Searcharxiv⌕ Search

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 145 records · Page 8Linked to original sources
arXiv2026

Predicting Supermassive Black Hole-Host Mass Offsets from Broadband Photometry Across Cosmological Simulations with Forecasts for LSST

The possibility of over-massive black holes suggested by James Webb Space Telescope photometric discoveries of 'little red dots', may disfavor light supermassive black hole (SMBH) seeds. However, what should constitute the mass (range) of 'heavy' seeds remains relatively unconstrained. Moreover, Vera C Rubin Observatory's Legacy Survey of Space and Time will photometrically characterize galaxies without direct black hole mass measurements. We forward-model the SIMBA, IllustrisTNG, and EAGLE cosmological simulations into the photometric bands of LSST to train an ensemble machine learning classifier. Our framework achieves $91\%$--$94\%$ accuracy across SIMBA and IllustrisTNG in distinguishing between over-massive and under-massive SMBH growth regimes under LSST magnitude limits, using only broadband photometry. Furthermore, cross-simulation transfer experiments (training on one cosmological simulation and evaluating on another using rank-normalized features) achieve $83\%$--$89\%$ accuracy. This suggests the relative photometric ordering of growth regimes is largely preserved even across fundamentally different sub-grid SMBH feedback prescriptions. Signal decomposition shows our classification is driven by host galaxy colors ($82\%$--$87\%$ accuracy) and, relatedly, the accretion-state's spectral energy distribution shape as opposed to an inversion of our forward model's analytical luminosity prescription. Given that the evaluated simulations employ heavy seed prescriptions ($\geq 10^{4}~M_\odot$), our methodology establishes a validated baseline for classifying post-seeding growth regimes.

astro-ph.GA↗
arXiv2026

Representation Before Training: A Practical Benchmark for Generative Medical Event Model Tokenization

Generative medical event models use tokenized sequences of patient timelines as input, but practical guidance on the many decisions around tokenization is limited. We benchmark quantization granularity, reference-range anchoring, code--value fusion, numeric and temporal encodings, and native versus harmonized event representations from an expert-mapped common data model. Using both Llama and Qwen architectures, 156 models were trained on full hospitalizations from three initialization seeds, with each configuration following a shared training recipe for up to five epochs. We evaluated learned representations from the first 24 hours of hospitalization with linear probes to predict binary and continuous outcomes during hours 24-48. Fused tokens pairing codes with value deciles increased performance across all eight outcome families relative to the equivalent unfused tokenized input with area under the receiver operating characteristic curve (AUROC) gains of $+0.002$ to $+0.033$ and Spearman correlation gains of $+0.025$ to $+0.114$. Neither anchoring value bins to reference ranges nor increasing quantization granularity consistently improved performance, while xVal variants underperformed both discrete and soft encodings. Alternatives to explicit time tokens, such as event-order and admission-relative rotary position embeddings (RoPE), yielded higher family-mean point estimates across all eight families while reducing input length. When evaluating native input against input mapped to the Common Longitudinal Intensive Care Unit Data Format (CLIF), the CLIF full-hospitalization training sequences contained 28.6% as many tokens as the native sequences and improved performance across six of eight outcome families. These findings show that tokenization and event encoding are consequential design choices when learning patient representations for downstream classification and regression tasks.

cs.LG↗
arXiv2026

Evolution as fitness landscape navigation: concepts, measures, and emerging questions

Fitness landscapes are mappings between genotypes, phenotypes, and fitness that shape evolution. In recent years, empirical work and theoretical models have greatly advanced our understanding of how populations navigate rugged fitness landscapes. Here, we provide a timely review of the theoretical aspects of this field. Its rapidly growing literature employs a wide range of terms, which are sometimes used ambiguously or inconsistently. We therefore begin by defining the major concepts and the field's vocabulary, highlighting our own terminology choices wherever needed. We then review key results on the relationships between epistasis, ruggedness, accessibility, and navigability for genotype-fitness maps, highlighting several complex and sometimes counterintuitive connections that have emerged. Further, we review how the conserved structural properties of the underlying genotype-phenotype map, which can lead to the formation of large connected neutral networks of genotypes, influence dynamics on fitness landscapes. We then compare the two levels to study landscape navigation: the level of genotype-phenotype maps and the level of genotype-fitness maps. Our review leads us to propose a new measure of navigability, based on evolutionary outcomes, that is broadly applicable and overcomes limitations of existing measures. Finally, we highlight examples from the smaller body of work that relaxes the common assumption of fitness-monotonic paths on static landscapes, and discuss how this can fundamentally change the nature of fitness landscape navigation. Throughout the review, we identify directions for future work to fill existing gaps and to synthesize the disparate strands of research within the field.

q-bio.PE↗
arXiv2026

On quantum functionals for higher-order tensors

Upper and lower quantum functionals, introduced by Christandl, Vrana and Zuiddam (STOC 2018, J. Amer. Math. Soc. 2023), are families of monotone functions of tensors indexed by a weighting on the set of subsets of the tensor legs. Inspired by quantum information theory, they were crafted as obstructions to asymptotic tensor transformations, relevant in algebraic complexity theory. For tensors of order three, and more generally for weightings on singletons for higher-order tensors, the upper and lower quantum functionals coincide and are spectral points in Strassen's asymptotic spectrum. Moreover, the singleton quantum functionals characterize the asymptotic slice rank, whereas general weightings provide upper bounds on asymptotic partition rank. It has been an open question whether the upper and lower quantum functionals also coincide for other cases, or more generally, how to construct further spectral points, especially for higher-order tensors. In this work, we show that upper and lower quantum functionals generally do not coincide, but that they anchor new spectral points. With this we mean that there exist new spectral points, which equal the quantum functionals on the set of tensors on which upper and lower coincide. The set is shown to include embedded three-tensors and W-like states and concerns all laminar weightings, significantly extending the singleton case. Moreover, it is shown that these spectral points provide obstructions to asymptotic restriction beyond the previously known spectral points.

math.AG↗
arXiv2026

Cross Subtype Transferability of Machine Learning Photometric Redshift Relations in Low Redshift Seyfert AGN

Photometric redshift estimation for active galactic nuclei (AGN) is complicated by the combined effects of host-galaxy light, nuclear emission, dust attenuation, and broadband spectral diversity. We investigate whether machine learning photo-z relations trained on one low-redshift Seyfert subtype remain valid when transferred to another, and whether probabilistic subtype classification can be used to identify sources for which a specialised regressor is reliable. Using spectroscopically selected Seyfert I and Seyfert II samples from SDSS, matched to AllWISE photometry over 0 < z_spec <= 0.6, we constructed a common 45-feature representation from SDSS ugriz and WISE W1-W4 data. Random Forest and XGBoost regressors were evaluated within each subtype, followed by controlled cross-subtype transfer tests, redshift and sample size-matched experiments, feature ablations, and an independent classifier-gated regression test. The subtype specific models achieved strong within-sample performance, with the Seyfert II model reaching R2 = 0.965 and sigma_NMAD = 0.0169. However, transfer between Seyfert I and Seyfert II produced a clear and asymmetric degradation in accuracy that persisted after matching the samples and restricting the photometric inputs. A probabilistic Seyfert classifier further identified subsets for which the Seyfert II regressor was more reliable, while extrapolation beyond the redshift range represented in training produced systematic underestimation. These results demonstrate that AGN photo-z performance depends strongly on the population and redshift domain represented in the training data, supporting subtype-aware calibration and applicability-based source selection.

astro-ph.GA↗
arXiv2026

Flavon assisted low scale leptogenesis

Low-scale leptogenesis scenarios, such as the resonant leptogenesis, typically require a highly degenerate mass spectrum of right-handed neutrinos (RHNs). This requirement can be circumvented by extending the seesaw framework with a scalar singlet $S$ that couples to RHNs via the $S N^{}_I N^{}_J$ terms (with $I \neq J$), which opens up new decay channels $N^{}_I \to N^{}_J S$ and provides additional sources of CP violation, thereby enabling successful leptogenesis at the TeV scale without the need for mass degeneracy. In this work, for the first time, we point out that the flavon fields, which are introduced in many flavor-symmetry neutrino mass models to be responsible for the generation of RHN masses through the acquisition of non-zero vacuum expectation values, serve as ideal candidates for the $S$ field. Taking as an example a flavor-symmetry neutrino mass model that naturally realizes the experimentally allowed TM1 mixing pattern and has the attractive features that only one flavon field plays the role of $S$ and that it couples to only two RHNs, we demonstrate that the observed neutrino masses and mixing angles can be consistently reproduced, while the observed baryon asymmetry can be achieved within a parameter space compatible with current experimental constraints.

hep-ph↗
arXiv2026

Exceptional poles of archimedean Rankin-Selberg L-functions for irreducible generic representations of GL(n,R)

For irreducible generic representations $π_1$ and $π_2$ of $\operatorname{GL}_n(\mathbb R)$, we prove that the notions of exceptional pole of type $1$ and type $2$ coincide at every level. When both representations are in general position, we use this identification to express the Rankin--Selberg $L$-function $L(s,π_1\timesπ_2)$ in terms of the exceptional $L$-factors attached to the irreducible constituents of their derivatives.

math.NT↗
arXiv2026

A bivariate cure copula model with zero-inflated gamma frailty for right-censored data

In biomedical studies, paired survival data arise naturally when two event times are observed within the same subject. Existing statistical models seldom accommodate both cure fractions and complex dependence structures. In this paper, we propose a novel bivariate cure frailty-copula model for paired survival data with a cure fraction. By incorporating a zero-inflated gamma frailty, the proposed framework simultaneously accommodates a cure fraction and continuous unobserved heterogeneity among uncured subjects. Dependence between cure statuses is modeled naturally via an odds-ratio parameter, while dependence between survival times conditional on frailty is captured through a copula. To provide interpretable measures of overall dependence in the presence of cure, we derive population-level rank correlation coefficients, namely tie-adjusted versions of Kendall's tau and Spearman's rho. For suitable choices of baseline marginal survival functions and copulas, the joint survival function admits a closed-form expression, enabling maximum likelihood estimation and likelihood ratio testing. Simulation studies demonstrate satisfactory finite-sample performance and favorable comparisons with existing bivariate cure models. In a real-data analysis, the proposed model is favored by information criteria. An R package, curecopula, implementing the proposed methods is publicly available on GitHub.

stat.ME↗
arXiv2026

JudgeSense: A Benchmark for Prompt Sensitivity in LLM-as-a-Judge Systems

Large language models are widely used to judge the output of other language models, yet whether a judge returns the same verdict when the same request is worded differently remains largely unexamined. We study that question across four evaluation tasks and twenty-five judges from six providers. To support the analysis we release JudgeSense, a benchmark of 880 items from human-labelled corpora, each issued under two instructions that differ in wording and not in what they ask, with the complete decision logs. Every score is reported against the judge's own agreement with itself on the identical prompt, so decoding noise is not charged to wording, and the release lets a reader ask the same of any judge not in our roster. Rewording costs agreement on all four tasks, and on two it clears the threshold we declare for a practically meaningful effect; the ordinal task is both the least stable and the one fewest judges are accurate on, and within a single family parameter count does not predict stability. A judge measured inside an agent harness yields a smaller estimate than the same judge reached through a direct API call, because its agreement with itself collapses faster than its agreement across wordings.

cs.CL↗
arXiv2026

Analysis and numerical simulations of a landfast ice model

In this manuscript, we consider a common modeling framework for Arctic landfast ice based on the work of Lemieux et al. [27], which is designed for use in large-scale climate models. This approach extends the classical viscous-plastic sea-ice model introduced by Hibler [18], which remains the most widely used model for simulating large-scale sea-ice dynamics. In particular, landfast ice refers to sea-ice that is attached to the coastline or grounded and therefore exhibits nearly vanishing motion. The model considered in this manuscript augments Hibler's viscous-plastic sea-ice model by tensile strength and a threshold-dependent basal stress. We present a rigorous analytical and numerical study of this landfast ice model. The analytical contributions are the local strong well-posedness, the global strong well-posedness in the absence of external forces and for initial data close to constant equilibrium solutions, and the existence of time-periodic solutions. The limiting case $k_{\mathrm{t}} \equiv 1$ is treated separately, since the pressure coupling then vanishes and a new smallness condition on the stress regularization parameter is required. For the system without scalar diffusion, we establish local strong well-posedness through a Lagrangian formulation and anisotropic maximal $\mathrm{L}^p$-regularity. Finally, we perform numerical simulations that illustrate key qualitative differences between landfast ice and classical viscous-plastic sea-ice models. The simulations reveal the formation of stationary equilibrium states characterized by vanishing ice velocity. These observations are consistent with the global-in-time existence result close to equilibria established in Theorem 4.1 and the time-periodic result in Theorem 5.2. The combined analytical and numerical results provide new insight into the structure, stability, and long-term behavior of landfast ice dynamics.

math.AP↗
arXiv2026

Personalizing Causal Audio-Driven Facial Motion via Dynamic Multi-modal Retrieval

Audio-driven facial animation is essential for immersive digital interaction, yet existing frameworks struggle to reconcile real-time streaming with high-fidelity personalization. Current methods either rely on latency-inducing audio look-ahead, or ask users to record scripted calibration sequences to pre-encode static identity embeddings that fail to capture dynamic idiosyncrasies. We present an end-to-end framework for personalized audio-driven facial motion generation, supporting causal, zero-lookahead streaming. We introduce two key innovations: (1) a causal multi-resolution motion tokenizer that captures both global temporal context and high-frequency articulatory details, and (2) a multi-modal style retriever that extracts stylistic priors from unstructured reference libraries by jointly querying ongoing audio and motion. Unlike prior retrieval mechanisms restricted to curated, fixed-size, or audio-only style banks, our design accepts arbitrary footage of the target identity, enabling personalization from a handful of casually recorded clips. By integrating these components, our method outperforms state-of-the-art approaches in lip-sync accuracy, identity consistency, and perceived realism, while preserving the streaming constraints of real-time telepresence. Code is available at https://github.com/xg-chu/Fallingwater.

cs.GR↗
arXiv2026

Decoding Delay Guarantees of Space Regulated Multiple Access Random Wireless Networks using Successive Interference Cancellation

This paper is focused on decoding delay guarantees in wireless networks, where messages have a given signal-to-interference-plus-noise ratio threshold $η_0$ to meet in order to be successfully decoded, and where transmissions should occur within some strict time constraints. Its main contribution consists in quantifying the worst-case transmissions decoding delays in the uplink of cellular and cell-free networks using successive interference cancellation. We show how such decoding delay guarantees can be obtained using spatial network calculus, a new tool introduced recently, and in particular spatial regulation. The results rely on the assumption of absence of fading. We nevertheless outline what this approach will lead to in the fading case for cellular networks.

cs.NI↗
arXiv2026

Tree-independence number of $P_5$-free graphs with no large bicliques

The tree-independence number of a graph is the minimum, over all tree-decompositions of the graph, of the maximum size of an independent set contained in a bag. Graph classes of bounded tree-independence number have strong structural and algorithmic properties; however, the parameter can be unbounded even in quite restricted classes. In particular, the presence of an induced biclique $K_{\ell,\ell}$ forces tree-independence number at least $\ell$. This leads to the question whether large induced bicliques are the only obstruction to bounded tree-independence number in natural hereditary classes. A conjecture of Dallard, Krnc, Kwon, Milanič, Munaro, Štorgel, and Wiederrecht states that for all positive integers $t$ and $\ell$, ${\{P_t,K_{\ell,\ell}\}}$-free graphs have bounded tree-independence number. We prove this conjecture for ${t=5}$ by showing that every ${\{P_5,K_{\ell,\ell}\}}$-free graph has tree-independence number at most ${4\ell-4}$. We also obtain related bounds for the weaker parameter of $α$-degeneracy and answer a question of Hilaire, Milanič, and Vasić whether tree-independence number of ${\{P_5,K_{\ell,\ell}\}}$-free graphs exceeds $\ell$ by at most an additive constant.

math.CO↗
arXiv2026

On Unbiased Parameter Estimation and Signal Reconstruction

In this paper, we extend the theory of depth-unbiased source localization to unbiased parameter estimation and signal reconstruction for an arbitrary number of non-zero parameters. The topic touches on exact reconstructibility, most commonly studied in compressed sensing and multisource estimation across various imaging problems. The theoretical results derive upper bounds on the number of recoverable parameters in the noiseless case, and define a probability measure to assess the likelihood of recovering all non-zero parameters with correct magnitude order. The work provides a mathematical explanation of the open question regarding the noise robustness of standardized and unbiased methods. The paper also reveals a trade-off between the number of sensors and the signal-to-noise ratio. Numerical experiments demonstrate the theoretical findings.

cs.IT↗
arXiv2026

On the inhomogeneous discounted Hamilton-Jacobi equations

In this paper, we study the family of inhomogeneous discounted Hamilton-Jacobi equations \begin{equation}\label{hjs1} λ(x)u+h(x,d_x u)=c \quad \tag{$\ast$} \end{equation} on a closed manifold $M$ with a non-identically vanishing discount factor $λ(x)$. There is a critical value $c_0\in[-\infty,\infty)$ such that \eqref{hjs1} admits a viscosity solution if $c>c_0$ and no solution if $c 0$ , we show that the equation admits an asymptotically stable solution if and only if $c>c_0$. In this case, we determine the basin of the stable solution and investigate the long time behavior of the solution semigroup associated to \eqref{hjs1}. In particular, we relate the lowest convergence rate to the integral of $λ$ over Mather measures, which leads to an asymptotic behavior of Mather measures when $c$ goes to infinity. Assuming $c\geqslant c_0$ and the equation admits a solution, we classify ergodic Mather measures and locate their distribution in the phase space.

math.AP↗
arXiv2026

Constraint Decay: The Fragility of LLM Agents in Backend Code Generation

Large Language Model (LLM) agents demonstrate strong performance in autonomous code generation under loose specifications. However, production-grade software requires strict adherence to structural constraints, such as architectural patterns, databases, and object-relational mappings. Existing benchmarks often overlook these non-functional requirements, rewarding functionally correct but structurally arbitrary solutions. We present a systematic study evaluating how well agents handle structural constraints in multi-file backend generation. By fixing a unified API contract across 80 greenfield generation tasks and 20 feature-implementation tasks spanning eight web frameworks, we isolate the effect of structural complexity using a dual evaluation with end-to-end behavioral tests and static verifiers. Our findings reveal a phenomenon of constraint decay: as structural requirements accumulate, agent performance exhibits a substantial decline. Evaluated configurations lose 27.28 points on average in assertion pass rates from baseline to fully specified tasks. Framework sensitivity analysis exposes performance disparities: mid-tier models succeed in minimal, explicit frameworks (e.g., Flask) but perform substantially worse on average in convention-heavy environments (e.g., FastAPI, Django). Finally, error analysis identifies data-layer defects (e.g., incorrect query composition and ORM runtime violations) as the leading root causes. This work highlights that jointly satisfying functional and structural requirements remains a key open challenge for coding agents.

cs.SE↗
arXiv2026

BoostAPR: Boosting Automated Program Repair via Execution-Grounded Reinforcement Learning with Dual Reward Models

Reinforcement learning for program repair is hindered by sparse execution feedback and coarse sequence-level rewards that obscure which edits actually fix bugs. We present BoostAPR, a three-stage framework addressing these challenges: (1) supervised fine-tuning on execution-verified demonstrations with reasoning traces, (2) training dual reward models--a sequence-level assessor and a line-level credit allocator--from execution outcomes, and (3) PPO optimization where the line-level model redistributes rewards to critical edit regions. This line-level credit assignment operates at an intermediate granularity naturally suited to code changes. Trained on SWE-Gym and evaluated on four benchmarks, BoostAPR achieves 40.7% on SWE-bench Verified (+22.9pp over base model), 24.8% on Defects4J (Python-to-Java transfer), 84.5% on HumanEval-Java, and 95.0% on QuixBugs, achieving competitive results among open-source models with strong cross-language generalization.

cs.AI↗