional feature supervision is most useful. We analyze the teacher-student feature mismatch across Transformer blocks and denoising timesteps, and find that the mismatch is concentrated in deeper Transformer blocks during early denoising and persists after OPD distillation, indicating ineffective distillation. Motivated by this observation, we introduce FocusOPD, which augments the base OPD objective with token-wise directional feature alignment restricted to deep layers and early denoising steps. Across two rectified-flow backbones and multiple OPD objectives, FocusOPD improves capability transfer in compositional generation, text rendering, and preference-based evaluation. On the primary benchmark, it improves over DiffusionOPD from 0.953 to 0.967 on GenEval, from 23.797 to 23.944 on PickScore, and from 6.052 to 6.199 on LAION-Aesthetic. Ablations show that selective deep and early supervision is more effective and efficient than alignment over the full network and trajectory.

cs.CV↗

arXiv2026

Do VLAs Really Follow Instructions? Diagnosing How-Level Instruction Following in Vision-Language-Action Models

Vision-language-action (VLA) models are expected to follow instructions like their language backbones, yet prior work has focused on What-level instructions that specify the task rather than How-level instructions that specify how to carry it out. Through our diagnoses, we find that How-level instructions do not help current VLAs and lower task success. We hypothesize that pairing each training episode with a single instruction limits How-level instruction following, and show that training with segment-level enriched instructions makes actions respond to such instructions without consistently raising task success. To measure How-level instruction following beyond standard benchmarks, we introduce SIMPLER-IN, where success requires conditions given only in the instruction. It shows that the gained responsiveness does not extend to demanding or out-of-distribution How-level instructions. How-level instruction following thus remains an open challenge.

cs.CV↗

arXiv2026

Finite-History Outage Certification for Partially Observed Fluid Antenna Receivers

Fluid antenna receivers can select a service position from partial port observations, but finite historical data leave uncertainty in the covariance used for that decision. We develop deployment-before-use reliability certification for a frozen partial-observation rule. A proper predictive density defines a history-based parameter region. Gaussian covariance envelopes, shared conditional-probability bounds and selector enclosures provide simultaneous risk bounds over continuous parameters and discontinuous port decisions. Separate history-coverage and integration budgets limit erroneous positive certification to 0.05 for a single qualification procedure under the stated stationary model and numerical assumptions. Confidence-set inclusion allows a new independent history to reuse the risk computation with the rule unchanged. In a structured Gaussian example with 32 candidate positions, five noisy probes and target outage 0.001, all 282 risk groups are qualified; the largest upper bound is 0.000999748. A separate coupled comparison gives a current-rule outage interval of $[7.6820,9.1266]\times10^{-4}$ and resolves lower outage than two observed-port controls using the same historical fit. One independently recalibrated history reuses the certificate without additional risk sampling. Unresolved and adverse-member controls delimit this reuse.

eess.SP↗

arXiv2026

What do AI chatbots do after users decline therapeutic redirection for suicide thoughts?

Safety evaluations of artificial intelligence (AI) chatbots focus on therapeutic redirection, statements urging users who disclose suicide thoughts to seek help from crisis lines or other human sources. How chatbots respond when users refuse that help is unclear. This cross-sectional audit ran 2,550 scripted multi-turn conversations in September 2026. Each conversation opened with a disclosure of suicide thoughts at 3 levels of severity, answered every referral with up to 10 standardized refusals and then changed the subject. All 2,550 conversations included a referral, 2,542 (99.7%) in the first reply. However, mean persistence ranged from 0.01 to 9.94 refusals across chatbots. Newer versions of all 5 chatbots tested in 2 versions persisted less than older versions (each P < .05). Compared with passive ideation, a disclosure of intent was associated with 1.26 more refusals sustained (95% CI, 1.06-1.46) and higher odds of renewed referral after the topic change (odds ratio, 10.07; 95% CI, 7.43-13.66). Chatbots that persisted less in referral more often presented themselves as the user's support (Spearman rho = -0.80; P < .001). Of 8,818 refusal-phase replies urging outside help, only 173 (2.0%) asked about barriers to seeking it. Chatbots uniformly recommended outside help after disclosures of suicide thoughts but varied widely in persisting in those recommendations, with newer versions persisting less. Safety evaluations should test what happens after a crisis referral is refused, not only whether one is offered.

cs.HC↗

arXiv2026

LLM-Based Weak Annotation for Cybersecurity Tweet Classification: A Teacher-Student Framework

Cyber threat intelligence (CTI) relies on the timely identification of emerging threats, enabling organizations and security analysts to support situational awareness and decision-making. Social media platforms such as X (formerly Twitter) have become valuable CTI sources because vulnerabilities and attack activities are often discussed there before appearing in formal reports. However, the rapid growth of cybersecurity discussions on social media has created more content than domain experts can realistically review and annotate, making labeled data a major bottleneck for developing supervised learning approaches. While large language models (LLMs) have shown promise as weak annotators in other domains, their effectiveness for cybersecurity tweet classification and downstream training of domain-specific pre-trained language models (PLMs) remains underexplored. To address this challenge, we propose a Teacher-Student weak supervision framework in which a selected LLM serves as a Teacher to generate weak labels for unlabeled cybersecurity tweets, enabling the training of cybersecurity-specific PLMs without large-scale manual annotation. Experimental results show that PLMs trained solely on LLM-generated labels consistently achieve F1 scores above 0.92, remaining within 3 percentage points of fully human-supervised models across all evaluated architectures. These findings suggest that LLM-based weak annotation offers a practical alternative for constructing cybersecurity-specific training data when expert annotation resources are limited.

cs.CL↗

arXiv2026

On coefficient theorems for starlike and quasi-convex of type B mappings in complex Banach spaces

In this paper, we study biholomorphic mappings $F: \mathbb{B} \to X$ of the form $F(z) = f(z)z$ for $z \in \mathbb{B}$, where $f: \mathbb{B} \to \mathbb{C}$ is a holomorphic function satisfying $f(0) = 1$ and $X$ is a complex Banach space with the unit ball $\mathbb{B}$. The main contributions of this paper are threefold. First, we establish a sufficient coefficient condition for a function $F(z)=zf(z)$ to be starlike and quasi-convex of type $B$ in a complex Banach space. Second, we prove that the Bieberbach conjecture holds for these classes of functions. Finally, we analyze the difference of coefficients for such functions in this setting. The results established here directly extend the previous results to the broader setting of complex Banach spaces.

math.CV↗

arXiv2026

Isolating a Language Model's Judgment with a Label-Free Subspace

Language models are often asked to make judgments: whether a proposed answer is correct, whether one statement follows logically from another, or whether a response is safe. Yet changing the instructions that specify how the judgment should be expressed can make the final answer less reliable, even for the same input. We ask whether this reflects a change in internal judgment or a failure to express it. Comparing hidden-state vectors for the same input under different instructions, we find that their differences concentrate in a low-dimensional expression subspace. We identify this subspace without benchmark labels and remove its contribution from a linear predictor to obtain an isolated judgment. Across correctness, entailment, and safety, the isolated judgment remains reliable under instructions not used to identify the subspace, even when the model's final answer and the original predictor degrade. When the judgment question is negated, isolation raises the predictor's AUROC from 0.504 to 0.825 for entailment and from 0.480 to 0.891 for safety. We also find that judgment is readable in earlier layers, before the expression subspace begins to disrupt the predictor. Together, these results provide a way to separate judgment from expression and recover internal judgments even when the model's final answers are unreliable.

cs.LG↗

arXiv2026

Fractional Power-of-Two Quantization for Efficient and Effective Multiplier-Free LLM Inference

Large Language Models (LLMs) demand efficient inference, where the multiply-accumulate (MAC) operations in linear projections dominate compute and energy cost. Post-training quantization (PTQ) reduces this cost by mapping weights and activations to low-bit integers. Power-of-Two (PoT) weight quantization promises efficient multiplier-free LLM inference via shift-add logic, but suffers significant accuracy degradation at low bitwidths due to its coarse base-2 quantization grid. We introduce Fractional PoT (FPoT), a base-$\sqrt{2}$ (half-octave) fractional PoT grid that achieves practical 4-bit weight quantization while preserving a fully integer shift-add datapath. The method combines (i) the base-$\sqrt{2}$ weight grid with uniform integer activations, (ii) a calibration-time grid-alignment refinement absorbed into the static per-channel scale, and (iii) a lightweight 1-adder $\sqrt{2}$ approximation whose error is absorbed during weight calibration via approximate-grid quantization, yielding a processing element (PE) with significantly fewer gates than a 4-bit integer multiplier. On Llama-3-8B with 4-bit weight and activation quantization, FPoT improves zero-shot commonsense reasoning performance by 8.72% relative to the PoT baseline, while maintaining performance close to the uniform integer baseline, with only a negligible 0.4%p degradation. These results hold across various configurations spanning seven models and three bit-widths. Furthermore, FPoT can be integrated with state-of-the-art LLM PTQ methods, demonstrating framework-agnostic applicability.

cs.LG↗

arXiv2026

Rigidity for formally determined point source scattering for an anisotropic acoustic medium

Consider an acoustic medium, occupying three dimensional space, whose anisotropic sound speed is modeled by a Riemannian metric which is Euclidean outside a compact region $D$. The medium is probed by acoustic waves generated by a finite number (dimension dependent) of point sources outside $D$ and the medium response is measured on the boundary of $D$ for a finite time period. We show that if these measurements agree with that of an Euclidean metric then the metric is Euclidean up to a diffeomorphism. This result is similar to the result in \cite{ORS26a} where they used plane waves instead of point sources. Using point sources instead of plane waves presents new difficulties, dealt with in this article.

math.AP↗

arXiv2026

Energy Flow Shapes the Decomposition Dynamics of a Reactive Atmospheric Intermediate

The unimolecular dynamics of Criegee intermediates depend strongly on their internal energy, which can differ substantially between nascent atmospheric species and laboratory-prepared reactants. Here, the decomposition dynamics of {\it anti-}CH$_3$CHOO are investigated using molecular dynamics simulations on a neural network-based potential energy surface under thermal and mode-selective vibrational excitation. Simulations were performed at internal energies representative of those released during ozonolysis (chemical/thermal preparation, TP) or following vibrational excitation (VibEx) along the CH-stretching mode with two or three quanta, as can be done in laboratory studies. These excitation conditions strongly affect reaction yields and kinetics, with VibEx promoting passage across the initial cyclization barrier. For TP, 35 kcal/mol of internal energy are required to form molecular products, whereas for VibEx, 27 kcal/mol are sufficient. Although TP and VibEx ensembles access largely the same fragmentation pathways with comparable relative yields, the absolute product yields, characteristic timescales, and kinetic behavior differ substantially. Methyldioxirane is found to be a kinetic trap along the pathway. The resulting molecular-level description of {\it anti-}CH$_3$CHOO decomposition provides guidance for interpreting experiments and contributes to a broader understanding of energy-dependent Criegee-intermediate dynamics.

physics.chem-ph↗

arXiv2026

Alpha-convex functions in complex Banach spaces

Let $\mathbb{B}$ be the unit ball in a complex Banach space ${\mathbb X}$. We study biholomorphic mappings $F: \mathbb{B} \to {\mathbb X}$ of the form $F(z) = f(z)z$ for $z \in \mathbb{B}$, where $f: \mathbb{B} \to \mathbb{C}$ is a holomorphic function satisfying $f(0) = 1$. Our main contributions are as follows: Inclusion Results: We establish inclusion conditions for $\mathcal{M}_α(\mathbb{B})$, class of $α$-convex functions in $\mathbb{B}$ to belong to the classes of starlike mappings or quasi-convex mappings of type $B$ on $\mathbb{B}$ for specified ranges of $α$. Characterization: We prove that $F(z) = \frac{f(l_{u}(z))}{l_{u}(z)}z \in \mathcal{M}_α(\mathbb{B})$ if and only if $f \in \mathcal{M}_α(\mathbb{U})$, where $f \in \mathcal{S}(\mathbb{U})$, $u \in {\mathbb X}$ with $\|u\| = 1$, and $l_{u} \in T(u)$. Duality: We present a duality relationship between the classes $\mathcal{S}^*(\mathbb{U})$ and $\mathcal{M}_α(\mathbb{B})$ for $α\ge 0$. Hankel Determinant Bounds: We determine sharp upper bounds for the second Hankel determinant of mappings in $\mathcal{M}_α(\mathbb{B})$ when $α\ge 0$. The results established here directly extend the previous results to the broader setting of complex Banach spaces.

math.CV↗

arXiv2026

Behavioral Compositionality of Sparse Autoencoder Representations

Sparse autoencoders (SAEs) represent model activations with sparse, linearly decoded features, but the downstream effects of those features need not compose additively. We introduce SAE Additivity Residuals (SAR), which measure the behavioral variation left unexplained above a chosen interaction order. Under independent feature subset sampling, population SAR is exactly the normalized energy of interactions above that order and bounds coalition ordering error for the optimal low order approximation. To distinguish compositionality associated with the SAE representation from that induced by downstream computation, we compare learned SAE coordinates with sum preserving random rotations of the same selected subspace and find lower first order residuals for the learned coordinates across all 18 model and behavior settings. The coordinates are nevertheless not fully additive: pairwise terms improve held out coalition ordering in every setting and outperform parameter matched models using randomly selected third order terms. In a combined steering experiment, pairwise modeling raises rank fidelity from 0.5874 to 0.7790 and reduces ordering error from 0.1444 to 0.0606. Together, these results show that SAE coordinates support comparatively compositional behavioral descriptions while faithful joint explanations and interventions still require interactions beyond individual feature effects. Code is available at https://github.com/OpenSAE/SAE-Behavioral-Compositionality.

cs.LG↗

arXiv2026

Rethinking Learning from Label Proportions via Moment Matching

Learning from label proportions (LLP) is a weakly supervised setting where training data are grouped into bags with only bag-level label proportions observed, and the goal is to learn a classifier that predicts labels for individual instances. Recently, a large body of methods has emerged; however, most of them rely on the assumption that instance-label pairs are sampled i.i.d. and randomly assembled into bags, an assumption often violated in real-world scenarios. Our experimental results indicate that these methods perform suboptimally under non-random settings. In this work, we adopt a more realistic assumption: bags are sampled i.i.d., and instance labels are conditionally independent given the instances. Under this setting, we show that the label counts follow a Poisson multinomial distribution. Motivated by this observation, we propose LLP via moment matching (LLP-MM), a simple yet effective approach which leverages multi-order factorial moments as training objectives, encouraging the classifier predictions to match these moments and thereby more fully exploit the underlying statistical structure of the data. Extensive experiments on benchmark datasets under various bag construction strategies demonstrate the effectiveness of our approach while maintaining high computational efficiency. Our code is available at https://github.com/TianhaoMa5/LLP-MM.

cs.LG↗