# Research Papers — 2026-05 · bundle page 326 of 326

> Full-text markdown of papers #16251–16266 of 16266 published in 2026-05, newest first. AI-readable mirrors of the full 16266-paper 2026-05 archive (50 papers per file).

Canonical page: https://aipod.dev/papers/archive/2026-05/page/326
[← Previous bundle](https://aipod.dev/papers/archive/2026-05/page/325.md)

### AgentFloor: How Far Up the tool use Ladder Can Small Open-Weight Models Go?

Authors: Ranit Karmakar, Jayita Chatterjee
Published: 2026-05-01
Categories: cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2605.00334v1

Production agentic systems make many model calls per user request, and most of those calls are short, structured, and routine. This raises a practical routing question that existing evaluations do not directly answer: which parts of an agent workflow truly require large frontier intelligence, and which can be handled by smaller models? We introduce AgentFloor, a deterministic 30-task benchmark organized as a six-tier capability ladder, spanning instruction following, tool use, multi-step coordination, and long-horizon planning under persistent constraints. We evaluate 16 open-weight models, from 0.27B to 32B parameters, alongside GPT-5 across 16,542 scored runs. Our results reveal a clear boundary of model necessity. Small and mid-sized open-weight models are already sufficient for much of the short-horizon, structured tool use work that dominates real agent pipelines, and in aggregate, the strongest open-weight model matches GPT-5 on our benchmark while being substantially cheaper and faster to run. The gap appears most clearly on long-horizon planning tasks that require sustained coordination and reliable constraint tracking over many steps, where frontier models still hold an advantage, though neither side reaches strong reliability. We also find that this boundary is not explained by scale alone: some failures respond to targeted interventions, but the effects are model-specific rather than universal. These findings suggest a practical design principle for agentic systems: use smaller open-weight models for the broad base of routine actions, and reserve large frontier models for the narrower class of tasks that truly demand deeper planning and control. We release the benchmark, harness, sweep configurations, and full run corpus.

---

### Borrowed Geometry: Cross-Distribution Head-Importance Fingerprints of Frozen Pretrained Gemma 4 31B

Authors: Abay Bektursun
Published: 2026-05-01
Categories: cs.LG, cs.CL
arXiv: https://arxiv.org/abs/2605.00333v2

Frozen Gemma 4 31B weights pretrained exclusively on text, unmodified, transfer through a thin trainable interface to non-text modalities the substrate has never processed. On the L24--L29 slice (192 attention heads), an English-text TxtCopy attention probe (95 sentences) and per-head ablation impact on four non-language token-pattern tasks (binary copy, associative recall, 1D cellular automaton Rule 90, binary addition) jointly classify four heads -- L26.28, L27.28, L27.2, L27.3 -- as top-tier on both signals. The slice-level joint coincidence is significant under hypergeometric null ($P = 0.0013$, $N=192$, $K=38$, $n=4$) and survives multiplicity-aware permutation tests ($P_{V4} = 0.013$). Pretrained Gemma L26 reaches 60.22% on OGBench cube-double-play-task1 vs ~1% for random-init Gemma ($+59$pt at $n=3$); a FrozenRandom-GPT2 control with correct $1/\sqrt{d_k}$ scaling also fails. Head-level causal validation: zeroing L26.28 in the trained cube-task1 IQL agent drops success $63.3\% \to 10.0\%$ vs $46.7\%$ for a layer-matched low-TxtCopy negative control ($3.2\times$ specificity at $n=30$; $n=5$ paired-$t$ $p=0.039$). A full L26 sweep places L26.28 at rank 4 of 32. Honest negatives: within-L26 Spearman $ρ(\text{TxtCopy, drop}) = +0.37$ (opposite of within-layer causal reading); single-head activation patching does not transfer the matching variable; the 4 named heads alone do not suffice on any task; Walker2d-DT and scene-task1 recruit L24 outside the named slice and show null head-ablation specificity. We frame the contribution as a cross-distribution importance fingerprint at the slice level plus head-level causal evidence on one cross-modality target.

---

### Conformalized Quantum DeepONet Ensembles: Towards Scalable Operator Learning with Distribution-Free Guarantees

Authors: Purav Matlia, Christian Moya, Guang Lin
Published: 2026-05-01
Categories: cs.LG
arXiv: https://arxiv.org/abs/2605.00330v3

Operator learning enables fast surrogate modelling of high-dimensional dynamical systems, but existing approaches face two fundamental limitations: the quadratic cost of dense neural layers and unreliable uncertainty quantification in safety-critical settings. We propose Conformalized Quantum DeepONet Ensembles, a framework that addresses both challenges simultaneously. Using Quantum Orthogonal Neural Networks (QOrthoNNs), we characterize the resource regime in which their established $\widetilde{\mathcal{O}}(n)$ hidden-layer running-time scaling improves on the $\mathcal{O}(n^2)$ cost of a classical dense layer. To quantify uncertainty, we combine ensemble predictions with split conformal calibration. For a new input-output function pair jointly exchangeable with the calibration pairs, we prove a finite-sample, distribution-free lower bound on the expected fraction of covered query locations. As a secondary proof-of-concept, we explore hybrid classical-quantum architectures and superposed execution to manage ensemble runtime and hardware requirements. Experiments on synthetic operator benchmarks and real-world power-system dynamics show accurate predictions under ideal simulation and empirical coverage near the target under both ideal conditions and selected compact-circuit simulations with depolarizing or device-calibrated composite noise. Together, these results connect resource-aware scalability analysis with finite-sample uncertainty guarantees for quantum operator learning.

---

### Conformalized Quantum DeepONet Ensembles for Scalable Operator Learning with Distribution-Free Uncertainty

Authors: Purav Matlia, Christian Moya, Guang Lin
Published: 2026-05-01
Categories: cs.LG
arXiv: https://arxiv.org/abs/2605.00330v2

Operator learning enables fast surrogate modeling of high-dimensional dynamical systems, but existing approaches face two fundamental limitations: quadratic inference complexity and unreliable uncertainty quantification in safety-critical settings. We propose Conformalized Quantum DeepONet Ensembles, a framework that addresses both challenges simultaneously. By leveraging Quantum Orthogonal Neural Networks (QOrthoNNs), we reduce operator inference complexity from O(n^2) to O(n), enabling scalable evaluation over fine discretizations. To provide rigorous uncertainty quantification, we combine ensemble-based epistemic modeling with adaptive conformal prediction, yielding distribution-free coverage guarantees. A key challenge in ensembling is that naive parallelism scales hardware resources linearly with the number of models. We resolve this by using Superposed Parameterized Quantum Circuits (SPQCs), which compress multiple ensemble members into a single circuit and enable simultaneous multi-model execution. Experiments on synthetic partial differential equations and real-world power system dynamics demonstrate that our approach achieves accurate predictions while maintaining calibrated uncertainty under realistic quantum noise. These results establish a practical pathway toward scalable, uncertainty-aware operator learning in quantum machine learning.

---

### DynamicPO: Dynamic Preference Optimization for Recommendation

Authors: Xingyu Hu, Kai Zhang, Jiancan Wu, Shuli Wang, Chi Wang, Wenshuai Chen, Yinhua Zhu, Haitao Wang, Xingxing Wang, Xiang Wang
Published: 2026-05-01
Categories: cs.IR, cs.AI
arXiv: https://arxiv.org/abs/2605.00327v4

In large language model (LLM)-based recommendation systems, direct preference optimization (DPO) effectively aligns recommendations with user preferences, requiring multi-negative objective functions to leverage abundant implicit-feedback negatives and sharpen preference boundaries. However, our empirical analyses reveal a counterintuitive phenomenon, preference optimization collapse, where increasing the number of negative samples can lead to performance degradation despite a continuously decreasing training loss. We further theoretically demonstrate that this collapse arises from gradient suppression, caused by the dominance of easily discriminable negatives over boundary-critical negatives that truly define user preference boundaries. As a result, boundary-relevant signals are under-optimized, weakening the model's decision boundary. Motivated by these observations, we propose DynamicPO (Dynamic Preference Optimization), a lightweight and plug-and-play framework comprising two adaptive mechanisms: Dynamic Boundary Negative Selection, which identifies and prioritizes informative negatives near the model's decision boundary, and Dual-Margin Dynamic beta Adjustment, which calibrates optimization strength per sample according to boundary ambiguity. Extensive experiments on three public datasets show that DynamicPO effectively prevents optimization collapse and improves recommendation accuracy on multi-negative preference optimization methods, with negligible computational overhead. Our code and datasets are available at https://github.com/xingyuHuxingyu/DynamicPO.

---

### Prompt-Induced Score Variance in Zero-Shot Binary Vision-Language Safety Classification

Authors: Charles Weng, Dingwen Li, Alexander Martin
Published: 2026-05-01
Categories: cs.CL, cs.CV
arXiv: https://arxiv.org/abs/2605.00326v1

Single-prompt first-token probabilities from zero-shot vision-language model (VLM) safety classifiers are treated as decision scores, but we show they are unreliable under semantically equivalent prompt reformulation: even when the binary label is constrained to a fixed output position, equivalent prompts can induce materially different unsafe probabilities for the same sample. Across multimodal safety benchmarks and multiple VLM families, cross-prompt variance is strongly associated with prompt-level disagreement and higher error, making it a useful fragility diagnostic. A training-free mean ensemble improves NLL on all 14 dataset-model evaluation pairs and ECE on 12/14 relative to a train-selected single-prompt baseline, and wins more head-to-head NLL comparisons than labeled temperature scaling, Platt scaling, and isotonic regression applied to the same prompt. Ranking gains are consistent against the train-selected baseline on both AUROC and AUPRC, and against the full 15-prompt distribution remain consistent on AUPRC while softening on AUROC. Labeled calibration on top of the mean provides further gains when labels are available, identifying prompt averaging as a strong label-free first stage rather than a replacement for calibration. We frame this as a reliability stress test for zero-shot VLM first-token safety scores and recommend prompt-family evaluation with mean aggregation as a standard label-free reliability baseline.

---

### Intelligent Elastic Feature Fading: Enabling Model Retrain-Free Feature Efficiency Rollouts at Scale

Authors: Jieming Di, Xiaoyu Chen, Ying She, Siyu Wang, Lizzie Liu, Fenggang Wu, Jiaoying Mu, Tony Tsui, Amr Elroumy, Hsing Tang, Zewei Jiang, Qiao Yang et al.
Published: 2026-05-01
Categories: cs.IR, cs.LG
arXiv: https://arxiv.org/abs/2605.00324v1

Large-scale ranking systems depend on thousands of features derived from user behavior across multiple time horizons. Typically requires model retraining -- resulting in long iteration cycles (3--6 months), substantial GPU resource consumption, and limited rollout throughput. We introduce Intelligent Elastic Feature Fading (IEFF), a production infrastructure system that enables retrain-free feature efficiency rollouts by elastically controlling feature coverage and distribution at serving time. IEFF supports incremental feature coverage adjustments while models adapt through recurring training, eliminating dependencies on explicit retraining cycles. The system incorporates strict safety guardrails, reversibility mechanisms, and comprehensive monitoring to ensure stability at scale. Across multiple production use cases, IEFF accelerates efficiency-related rollouts by 5$\times$, eliminates retraining-related GPU overhead, and enables faster capacity recycling. Extensive offline and online experiments demonstrate that gradual feature fading prevents 50--55\% of online performance degradation compared to abrupt feature removal, while maintaining stable model behavior. These results establish elastic, system-level feature fading as a practical and scalable approach for managing feature efficiency in modern industrial ranking systems.

---

### Online Self-Calibration Against Hallucination in Vision-Language Models

Authors: Minghui Chen, Chenxu Yang, Hengjie Zhu, Dayan Wu, Zheng Lin, Qingyi Si
Published: 2026-05-01
Categories: cs.CV, cs.LG
arXiv: https://arxiv.org/abs/2605.00323v1

Large Vision-Language Models (LVLMs) often suffer from hallucinations, generating descriptions that include visual details absent from the input image. Recent preference alignment methods typically rely on supervision distilled from stronger models such as GPT. However, this offline paradigm introduces a Supervision-Perception Mismatch: the student model is forced to align with fine-grained details beyond its perceptual capacity, learning to guess rather than to see. To obtain reliable self-supervision for online learning, we identify a Generative-Discriminative Gap within LVLMs, where models exhibit higher accuracy on discriminative verification than open-ended generation. Leveraging this capability, we propose \textbf{O}nline \textbf{S}elf-\textbf{CA}lib\textbf{R}ation (OSCAR), a framework that integrates Monte Carlo Tree Search with a Dual-Granularity Reward Mechanism to construct preference data and iteratively refines the model via Direct Preference Optimization. Extensive experiments demonstrate that OSCAR achieves state-of-the-art performance on hallucination benchmarks while improving general multimodal capabilities.

---

### Federated Weather Modeling on Sensor Data

Authors: Shengchao Chen, Guodong Long
Published: 2026-05-01
Categories: cs.LG
arXiv: https://arxiv.org/abs/2605.00322v1

Federated weather modeling on sensor data is a distributed system underpinned by federated learning, enabling multiple sensor data sources, including ground weather stations, satellites and IoT devices, to collaboratively train deep learning models without sharing raw data. This method safeguards data privacy and security while leverages diverse, geographically distributed datasets to improve the accuracy and robustness of global/regional weather modeling tasks such as forecasting and anomaly detection.

---

### Structure-Aware Chunking for Tabular Data in Retrieval-Augmented Generation

Authors: Pooja Guttal, Varun Magotra, Vasudeva Mahavishnu, Natasha Chanto, Sidharth Sivaprasad, Manas Gaur
Published: 2026-05-01
Categories: cs.CL, cs.IR
arXiv: https://arxiv.org/abs/2605.00318v1

Tabular documents such as CSV and Excel files are widely used in enterprise data pipelines, yet existing chunking strategies for retrieval-augmented generation (RAG) are primarily designed for unstructured text and do not account for tabular structure. We propose a structure-aware tabular chunking (STC) framework that operates on row-level units by constructing a hierarchical Row Tree representation, where each row is encoded as a key-value block. STC performs token-constrained splitting aligned with structural boundaries and applies overlap-free greedy merging to produce dense, non-overlapping chunks. This design preserves semantic relationships between fields within a row while improving token utilization and reducing fragmentation. Across evaluations on the MAUD dataset, STC reduces chunk count by up to 40% and 56% compared to standard recursive and key-value based baselines, respectively, while improving token utilization and processing efficiency. In retrieval benchmarks, STC improves MRR from 0.3576 to 0.5945 in a hybrid setting and increases Recall@1 from 0.366 to 0.754 in BM25-only retrieval. These results demonstrate that preserving structure during chunking improves retrieval performance, highlighting the importance of structure-aware chunking for RAG over tabular data.

---

### Unbox Responsible GeoAI: Navigating Climate Extreme and Disaster Mapping

Authors: Hao Li, Steffen Knoblauch
Published: 2026-05-01
Categories: cs.CY, cs.AI
arXiv: https://arxiv.org/abs/2605.00315v1

As climate extreme and disaster events become more frequent and intense, Geospatial Artificial Intelligence (GeoAI) has emerged as a transformative approach for large-scale disaster mapping and risk reduction. However, the purely mechanical, performance-driven deployment of GeoAI models can result in amplifying inherent spatial inequalities, preventing effective emergency decision-making, and producing severe environmental carbon footprint. To unbox the concept of responsible GeoAI, this position paper examines its emerging role, e.g., in climate extreme and disaster mapping, from a critical GIS perspective. We address the nexus of responsible GeoAI into four interrelated theoretical dimensions, specifically Representativeness, Explainability, Sustainability, and Ethics, with examples from climate extreme and disaster mapping. Moreover, targeting at the operational practice, we then propose a conceptual governance Model of responsible GeoAI that categorizes its governance practices into Data, Application, and Society scopes. Last, this position paper aims to raise the attention in the broader GIS community that the future of climate resilience relies not just on building better algorithms, but on fostering a governance ecosystem where GeoAI is deployed responsibly, ethically, and sustainably.

---

### Semia: Auditing Agent Skills via Constraint-Guided Representation Synthesis

Authors: Hongbo Wen, Ying Li, Hanzhi Liu, Chaofan Shou, Yanju Chen, Yuan Tian, Yu Feng
Published: 2026-05-01
Categories: cs.CR, cs.AI, cs.PL
arXiv: https://arxiv.org/abs/2605.00314v1

An agent skill is a configuration package that equips an LLM-driven agent with a concrete capability, such as reading email, executing shell commands, or signing blockchain transactions. Each skill is a hybrid artifact-a structured half declares executable interfaces, while a prose half dictates when and how those interfaces fire-and the prose is reinterpreted probabilistically on every invocation. Conventional static analyzers parse the structured half but ignore the prose; LLM-based tools read the prose but cannot reproducibly prove that a tainted input reaches a high-impact sink. We present Semia, a static auditor for agent skills. Semia lifts each skill into the Skill Description Language (SDL), a Datalog fact base that captures LLM-triggered actions, prose-defined conditions, and human-in-the-loop checkpoints. Synthesizing a fact base that is both structurally sound and semantically faithful to the original prose is the central challenge; we address it with Constraint-Guided Representation Synthesis (CGRS), a propose-verify-evaluate loop that refines LLM candidates until convergence. Security properties (e.g., indirect injection, secret leakage, confused deputies, unguarded sinks, etc.) over an agent skill can then be reduced to Datalog reachability queries. We evaluate Semia on 13,728 real-world skills from public marketplaces. Semia renders all of them auditable and finds that more than half carry at least one critical semantic risk. On a stratified sample of 541 expert-labeled skills, Semia achieves 97.7% recall and an F1 of 90.6%, substantially outperforming signature-based scanners and LLM baselines.

---

### Beyond Structure: Revolutionising Materials Discovery via AI-Driven Synthesis Protocol-Property Relationships

Authors: Guillaume Lambard
Published: 2026-05-01
Categories: cond-mat.mtrl-sci, cs.AI
arXiv: https://arxiv.org/abs/2605.00313v1

The current structure-centric paradigm in artificial intelligence (AI)-driven materials discovery, despite delivering thousands of candidate structures, is stalling at a critical barrier: the synthesizability gap. We argue that closing this gap demands a pivot to a synthesis-first paradigm in which executable synthesis protocols, not just atomic configurations, are treated as primary design variables. We outline a roadmap built on three pillars: (i) representing synthesis procedures as machine-readable protocols, (ii) deploying generative and inverse-design models to propose actionable reaction pathways and recipes, and (iii) integrating closed-loop optimisation to refine protocols against experimental realities and sustainability constraints. Framed in terms of the causal backbone P->X->y from protocol P to structure X and properties y, this perspective sets out methodological building blocks, standards needs and self-driving laboratory (SDL) integration strategies to accelerate reproducible, data-first materials discovery.

---

### Beyond Visual Fidelity: Benchmarking Super-Resolution Models for Large-Scale Remote Sensing Imagery via Downstream Task Integration

Authors: Zhili Li, Kangyang Chai, Zhihao Wang, Xiaowei Jia, Yanhua Li, Gengchen Mai, Sergii Skakun, Dinesh Manocha, Yiqun Xie
Published: 2026-05-01
Categories: cs.CV, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2605.00310v2

Super-resolution (SR) techniques have made major advances in reconstructing high-resolution images from low-resolution inputs. The increased resolution provides visual enhancement and utility for monitoring tasks. In particular, SR has been increasingly developed for satellite-based Earth observation, with applications in urban planning, agriculture, ecology, and disaster response. However, existing SR studies and benchmarks typically use fidelity metrics such as PSNR or SSIM, whereas the true utility of super-resolved images lies in supporting downstream tasks such as land cover classification, biomass estimation, and change detection. To bridge this gap, we introduce GeoSR-Bench, a downstream task-integrated SR benchmark dataset to evaluate SR models beyond fidelity metrics. GeoSR-Bench comprises spatially co-located, temporally aligned, and quality-controlled image pairs from about 36,000 locations across diverse land covers, spanning resolutions from 500m to 0.6m. To the best of our knowledge, GeoSR-Bench is the first SR benchmark that directly connects improved image resolution from SR models with downstream Earth monitoring tasks, including land cover segmentation, infrastructure mapping, and biophysical variable estimation. Using GeoSR-Bench, we benchmark GAN, transformer, neural operator, and diffusion-based SR models on perceptual quality and downstream task performance. We conduct experiments with 270 settings, covering 2 cross-platform SR tasks, 9 SR models, 3 downstream task models, and 5 downstream tasks for each SR task. The results show that improvements in traditional SR metrics often do not correlate with gains in task performance, and the correlations can be negative, indicating that these metrics provide limited guidance for selecting superior models for downstream tasks. This reveals the need to integrate downstream tasks into SR model development and evaluation.

---

### A Model-based Visual Contact Localization and Force Sensing System for Compliant Robotic Grippers

Authors: Kaiwen Zuo, Shuyuan Yang, Zonghe Chua
Published: 2026-05-01
Categories: cs.RO, cs.CV
arXiv: https://arxiv.org/abs/2605.00307v1

Grasp force estimation can help prevent robots from damaging delicate objects during manipulation and improve learning-based robotic control. Integrating force sensing into deformable grippers negotiates trade-offs in cost, complexity, mechanical robustness, and performance. With the growing integration of RGB-D wrist cameras into robotic systems for control purposes, camera-based techniques are a promising solution for indirect visual force estimation. Current approaches mostly utilize end-to-end deep learning, which can be brittle when generalizing to new scenarios, while existing model-based approaches are unsuited to grasping and modern grasper geometries. To address these challenges, we developed a model-based visual force sensing approach integrating an iterative contact localization with generalization to unseen objects. The system extracts structural key points from wrist camera RGB-D images of deforming fin-ray-shaped soft grippers, and uses these key points to define parameters of an inverse finite element analysis simulation in Simulation Open Framework Architecture. The iterative contact localization sub-system utilizes a deep learning-based online 3D reconstruction and pose estimation pipeline to dynamically update contact location, and is robust to visual occlusion and unseen objects. Our system demonstrated an average root mean square error of 0.23 N and normalized root mean square deviation of 2.11% during the load phase, and 0.48 N and 4.34% over the entire grasping process when interacting with different objects under various conditions, showcasing its potential for real-time model-based indirect force sensing of soft grippers.

---

### Token Arena: A Continuous Benchmark Unifying Energy and Cognition in AI Inference

Authors: Yuxuan Gao, Megan Wang, Yi Ling Yu
Published: 2026-05-01
Categories: cs.AI, cs.DC, cs.LG, cs.PF
arXiv: https://arxiv.org/abs/2605.00300v1

Public inference benchmarks compare AI systems at the model and provider level, but the unit at which deployment decisions are actually made is the endpoint: the (provider, model, stock-keeping-unit) tuple at which a specific quantization, decoding strategy, region, and serving stack is exposed. We introduce TokenArena, a continuous benchmark that measures inference at endpoint granularity along five core axes (output speed, time to first token, workload-blended price, effective context, and quality on the live endpoint) and synthesizes them, together with a modeled energy estimate, into three headline composites: joules per correct answer, dollars per correct answer, and endpoint fidelity (output-distribution similarity to a first-party reference). The framework's novelty is empirical and methodological. Across 78 endpoints serving 12 model families, the same model on different endpoints differs in mean accuracy by up to 12.5 points on math and code, in fingerprint similarity to first party by up to 12 points, in tail latency by an order of magnitude, and in modeled joules per correct answer by a factor of 6.2. We further show that workload-aware blended pricing reorders the leaderboard substantially: 7 of 10 top-ranked endpoints under the chat preset (3:1 input:output) fall out of the top 10 under the retrieval-augmented preset (20:1), and the reasoning preset (1:5) elevates frontier closed models that the chat preset penalizes on price. We release the framework, schema, probe and eval harness, and a v1.0 leaderboard snapshot under CC BY 4.0. TokenArena is a methodology, not a single ranking; we publish full provenance and limitations and welcome external replication.

---