Search arXiv⌕ Search

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 1,405 records · Page 78Linked to original sources
arXiv2026

Agentic Knowledge-Model Co-Evolution via Consistency Regularization

LLM agents can evolve by learning from accumulated experience. Existing approaches extract reusable external knowledge, such as skills and memories, from experience to guide subsequent task solving, with some further internalizing this knowledge through model training. However, different ways of analyzing and transforming the same experience can yield knowledge with substantially different utility, even when the resulting guidance appears reasonable. Inspired by consistency regularization, we identify consistent utility gains across different ways of analyzing the same experience, rather than relying on any single way of analyzing it. We introduce SPLICE, a framework for consistency-regularized knowledge-model co-evolution. It constructs evolution views by perturbing how the same experience is analyzed, producing different knowledge candidates. Knowledge updates are then guided by common gain, which accounts for utility gains and their consistency across views. During model evolution, self-distillation internalizes the evolved knowledge alongside reinforcement learning, using knowledge associated with consistent utility degradation as a contrastive reference. The two processes interleave, with model updates driving further knowledge evolution. Experiments across diverse domains show that SPLICE outperforms knowledge evolution and model training baselines by up to 7.1 and 4.6 points in average accuracy, while its evolved knowledge reduces output tokens per correct solution by up to 44% and transfers across model sizes and families.

cs.AI↗
arXiv2026

Gandalf: Flash Shall Not Die! Increasing Flash Storage Lifetime for KV Cache Offloading in LLM Serving

KV caching reduces the serving cost of large language models (LLMs), but the growing KV cache footprint is increasingly exceeds GPU and host-memory capacity. Therefore, people use SSD to expand the KV cache storage capacity with high I/O bandwidth. However, in production LLM workload, such as multi-round conversation and LLM agents, the total amount of KV cache writes significantly exceeds SSD capacity, shortening SSD lifetime from years to months. To improve SSD lifetime, a natural solution is to reduce the KV cache writes by gating the KV cache writes via historical access patterns. However, that significantly lowers the KV cache hit rate (and increase the LLM serving cost) since KV caches from different attention types might have identical historical access pattern but different future access patterns. Our insight is that, to reduce KV cache writes without hurting KV cache hit rate, one need to not only consider the access pattern, but also application-level information such as the attention type of KV cache. Leveraging this observation, we propose Gandalf , the first SSD KV cache offloading system with SSD life time awareness. Gandalf prioritize the admission of full-attention KV cache into SSD over the cache generated by other types of attention, which significantly increases SSD lifetime without sacrificing cache hit rate. Our evaluation shows that Gandalf extends the SSD lifetime by 7.3X on average, while having negligible impact on cache hit rate across multiple workload traces and wide range of hardware capacity. Gandalf design can be extended to other flash-based storage, such as High Bandwidth Flash (HBF).

cs.LG↗
arXiv2026

Trace2World: Reconstructing App Worlds from Real Interaction Traces for Scalable GUI Agent Training

Scaling GUI reinforcement learning requires environments that preserve real-app behavior. Yet coupling behavior inference with coding can turn partial traces into runnable mocks with incorrect state transitions. We introduce \textbf{Trace2World}, which separates global behavior reconstruction from implementation. It reconciles traces into a shared, source-linked app model, then compiles an executable environment. Independent replay of source workflows localizes mismatches for bounded implementation repair. Validated worlds then support learner-adaptive task and environment augmentation within source-supported behavior. From 1,570 traces, Trace2World constructs 50 environments. On 200 workflows across ten apps, 91.1% are reproduced end to end, versus 59.8% for the strongest matched task-local baseline. Source-grounded validation retains 94.0% of simulated training gains on real devices, versus 54.7% under self-validation. Learner guidance raises benchmark success from 56.9% to 61.1%. Training in Trace2World further improves the model by 20.2 points on our benchmark, 5.4--16.2 points across four external GUI benchmarks, and 16.3 points on 100 real-device tasks. Reliable reconstruction separates what behavior source evidence supports from whether the implementation preserves it.

cs.LG↗
arXiv2026

Advancing oxygen-enhanced MRI acquisition and modelling through dynamic $ΔR_{1}$ and $ΔR_2^*$ sensing

Purpose: To develop a multiparametric mapping framework for quantifying fat- and water-specific longitudinal and effective transverse relaxation rates, as well as changes in these parameters induced by a gas challenge. Theory and Methods: The proposed framework, termed SR-Fat DESPOT (single-reference fat-water separated driven equilibrium single-pulse observation of $T_1$), combines a two-component signal model with a single-reference variable flip angle (SR-VFA) approach. The signal model explicitly incorporates gas challenge-induced changes in relaxation rates. The single-reference acquisition strategy reduces data acquisition requirements while preserving sensitivity to both baseline parameters and dynamic changes. Sequence parameters were optimized using Cramer-Rao lower bound theory. SR-Fat DESPOT was evaluated using a modified open-surce abdominal numerical phantom and in two in vivo experiments: a single participant test-retest experiment and a pilot study with six participants. Results: In the abdominal numerical phantom, SR-Fat DESPOT was accurate for quantification of baseline relaxation rates with a normalized median absolute error (nMedAE) <3% for signal-to-noise ratio (SNR) >30 when comparing ground truth and estimates. SR-Fat DESPOT was also able to accurately recover median changes in relaxation rates, with differences within 0.001 $s^-1$ when compared to ground truth. In vivo, SR-Fat DESPOT measured changes in relaxation rates for spleen, kidneys, liver, and muscle under 100% O2 breathing, consistent with reports in literature. Changes detected in the spleen and kidneys were statistically significant across participants. Conclusion: SR-Fat DESPOT can quantify changes in relaxation rates induced by a gas challenge while reducing data requirements compared to more conventional approaches used in MR-oximetry literature.

physics.med-ph↗
arXiv2026

VLTest: Finding Joint Image-Text Failures in Vision-Language Models

Vision-language models (VLMs) are increasingly embedded in software systems. Testing them raises two problems. Mutating either input can change the expected answer, so an output change is not by itself a failure. Some failures appear only when both inputs are mutated; attributing them to the combination requires testing each mutation alone. We present VLTest, a black-box VLM testing approach. It generates image variants by editing VQGAN latent codes and text variants with a rewrite-expand-prune pipeline, filters variants likely to change the expected answer, and tests each variant alone before testing combinations, so every joint failure has four matched cells. A VLM judge confirms a failure only after checking that the input keeps the expected answer. On three tasks with eight VLMs, including two commercial API models, the oracle confirms 81.7% of the candidates, with 88.8% to 94.6% precision against human labels, and four-cell confirmation rejects 22.5% of joint candidates. Under the same budget, VLTest's generators find 1.8 to 2.5 times as many confirmed failures as SINVAD and TextFooler, and its joint tests find 91 to 197 attributed joint failures per 1,000 joint calls, 1.3 times as many as the same pairing applied to the baselines' variants.

cs.SE↗
arXiv2026

A Modified Direct Eigenvalue Swapping Algorithm with Guaranteed Stability

For a given matrix \( A \) in the real Schur form, the eigenvalue swapping algorithm is used to swap two adjacent diagonal blocks. In LAPACK, this operation is performed by the subroutine~\texttt{SLAEXC}/\texttt{DLAEXC}, which is based on a direct swapping algorithm proposed by Bai and Demmel in 1993. In this work, we present a thorough error analysis of a modified direct swapping algorithm. By a carefully designed pivoted Givens QR factorization, this algorithm can achieve an error bound of the form \( O( \boldsymbol{u} \lVert A \rVert ) \) under mild assumptions, where \( \boldsymbol{u} \) is the unit roundoff. We present empirical evidence to illustrate the superior stability of our algorithm over existing approaches.

math.NA↗
arXiv2026

GIFT: Gradient Informed Filtering for Distillation Augmented Reasoning RL

Training large language models for complex reasoning tasks relies on two complementary signals: sparse, outcome-grounded reinforcement learning (RL) for correctness, and dense, token-level distillation for guidance. However, complementarity in supervision does not guarantee compatibility in optimization; imitation-focused distillation can conflict with the outcome-driven RL objective, and existing hybrids lack a principled resolution. We propose \textbf{GIFT} (\textbf{G}radient-\textbf{I}nformed \textbf{F}il\textbf{T}ering for Distillation-Augmented Reasoning RL), a novel method that directly addresses the allocation problem of which teacher updates to apply. GIFT performs a token-level geometric analysis, decomposing the distillation gradient into a component \textbf{parallel} to the RL gradient, which measures first-order conflict and provides a \textbf{safety} filter, and a component \textbf{orthogonal} to it, which quantifies novel directional guidance and provides a measure of \textbf{value}. This decomposition yields a principled scoring rule to rank token-level guidance. We combine this with a progressive \textbf{coverage annealing} schedule that shrinks the budget of tokens receiving distillation, ensuring that the most conflicting updates are discarded first while the most complementary ones are retained at full strength. This facilitates a smooth transition from dense, geometrically-filtered guidance to a pure, outcome-grounded objective at convergence. Extensive experiments on mathematical and logical reasoning benchmarks show that GIFT consistently outperforms strong baselines across multiple model scales, achieving relative gains of up to 9.5\% on difficult mathematical reasoning tasks. Code: https://github.com/lilifree/GIFT

cs.LG↗
arXiv2026

OpenAI partially disproves Littlewood's 1961 $L^α$-flat theorem

We present a counterexample, produced by OpenAI, to Littlewood's 1961 theorem on $L^α$-flat polynomials. The theorem asserts that no $L^α$-flat polynomials exist within the class of trigonometric polynomials of the form$$\sum_{m=1}^{n} a_m \cos(mθ+α_m), \qquad n \in \mathbb{N}^*,$$satisfying$$k\, n^2 \sum_{m=1}^{n} a_m^2 \;\leq\; \sum_{m=1}^{n} m^2 a_m^2,$$where $k$ is an absolute constant (alternatively expressed on the right-hand side as $K = 1/k$). We show that OpenAI's counterexample is contained within the class of hyper-lacunary sequences. Consequently, it offers only a limited refutation, as such sequences fail to satisfy an implicit assumption of Littlewood's theorem. We also discuss broader questions surrounding this counterexample and provide a brief historical overview.

math.NT↗
arXiv2026

Amplifying Latency and Cost of LLM Inference via Rowhammer Attack

Large language models (LLMs) achieve high inference efficiency using optimized GPU kernels generated by just-in-time (JIT) compilation frameworks such as Triton. Triton compiles GPU kernels through an MLIR/LLVM-based pipeline. We investigate whether a Rowhammer-induced bit-flip in the CPU's DRAM can degrade GPU-based LLM inference performance while preserving output correctness. We present LLVMhammer, the first Rowhammer attack that causes LLM inference slowdown by inducing a bit-flip in Triton's compilation pipeline. A bit-flip in a performance-critical optimization pass causes Triton to generate a functionally identical but suboptimal GPU kernel. Since Triton caches the generated kernel on disk, a single successful bit-flip causes persistent slowdown across subsequent inference executions. Triton's target-independent stages are shared across GPU vendors, which makes our attack GPU vendor-agnostic. To identify exploitable bit-flip locations without manual inspection, we build Karnage, a tool that isolates useful bit-flip locations in Triton's target-independent compilation pipeline. We evaluate LLVMHammer across a diverse set of LLMs, including Qwen2.5-7B-Instruct, Llama-3.1-8B-Instruct, Mistral-7B-Instruct-v0.3, and Deepseek-llm-7B-chat, across NVIDIA RTX 4090 and AMD RX 7900 XTX GPUs. LLVMhammer increases inference latency of an LLM inference server running on a single GPU by 4-16\%, which, in turn, can translate into additional annual operating costs at the scale of multi-GPU deployments. We further show that LLVMhammer can also corrupt LLM outputs across different GPU vendors, reducing BERTScore to as low as 25\%, illustrating the broader security implications of bit-flips in Triton's compilation pipeline.

cs.CR↗
arXiv2026

WovenAnchor Matcher: Specialized Intra- and Inter-Image Context Modeling for Feature Matching

Semi-dense local feature matching commonly aggregates contextual information before coarse-to-fine correspondence estimation. However, intra-image and inter-image aggregation serve different roles: the former should propagate spatial evidence within each image while retaining structured two-dimensional dependencies, whereas the latter should gather information from plausible corresponding regions while limiting noise from unrelated locations. We propose WovenAnchor Matcher (WAM), a semi-dense matching framework based on role-specialized context aggregation. WAM introduces two complementary operators. WovenMamba performs intra-image aggregation through horizontal-then-vertical state-space propagation, allowing vertical updates to operate on horizontally contextualized features and thereby encouraging a structured two-dimensional receptive-field bias. AnchorCrossAttention performs inter-image aggregation by using cross-attention to gather context from plausible corresponding regions across the image pair. It performs this exchange over local representative anchors that aggregate feature evidence within each window. On MegaDepth, WAM achieves 66.1 pose AUC@5$^\circ$, improving on the JamMa baseline by 2.0 AUC points under the same evaluation setup. Ablations show that removing or replacing either specialized component reduces accuracy, providing evidence that intra-image propagation and inter-image retrieval benefit from role-specific aggregation designs. Code and pretrained weights are available at https://github.com/mmaiLab/WovenAnchorMatcher.

cs.CV↗
arXiv2026

When are metrizable linear subspaces of topological vector spaces finite-dimensional?

We study conditions that imply finite-dimensionality of metrizable or Frechet-Urysohn linear subspaces of topological vector spaces (TVS). We consider the classes $\mathscr{D}_M$ and $\mathscr{D}_{FU}$ of TVS: a space $L$ belongs to $\mathscr{D}_M$ (respectively, $\mathscr{D}_{FU}$) if every metrizable (respectively, Frechet-Urysohn) linear subspace of $L$ is finite-dimensional. Although the classes of metrizable spaces and of Frechet-Urysohn spaces differ substantially, the classes $\mathscr{D}_M$ and $\mathscr{D}_{FU}$ are much closer to each other: in the Hrusak-Ramos-Garcia model, in which Malykhin's problem is solved, $\mathscr{D}_M=\mathscr{D}_{FU}$, whereas under MA+not CH, as shown in the paper, $\mathscr{D}_M\neq\mathscr{D}_{FU}$. For a locally convex space (LCS) $L$ we denote by $L_w$ the space $L$ with its weak topology and by $L'_{w^*}$ the dual space with the weak$^*$ topology. This paper investigates which LCS $L$ have the property that $L_w$ and $L'_{w^*}$ belong to $\mathscr{D}_M$ or $\mathscr{D}_{FU}$. Although the classes are close, different methods are used to study them: topological-algebraic ones for $\mathscr{D}_M$ (a criterion is obtained) and methods of measure theory and probability for $\mathscr{D}_{FU}$ (a sufficient condition is obtained). We prove that if an LCS $L$ admits a continuous norm, then $L_w\in\mathscr{D}_{FU}$; in particular, if $L$ admits a continuous norm and $L_w$ is Frechet-Urysohn, then $L$ is finite-dimensional. If the topology of an LCS $L$ is coarser than the topology induced by some complete norm on $L$ (for example, if $L$ is Banach), then $L'_{w^*}\in\mathscr{D}_{FU}$; completeness of the norm is essential here. The methods also apply to spaces of real-analytic functions and to the space of polynomials with the topology of pointwise convergence: these also belong to $\mathscr{D}_{FU}$.

math.GN↗
arXiv2026

A Computationally Efficient and Rate-Optimal Algorithm for Adversarial Linear MDPs with (Aggregate) Bandit Feedback

We study episodic linear Markov decision processes with unknown transitions, adversarial losses, and bandit feedback. The state-of-the-art result for this setting (Liu et al., 2024) either achieves $\sqrt{K}$ regret with a computationally inefficient algorithm or achieves suboptimal $K^{3/4}$ regret, where $K$ is the number of episodes. Our work gives the first $\sqrt{K}$ regret with a polynomial-time algorithm. Our method reduces adversarial linear MDPs to adversarial linear bandits whose action set consists of policies' expected feature vectors. The central challenge is that this action set is unknown and can be estimated only through exploration. We address this challenge by constructing an approximate separation oracle for the action set that either generates executable policies from queried features, or refines an outer approximation of the action set. This approach draws inspiration from the reduction of van Erven et al. (2025) for linear contextual bandits and the oracle-efficient linear bandit algorithm of Ito et al. (2019), but additionally deals with the imprecision of the oracle. The resulting algorithm requires only aggregate loss feedback, and to our knowledge. This is the first polynomial-time algorithm for adversarial MDPs with aggregate feedback beyond the tabular case.

cs.LG↗
arXiv2026

Is spectral predictability enough? Testing Ω-matching for synthetic pretraining under distribution shift

Synthetic pretraining under distribution shift requires deciding what synthetic data to use. A natural criterion is to match the spectral predictability $Ω$ of the source to the target, but is $Ω$ sufficient? We test this hypothesis, which we call predictability-aware pretraining, across three architectures (LSTM, PatchTST, TCN) and three seeds. Under ideal synthetic-to-synthetic conditions, $Ω$-matched pretraining can outperform even an oracle baseline (+46.8% MSE reduction, LSTM), though the benefit is already architecture-dependent. On the real-world ETTm1 benchmark, however, this relationship does not hold: sine-plus-noise pretraining at matched $Ω$ yields only a marginal mean improvement +2.3% averaged over all three architectures, up to +65.5% on individual channels), whereas $Ω$ matched chaotic ODE trajectories systematically degrade performance -13.2%). The two sources agree on the direction of transfer for only 53% of configurations across all three architectures (Pearson r=-0.02), indicating that $Ω$ carries little information about which source will transfer beneficially. We further observe an architecture-dependent asymmetry in the response to the direction of the $Ω$ gap. We conclude that $Ω$ matching can help in controlled settings but is not sufficient to predict transfer across sources, and we discuss additional signal properties that a source-selection criterion may need to consider.

cs.LG↗
arXiv2026

What Lies Beyond Perception: A Partition Theory of the Limits of AI Evaluation and Oversight

Replacing the judge can move a benchmark score, and models can be trained to hide capabilities. Both put the evaluator in question: what can it certify? We develop a partition theory of evaluation from one principle: judgement is bounded by perception. An agent's discrimination is the partition of its inputs into classes it cannot tell apart. Discrimination is necessary for capability but not sufficient, so bounds on it bear directly on whether evaluation can show that a model lacks an ability. We prove a one-sided bound. An examinee whose partition refines its evaluator's has the discrimination to pass every test the evaluator can grade, provided each input has an acceptable reply. We define the echo test, which accepts any reply the evaluator perceives as identical to the input, and show that it is complete for the tests the evaluator can grade. Passing therefore certifies at most a floor on discrimination, never a ceiling, and only a sharper evaluator can raise it. Used to detect concealment, echo-test failure points the wrong way: under exhaustive probing it flags every honestly weaker examinee and no concealer that presents as the evaluator's peer. Exact imitation of a weaker model is undetectable by any behavioural evaluator. We extend the bound to ensembles of monitors and to channels a model controls, explain published findings on judge disagreement and sandbagging, and derive testable predictions. A safety case that argues from behavioural evaluation alone that a model lacks an ability must assume the model's cooperation.

cs.AI↗
arXiv2026

Numerical Solution of Large-Scale Low-Rank Coupled Generalized Sylvester Equations

This paper develops projection methods for computing low-rank solutions to large-scale coupled generalized Sylvester equations. The dominant subspaces of these solutions are shown to be spanned by specific resolvent directions, motivating a Petrov--Galerkin projection framework. Conditions for the unique solvability of the projected coupled generalized Sylvester equations are also discussed. Based on these dominant subspaces, two algorithms are proposed: an extended Krylov-based projection algorithm and a rational Krylov-based projection algorithm. The former uses fixed shifts, whereas the latter uses adaptive shifts. Numerical experiments on large-scale benchmark problems demonstrate that the proposed methods achieve high accuracy and fast convergence.

math.NA↗
arXiv2026

Distinct stellar-parameter dependencies of granulation and mesogranulation in red giants

Granulation and mesogranulation represent different spatial and temporal scales of convection, but are commonly assumed to follow the same stellar atmospheric scaling relations in red giants. We test this assumption using the largest sample assembled for this purpose to date -- 9,328 Kepler red giants spanning the red giant branch and red clump phase, cross-matched with APOGEE stellar parameters. We model each star's power spectrum background to obtain the radius-corrected amplitude and characteristic frequency of both components, and derive empirical scaling relations with effective temperature, surface gravity, and metallicity through Bayesian model comparison. Gaussian Process and symbolic regression provide independent validation. We find that granulation and mesogranulation are not self-similar: their surface-gravity dependencies are statistically indistinguishable, consistent with a shared envelope geometry, but their temperature dependencies diverge: mesogranulation's amplitude rises steeply with Teff, whereas granulation's shows no statistically justified dependence, and both frequencies respond weakly and similarly. We propose that this pattern is consistent with granulation forming in a thin near-surface layer, where radiation takes over from convection and the Mach number depends only weakly on Teff. Mesogranulation originates deeper, where convection carries the full flux and its Mach number and cell size increase with Teff. Metallicity raises both amplitudes but has little effect on the frequencies. Our results indicate that the self-similarity assumption underlying two-component background models does not hold, with direct implications for background modeling and its downstream application in Galactic archaeology.

astro-ph.SR↗
arXiv2026

A Weakly Group Strategy-Proof and Pareto-Stable Matching Mechanism with Two-Sided Indifferences under Gross Substitutes

We study finite many-to-one matching markets with indifferences on both sides and publicly known firm preferences over sets of workers. A standard approach breaks ties uniformly at random and runs deferred acceptance, which is strategy-proof for workers but may select a weakly stable matching that is Pareto dominated under the agents' original preferences. For the broad class of markets in which each firm's preferences admit a gross substitutes (GS) representation, we construct a deterministic generalized deferred acceptance mechanism that is weakly stable, Pareto efficient with respect to both sides, worker weakly group strategy-proof, and implementable in strongly polynomial time given value oracles for the GS representations. This domain includes nonresponsive firm preferences, such as additive scores under matroid constraints and assignment values. Our analysis bridges ordinal matching with indifferences and competitive equilibria under imperfectly transferable utility. Although the mechanism uses only ordinal comparisons, it returns a seller-optimal equilibrium matching of a tiered auxiliary market, and the equilibrium structure of this market yields the stability, efficiency, and incentive guarantees.

cs.GT↗
arXiv2026

Global Convergence of Detailed Balanced Mass Action Systems

We give a self-contained proof that, for a finite detailed balanced mass-action system with positive rate constants, each positive stoichiometric compatibility class contains exactly one positive equilibrium and every positive trajectory converges to it, together with a Lean 4 formalization of the complete argument from persistence to convergence. Boundedness and persistence come from the barrier argument of OpenAI (2026) for weakly reversible systems: an arbitrary-tolerance affine approximation yields a finite-minimum barrier whose superlevel sets confine trajectories and give one logarithmic bound for all solutions from a given initial state. The Horn-Jackson dissipation identity identifies the positive equilibria as an exponential family with parameters in the orthogonal complement of the stoichiometric subspace, which gives uniqueness in each class. A bounded-speed descent argument replaces LaSalle's invariance principle; it uses neither uniqueness of solutions nor continuous dependence on initial data, and it places every cluster point in that family. The route given here is direct and elementary, and its main theorem is checked by the Lean kernel against explicit definitions. The proofs, the Lean development, and the manuscript were produced in ClawsGO Science, an agent harness that packages a complete research environment: Claude Opus 5.5 made the research decisions, GPT 6 Astra carried them out, and the Lean kernel checked every formal step.

cs.LO↗