# Research Papers — 2026-09 · bundle page 259 of 266

> Full-text markdown of papers #12901–12950 of 13298 published in 2026-09, newest first. AI-readable mirrors of the full 13298-paper 2026-09 archive (50 papers per file).

Canonical page: https://aipod.dev/papers/archive/2026-09/page/259
[← Previous bundle](https://aipod.dev/papers/archive/2026-09/page/258.md)
[Next bundle →](https://aipod.dev/papers/archive/2026-09/page/260.md)

### EvoSCM: Scientific Belief Revision Through Causal Model Evolution and Experimentation

Authors: Qing Zhao, Haowei Li, Weijian Deng, Sibei Yang, Pengxu Wei, Liang Lin
Published: 2026-09-01
Categories: cs.AI
arXiv: https://arxiv.org/abs/2609.01526v2

Scientific discovery depends on the ability to form hypotheses, test them through experiments, and revise them when evidence disagrees. Existing LLM agents support this process by improving their reasoning or actions, but their scientific beliefs are often scattered across free-form reasoning and difficult to update coherently. This makes it difficult to identify what failed, what should change, and whether revisions remain consistent with prior evidence. We introduce EvoSCM, which represents scientific beliefs as a population of structural causal model (SCM) hypotheses that can be tested and revised across experiments. EvoSCM formulates scientific discovery as a closed loop in which causal hypotheses guide experimentation and experimental outcomes drive causal model evolution. Competing SCM hypotheses make falsifiable predictions and guide discriminative experiments that separate alternative explanations. When observations contradict these predictions, EvoSCM distills discrepancies into correction rules identifying which aspects of the hypotheses fail to explain the evidence. These rules guide revisions to causal dependencies, latent factors, mechanisms, and parameters. Revised hypotheses are validated against accumulated evidence and carried forward to guide subsequent experiments, allowing scientific beliefs to evolve cumulatively. We evaluate EvoSCM across physics, chemistry and materials, and biology. It consistently outperforms baseline agents and existing evolution methods, yielding more accurate explanations and predictions with more effective use of experimental budgets. The evolved SCMs also transfer across base models, suggesting reusable scientific knowledge beyond any single model's reasoning process.

---

### EvoSCM: Scientific Belief Revision Through Causal Model Evolution and Experimentation

Authors: Qing Zhao, Haowei Li, Weijian Deng, Pengxu Wei, Liang Lin
Published: 2026-09-01
Categories: cs.AI
arXiv: https://arxiv.org/abs/2609.01526v1

Scientific agents must learn not only how to reason, but also what to believe. However, existing LLM agents typically express scientific hypotheses in free-form text, leaving their beliefs implicit and difficult to test or revise. We introduce EvoSCM, which equips scientific agents with explicit structural causal models that evolve as new experimental evidence is collected. EvoSCM maintains a population of competing SCM hypotheses, each encoding a candidate causal explanation of the environment, and evolves them through a closed discovery loop. In each round, the agent abduces latent mechanisms from accumulated evidence, designs discriminative interventions, and commits to falsifiable predictions that it tests through experimentation. Discrepancies between prediction and observation are inductively distilled into correction rules that revise the causal structures and mechanisms of each hypothesis, and the agent then deductively validates the revised population against accumulated evidence and structural consistency to guide the next round. We evaluate EvoSCM on DiscoverPhysics, a benchmark requiring agents to uncover the hidden dynamics of noncanonical physical worlds through experimentation. EvoSCM consistently improves scientific discovery over baselines, yielding more accurate explanations and predictions while making more effective use of experimental interactions.

---

### Relational-Core Graph Analytics Querying graphs at SQL scale, and why the node/edge model is a performance tax, not a truer picture of connected data

Authors: Gene Zhang
Published: 2026-09-01
Categories: cs.DB, cs.AI, cs.PL
arXiv: https://arxiv.org/abs/2609.01525v1

A durable assumption holds that graph analytics requires a purpose-built graph engine, and that relational systems are ill-suited to connected data. We argue the opposite for the workloads enterprises actually run. A columnar relational engine fronted by a graph query language matches or exceeds native graph engines on analytical graph queries, and - decisively - scales past the point where in-memory graph engines fail. We further argue that the node/edge property graph is not a more faithful model of connected data but a re-encoding of relationships that already exist explicitly in relational tables; reconstructing them at query time is pure overhead. We present ClickGraph and its Databricks-dialect sibling DeltaGraph, systems that translate Cypher directly onto the native relational schema - the tables, columns, and foreign keys as they already exist - and execute in place on ClickHouse, Databricks, or in-process on lakehouse files, with no import and no separate cluster. Because the output is ordinary SQL, an underperforming query is an open optimization surface: it can be rewritten, and the engine itself extended. We support the argument with a peer system's own published benchmark, in which a columnar engine outruns Neo4j by two-to-four orders of magnitude, and with reproducible measurements across the LDBC Social Network Benchmark suite.

---

### When Guardrails Look Effective: Construct Validity Failures in LLM Agent Commerce Evaluation

Authors: Peiying Zhu, Sidi Chang
Published: 2026-09-01
Categories: cs.AI
arXiv: https://arxiv.org/abs/2609.01519v1

Interactive simulations increasingly evaluate policies in markets populated by language-model agents. Their outputs can look economic---prices, profits, consumer surplus, and welfare---without instantiating the behavior named in the claim. We audit this risk in a multi-turn buyer--seller testbed for configurable hotel transactions. An initial implementation reported welfare gains from two marketplace guardrails of +87.4, +35.0, and +28.8 across a Qwen2.5 1.5B--14B ladder. It also gave guarded and unguarded agents different offer schemas and choice procedures. Holding the schema and buyer chooser fixed changes the paired contrasts to +7.2, -13.9, and +23.8. The four largest 14B single-generation effects averaged +229; after three generations per profile-condition, they averaged +37.6 (95% bootstrap interval (-34.2, 109.3)), while generation residuals account for 49.9% of variation in this post-hoc probe. A seller-incentive check is non-monotone: increasing profit pressure produces less profit than the default seller prompt. Scripted positive controls show why this matters. A profit-maximizing seller already attains first-best welfare, so guardrails mostly redistribute and reduce welfare; they create welfare only when the seller is explicitly programmed to force inefficient bundles. We contribute a construct-validity contract separating incentive validity, protocol isolation, stochastic stability, and welfare accounting, and returning INVALID or INCONCLUSIVE before substantive policy claims. In our case, the original estimate is INVALID under protocol isolation, while the controlled study remains INCONCLUSIVE under incentive validity and stochastic stability. The case does not show that guardrails are ineffective; it shows their apparent value is unidentified until the simulated agents and protocol pass these checks.

---

### DualDiff3D: Dual Structure-Appearance Diffusion Priors for Reliability-Enhanced 3D Gaussian Splatting

Authors: Qian Wang, Yu Wang, Weiqi Li, Xinhua Cheng, Xiandong Meng, Ronggang Wang, Jian Zhang
Published: 2026-09-01
Categories: cs.CV
arXiv: https://arxiv.org/abs/2609.01516v1

While 3D Gaussian Splatting (3DGS) has revolutionized 3D reconstruction and novel-view synthesis, scenarios with limited input views often lead to poor reconstruction quality and artifacts in rendered novel views. Recent efforts attempt to utilize powerful diffusion priors, yet they typically process rendered and reference views concatenated along an additional dimension in a single network. These methods overlook an inherent nature that different views should maintain appearance similarity but differ in structure due to view shifts, leading to blur caused by conflicts between the two properties. In this paper, we propose DualDiff, a novel pipeline that leverages dual diffusion priors with a Structure-Appearance Attention (SAA) module to introduce reference guidance for refining low-quality novel views rendered from flawed 3D representations. Specifically, we retain one diffusion branch to focus on extracting structural information from the low-quality novel views, while introducing another branch to ensure appearance consistency with reference views. Furthermore, we present a 3D reconstruction framework named DualDiff3D, which integrates a reliability-enhanced Render-Refine-Optimize (RRO) loop to progressively and robustly incorporate the refined novel views, yielding more accurate 3DGS. Extensive experiments demonstrate that our approach outperforms state-of-the-art methods even in the inference-only setting, with further performance gains achievable through training. Our code and pre-trained weights are available at https://github.com/Akaneqwq/DualDiff3D.

---

### TempCloze: Can Video-LLMs Identify the Missing Middle?

Authors: Wenqi Pei, Henry Hengyuan Zhao, Yilai Liu, Jiahao Meng, Han Chen, Ziyu Wang, Hongyang Du
Published: 2026-09-01
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2609.01515v1

Temporal reasoning benchmarks for Video-LLMs are often mediated by language, leaving room for linguistic shortcuts from option wording, answer correlations, or language priors. To reduce such shortcuts, we introduce TempCloze, a video cloze benchmark for evaluating visual temporal reasoning in Video-LLMs. Given the beginning and ending clips of a video, models must identify the true missing middle from four candidates. TempCloze contains 1,521 carefully filtered videos from seven sources, mainly long-take and egocentric videos. We construct same-source distractors along three dimensions: Semantic asks what event should happen, Alignment probes when it should occur, and Progression tests how it should unfold, while shared scenes and objects reduce appearance cues. Our evaluation of 10 proprietary and 21 open-source Video-LLMs reveals Alignment as the primary bottleneck: models often recognize plausible semantic content and local event progression but struggle with temporal alignment. We further conduct error pattern and behavioral sensitivity analyses on TempCloze-Mixed and TempCloze-Hard with four representative models to examine where errors arise and how candidate order, context direction, visible span, frame density, and test-time scaling influence model choices.

---

### A Sensor-Adaptive Incremental Learning Framework for Artifact Detection in Satellite Precipitation Data

Authors: Andres F. Monsalve, Hernan A. Moreno, Christian D. Kummerow
Published: 2026-09-01
Categories: physics.ao-ph, cs.CV, stat.AP
arXiv: https://arxiv.org/abs/2609.01514v1

Historically, retrieving rainfall data from satellite imagery has been the domain of space agencies. However, in recent years, the development of cheaper, more compact satellites (SmallSats) capable of detecting rainfall proxies has led to a significant increase in private-sector initiatives for satellite launch and surface precipitation products. This rapid growth has yet to be matched by data validation efforts. Consequently, the need for a robust tool to detect anomalies in near-real-time data before it is disseminated to the public has become critical. In this paper, we present the development of an anomaly-detection system to identify artifacts in global satellite-based rainfall products. The developed framework leverages pre-trained computer vision models and incorporates scarce human-labeled data to detect specific anomalies. Our proposed anomaly detection strategy is tested on data from the Special Sensor Microwave Imager (SSMI) and the Special Sensor Microwave Imager/Sounder (SSMIS). Results demonstrate the efficacy of our approach at separating regular orbits from artifact-containing orbits for each satellite, with performance comparable to state-of-the-art in-place methods. Additionally, the framework offers explainability and the capacity for iterative refinement following false-positive or false-negative classifications.

---

### Benchmarking Spatial, Spectral, and Self-Supervised Cues for Face Forgery Detection under Realistic Degradation

Authors: Lucas Cunha, Lucas Sotomaior, Lucas Gasperin, Beatriz Caldas, Eduardo Pianovski, Rayson Laroca
Published: 2026-09-01
Categories: cs.CV
arXiv: https://arxiv.org/abs/2609.01511v1

Face forgery detectors often achieve strong results on controlled benchmarks, but their reliability under realistic image degradations remains limited. This paper presents a standardized benchmark for face forgery detection using the Multi-Dimensional Face Forgery Image (MFFI) dataset and evaluates performance on both clean and degraded test partitions. We compare six model families, including convolutional networks, transformer-based models, and a frozen self-supervised DINOv3 backbone, across spatial, spectral, and hybrid input representations. The results show that clean-set performance is not a reliable indicator of robustness under compression, resizing, and blurring. Xception with RGB obtains the best clean performance, reaching 0.884 mean ROC-AUC, but degrades substantially on the harder partition. In contrast, frozen DINOv3 achieves the strongest degraded-set result, with 0.726 mean ROC-AUC, while training only a linear classification head. The representation analysis indicates that Fourier-domain cues are most useful when combined with RGB information, whereas purely spectral inputs consistently underperform spatial representations. Qualitative attribution maps further suggest that convolutional detectors focus on localized artifacts, while DINOv3 relies on broader facial structure. These findings reinforce the need for degraded evaluation protocols and highlight self-supervised visual representations as a promising direction for robust face forgery detection. Our source code is publicly available at https://github.com/lucasdocunha/FaceForgery-Benchmark/.

---

### LatentPress: Context Compression Beyond Text and Vision

Authors: Zhengze Zhou, Hejian Sang
Published: 2026-09-01
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2609.01507v2

Compressed context is usually carried as human-readable text or as rendered images that must be decoded, even when its consumer is a language model. We introduce LatentPress, which writes conversational histories and long documents into a third representation: continuous memory tokens that a frozen decoder reads directly through its input-embedding interface, with no text reconstruction at inference. A small reader-matched writer compresses $4$-$16\times$ while training only an adapter (4.2M-26.2M parameters, $\sim\!0.1\%$ of the decoder). On LongMemEval, LatentPress reaches $0.504$ accuracy at $7.70\times$ compression versus $0.490$ for uncompressed evidence, outperforming text summaries (0.184) and OCR-based compression (0.426 to 0.312). On LongBench-QA, in-domain writers match or exceed raw-context reading at $4$-$8\times$ compression, while $16\times$ trails raw. Writing takes 43ms per conversation, roughly an order of magnitude faster than text summarization or OCR reconstruction, and reading is $5$-$9\times$ faster than raw context or cached OCR. We validate the interface under two transfer settings, zero-shot from UltraChat to LongMemEval memory QA and from LongMemEval-derived QA to unseen LongBench document domains, establishing direct soft tokens as a practical machine-facing context interface beyond text and vision. The implementation of the experiments could be found at: https://github.com/HJSang/LatentPress .

---

### LatentPress: Context Compression Beyond Text and Vision

Authors: Zhengze Zhou, Hejian Sang
Published: 2026-09-01
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2609.01507v1

Compressed context is usually carried as human-readable text or as rendered images that must be decoded, even when its consumer is a language model. We introduce LatentPress, which writes conversational histories and long documents into a third representation: continuous memory tokens that a frozen decoder reads directly through its input-embedding interface, with no text reconstruction at inference. A small reader-matched writer compresses $4$-$16\times$ while training only an adapter (4.2M-26.2M parameters, $\sim\!0.1\%$ of the decoder). On LongMemEval, LatentPress reaches $0.504$ accuracy at $7.70\times$ compression versus $0.490$ for uncompressed evidence, outperforming text summaries (0.184) and OCR-based compression (0.426 to 0.312). On LongBench-QA, in-domain writers match or exceed raw-context reading at $4$-$8\times$ compression, while $16\times$ trails raw. Writing takes 43ms per conversation, roughly an order of magnitude faster than text summarization or OCR reconstruction, and reading is $5$-$9\times$ faster than raw context or cached OCR. We validate the interface under two transfer settings, zero-shot from UltraChat to LongMemEval memory QA and from LongMemEval-derived QA to unseen LongBench document domains, establishing direct soft tokens as a practical machine-facing context interface beyond text and vision. The implementation of the experiments could be found at: https://github.com/xuyd16ai/context_softtoken_compress .

---

### Optimizing Byzantine Node Placement in Decentralized Federated Learning

Authors: Edoardo Gabrielli, Gabriele Tolomei
Published: 2026-09-01
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2609.01495v1

Security evaluations of decentralized federated learning (DFL) typically focus on how Byzantine participants behave, while largely overlooking which participants are compromised. Yet, because aggregation is distributed over a communication graph, the placement of Byzantine nodes determines how malicious influence propagates through the network. We therefore treat Byzantine placement as an explicit adversarial decision and formulate the attacker's objective as selecting, under a fixed compromise budget, the set of participants that maximizes its finite-time impact on honest nodes. To approximate this objective without executing the learning process for every candidate placement, we introduce Byzantine Placement Influence (BPI), a set-level measure derived from the actual gossip dynamics that quantifies the cumulative exposure of honest nodes to Byzantine sources over the training horizon. Unlike placement criteria based on node centrality heuristics, BPI directly accounts for weighted multi-hop propagation and interactions among compromised nodes. We develop efficient algorithms for optimizing BPI and evaluate them across six heterogeneous graph families, untargeted model poisoning, and backdoor attacks. BPI-guided placements consistently identify highly damaging configurations across different network structures and remain effective when the linear gossip assumption is relaxed through Byzantine-robust aggregation. Our results show that Byzantine placement is a critical but under-modeled dimension of DFL threat models and robustness evaluations.

---

### Rethinking Learnability in Offline Data-driven Optimization

Authors: Chao Qian, Chen-Guang Wang, Rong-Xi Tan, Ke Xue
Published: 2026-09-01
Categories: cs.LG, cs.AI, cs.NE
arXiv: https://arxiv.org/abs/2609.01493v2

Black-Box Optimization (BBO) has broad applications, while traditional algorithms such as evolutionary algorithms and Bayesian optimization face efficiency challenges as real-world BBO problems grow increasingly complex. Data-driven optimization has been the most popular paradigm to improve the efficiency of BBO, by learning from data. Offline data-driven optimization seeks high-quality solutions using only a fixed set of previous evaluations, attracting substantial attention because it requires no additional online evaluations. Many offline optimization methods have been proposed, but a fundamental question remains unanswered: what learnability is sufficient for offline optimization? Prior theoretical studies show that Probably Approximately Correct (PAC) learnability is insufficient, as the optimal region may remain poorly learned even when most regions are well learned. In this paper, we propose algorithm-dependent learnability, which requires accuracy only on the optimizer's trajectory. We prove that its value-query form is sufficient for representative discrete settings, including greedy and local search for submodular maximization, while its first-order analogue is sufficient for projected gradient descent on convex minimization. Motivated by this notion, we formalize a trajectory-learning framework comprising trajectory construction, trajectory modeling, and candidate generation, and analyze existing trajectory-based methods under it. We further propose Uncertainty-aware Gradient-guided Trajectory Learning (UGTL), which constructs locally coherent improvement trajectories reflecting plausible search paths, models them with conditional diffusion, and selects a diverse candidate set. Our experiments show that UGTL achieves the best average rank, 3.1/25, among 25 methods on Design-Bench tasks, and confirm that our trajectory construction plays a significant role in the improvement.

---

### Rethinking Learnability in Offline Data-driven Optimization

Authors: Chao Qian, Chen-Guang Wang, Rong-Xi Tan, Ke Xue
Published: 2026-09-01
Categories: cs.LG, cs.AI, cs.NE
arXiv: https://arxiv.org/abs/2609.01493v1

Black-Box Optimization (BBO) has found broad applications, but evolutionary algorithms and Bayesian optimization face efficiency challenges as real-world BBO problems grow increasingly complex. Data-driven optimization improves the efficiency of BBO algorithms by learning from data. Offline data-driven optimization seeks high-quality solutions using only a fixed set of previous evaluations, attracting substantial attention because it requires no additional online evaluations. Many offline optimization methods have been proposed, but a fundamental question remains unanswered: what learnability is sufficient for offline optimization? Prior theoretical studies show that Probably Approximately Correct (PAC) learnability is insufficient, as the optimal region may remain poorly learned even when most regions are well learned. In this paper, we propose algorithm-dependent learnability, which requires accuracy only on the optimizer's trajectory. We prove that its value-query form is sufficient for representative discrete settings, including greedy and local search for submodular maximization, while its first-order analogue is sufficient for projected gradient descent on convex minimization. Motivated by this notion, we formalize a trajectory-learning framework comprising trajectory construction, trajectory modeling, and candidate generation, and analyze existing trajectory-based methods under it. We further propose Uncertainty-aware Gradient-guided Trajectory Learning (UGTL), which constructs locally coherent improvement trajectories reflecting plausible search paths, models them with conditional diffusion, and selects a diverse candidate set. On five Design-Bench tasks, UGTL achieves the best aggregate mean rank, $3.1/25$, among 25 methods. Controlled trajectory analyses and cross-architecture replacements confirm that our trajectory construction plays a significant role in the improvement.

---

### GlossoGen: Emergent Language in Complex Multi-Agent LLM Interactions

Authors: Elias Stengel-Eskin, Newton Sander, Carlos Bonetti, Sasha Boguraev, James Bowler, Hale Sirin, Simon Kirby
Published: 2026-09-01
Categories: cs.CL, cs.AI, cs.MA
arXiv: https://arxiv.org/abs/2609.01491v1

The growing rate at which LLM agents interact with one another raises key questions about language evolution in multi-LLM-agent settings, with implications for safety and monitorability as well as for linguistic accounts of LLMs. To address these questions, we introduce GlossoGen, a novel platform for studying multi-agent language evolution in complex scenarios. Within GlossoGen, we build the SaveVeyru scenario, which requires agents with partial information to communicate under pressure. We find that language evolution does occur between LLM agents, that the resulting languages are compositional and morphologically productive, and that they deviate from the LLMs' English prior in ways that render them incomprehensible to humans. Moreover, we identify several qualities essential to this evolution: pressure towards efficiency; the strength of the models backing the agents; and access to a "postmortem" stage in which agents can agree on linguistic conventions. Importantly, we observe that different conditions govern the transmission of language to new agents. Specifically, we find that agents learn new languages from usage alone, take an active role in this learning, and that while stronger models are required for novel language emergence, weaker models can learn an existing language once it has emerged. Taken together, our results indicate that current LLMs have the potential for cumulative cultural evolution -- previously attested only in humans -- with mixed populations of agents developing capacities that go beyond their lowest common denominator.

---

### Defense-as-Skill: Evolving Runtime Guard Skill for Skill-Augmented Agents

Authors: Xiaofang Yang, Ziqi Miao, Dianbo Sui, Jing Shao, Lijun Li
Published: 2026-09-01
Categories: cs.CR, cs.AI
arXiv: https://arxiv.org/abs/2609.01487v1

Skill-augmented agents load reusable skills as persistent runtime context, improving task performance but also giving malicious skills a durable channel for steering future actions. Such skills may leak secrets, corrupt code, bypass approvals, or stage data for exfiltration only after a concrete user task and workspace state make the unsafe action appear useful. This makes pre-install vetting insufficient and calls for runtime, task-conditioned protection. We propose Defense-as-Skill, a defense paradigm that implements the runtime guard itself as an installable, inspectable, and editable skill. Our guard, SkillSonar, runs alongside untrusted task skills and checks sensitive actions against the user's task boundary, routing each action to an allow, replan, or confirmation decision without modifying the underlying agent runtime. To study this setting, we construct SCOPE-R, a task-conditioned dataset covering 6 risk families and 21 sub-categories, with 206 attack-confirmed malicious instances and 43 benign tasks. We then improve SkillSonar on the SCOPE-R training subset using runtime guard-skill evolution, a Monte-Carlo Tree Search procedure that evolves the on-disk guard skill from feedback on the rollouts. Across Claude Code and OpenClaw, the evolved guard substantially reduces attack success while maintaining a favorable safety-utility trade-off. On repeated GLM-5 runs, SkillSonar reduces ID ASR from 0.482 to 0.104 and OOD ASR from 0.606 to 0.115. Further analyses demonstrate transfer across victim models, held-out risk families, and external benchmarks, as well as retained protection against adaptive attackers. Ablations further show that explicit safety responsibility assignment and the skill-native representation are both important to the observed gains.

---

### Harness-of-Harness: Multi-Day Autonomous Software Development with Continual Improvement

Authors: Haoyang Yan, Min-le Su, Hangfan Zhang, Zhanhao Li, Chen Zhang, Shao Zhang, Yang Chen, Lei Bai, Shuyue Hu
Published: 2026-09-01
Categories: cs.AI
arXiv: https://arxiv.org/abs/2609.01481v1

This paper studies autonomous software development, in which LLM-based coding agents transform high-level requirements into complete, functional, and usable software systems without human intervention. We introduce Harness-of-Harness (HoH), a framework that enables coding agents to continually improve software during autonomous development. HoH operates on existing coding-agent harnesses, and organizes their executions into iterative planning-coding-testing loops. To sustain improvement across loops, HoH balances repair with capability growth, scopes development into small and verifiable increments, separates implementation-time testing from independent evaluation, and constrains verifiable outputs rather than prescribing agent workflows. It progressively exposes deliverables, role-specific tools, and skills, encourages reuse rather than recreation, and maintains versioned project histories. On GameCraft-Bench, FrontierSWE, and ProgramBench, three harness-model pairs (Codex with GPT-5.5, OpenCode with DeepSeek-V4-Pro, and Pi with MiniMax-M3), HoH consistently outperforms the corresponding standalone harnesses, achieving an average relative gain of 52.25 percent and a maximum gain of 82.86 percent after three iterations. In a multi-day deployment with more than 70 iterations, HoH autonomously develops a first-person-shooter game, featuring a coherent storyline, fully implemented core mechanics, human-playable experience, polished visuals and integrated audio. Github: https://github.com/Flesymeb/HarnessOfHarness Project Page: https://flesymeb.github.io/HarnessOfHarness/

---

### CameraEditor: Camera-Controlled Image Editing via Video-Prior Sequential Modeling

Authors: Xin Shen, Chengyou Jia, Keshuo Xing, Zifeng Zhu, Changliang Xia, Bowen Ping, Zhuohang Dang, Hangwei Qian, Minnan Luo
Published: 2026-09-01
Categories: cs.CV
arXiv: https://arxiv.org/abs/2609.01479v1

Beyond semantic content, camera parameters play a pivotal role in dictating the geometric perspective and appearance of any given image. While recent image editing models excel at semantic and stylistic manipulation, they struggle with explicit camera parameter control. When handling large perspective shifts, instruction-driven models face a dilemma: they either suffer from structural tearing or generate conservative outputs that ignore geometric instructions. To address this, we introduce CameraEditor, a framework that reformulates camera-controlled editing from a spatial problem into a temporal sequence prediction task. By leveraging the temporal coherence of video diffusion models, our approach integrates an explicit geometric perception module with a dynamic reference routing mechanism. This allows us to construct geometrically rigorous visual reference pairs via dynamic panorama cropping, overcoming the ambiguity of text-based instructions. Furthermore, CameraEditor strategically inserts intermediate transition frames to decompose large perspective shifts, providing a robust temporal buffer that preserves content identity and spatial coherence. We construct a training dataset of 5,760 instances. As an independent contribution, we introduce CamEditor-Bench, a model-agnostic evaluation suite of 462 test cases. Extensive experiments demonstrate that CameraEditor achieves state-of-the-art camera control precision and source identity preservation, outperforming existing methods.

---

### RadMatch: Auditable Radiology Report Evaluation via Finding-Level Matching

Authors: Charles Corbière, Léo Machado, Aubin Charley, Baptiste Callard, Pierre Manceron, Corentin Dancette
Published: 2026-09-01
Categories: cs.CV
arXiv: https://arxiv.org/abs/2609.01470v1

As AI systems are increasingly used to draft radiology reports, reliably evaluating their clinical quality remains a critical challenge. Large language model (LLM)-based metrics are now the best-correlated with radiologist judgment, yet they output a single opaque score that neither a clinician nor a model builder can easily interpret or audit. We introduce RadMatch, a multi-stage, LLM-based metric that decomposes report comparison into a structured finding-level matching with significance-aware scoring and error characterization across seven clinical attribute dimensions (status, location, severity, morphology, certainty, longitudinal comparison, and measurement). The main score is the actionable-error count, both interpretable and auditable. Candidate findings are graded correct, partial, or incorrect, and unmatched findings are counted as missed or hallucinated. Triage and actionable safety recall/precision and per-subset views add complementary, deployment-oriented lenses. Across two expert benchmarks, RadMatch is the most clinically aligned metric, matching inter-radiologist agreement on ReXVal and more than doubling the best prior metric on the harder RadEvalExpert. Relying only on few-shot prompting, it is designed to extend to other modalities and anatomies. We will release RadMatch as open-source code with an interactive dashboard for inspecting results.

---

### Parsing the Stream: A Live Trace Model for Long-Horizon Agents and Their Observers

Authors: Egor Pakhomov, Erik Nijkamp
Published: 2026-09-01
Categories: cs.AI
arXiv: https://arxiv.org/abs/2609.01466v1

A long-horizon agent's trace outgrows both of its consumers: the human observer monitoring the run, and the agent itself, whose bounded context the trace must be folded back into. We present a live trace model, an append-only event ledger folded incrementally into typed run state and compiled into per-consumer views, and evaluate it for both consumers against deterministic ground truth. For the observer side, evaluated with an LLM reader as proxy, the compiled view answers monitoring questions using approximately 14x and 15x fewer input tokens (by reader) and at 5-7x lower cost than a budget-capped single-call reading of the raw trace, with higher accuracy (0.85-0.87 versus 0.48). Because the questions were co-designed with the view schema, we treat the token and cost reduction, conditional on schema coverage, as the transferable result. For the agent, on 120-link sequential-dependency tasks, mechanisms that maintain the task's running statistic in per-step state succeed where full-context prompting fails (30/30 versus 8/30 under a clean protocol, n=30, labeled descriptive owing to benchmark-system co-development); a prompt-level scratchpad matches the fold's accuracy at lower cost, and a two-arm decomposition attributes the fold's accuracy to its deterministic aggregate and its cost advantage to its compactness. The fold's remaining value over cheaper alternatives is deterministic auditability and serving the observer from the same state. We derive eleven candidate requirements for trace folding from observed failures and delimit them with an order-sensitive task family on which the fold ceases to help. Code, benchmarks, a regenerable synthetic corpus, and all workbench traces are released.

---

### AutoConcept: Training-Free Concept-Guided Reranking for Metadata-Available Composed Image Retrieval

Authors: Tianyu Wang, Tianjiao Wu
Published: 2026-09-01
Categories: cs.IR, cs.CL
arXiv: https://arxiv.org/abs/2609.01456v1

Composed image retrieval (CIR) retrieves a target image from a reference image and a text modification. This paper studies metadata-available CIR reranking, where a fixed CIR model first returns a candidate pool and gallery metadata is then used for second-stage concept-guided scoring. We introduce AutoConcept, a training-free reranker that converts concept evidence into an interpretable memory. AutoConcept filters noisy concepts, activates query-relevant positive constraints with an auxiliary negative penalty, and combines base retrieval scores with metadata-based concept-candidate alignment through inference-time calibration. On FashionIQ, AutoConcept yields significant early-rank improvements over WeiMoCIR and consistent plug-in gains on LinCIR candidate pools. Metadata-aware controls show that structured concept memory adds signal beyond direct query-text and extracted-attribute matching, while a query-only variant further supports the effectiveness of concept-level reranking. A supplementary real-human concept-label study indicates that the same memory interface can consume participant-provided evidence. These results position AutoConcept as an interpretable concept-memory reranker for product-style CIR galleries with available metadata.

---

### When Safety Routing Breaks: Understanding Alignment Fragility under Benign Fine-Tuning

Authors: Yitong Guo, Xiaoyi Chen, Siyuan Zhang, Xiaofeng Wang, Haixu Tang
Published: 2026-09-01
Categories: cs.CR, cs.AI
arXiv: https://arxiv.org/abs/2609.01455v1

Benign fine-tuning severely weakens the safety alignment of large language models (LLMs), so we study why refusal behavior is so fragile. While prior work often attributes this failure to gradient conflict, we propose a fundamentally different Fisher-geometric explanation: safety Fisher is low-rank, and alignment makes the safety geometry flatter while preserving an output-routing pathway. After 100 benign fine-tuning examples, this pathway is selectively re-sharpened in output-side MLP modules, explaining the asymmetric fragility: safety can collapse to high attack success rates, while general utility degrades mildly. The routing view also explains why few safety examples can restore refusal behavior, indicating that internal safety-relevant representations are preserved. Finally, we show that LoRA and ASAM mitigate early collapse by suppressing output-side sharpness, but their protection weakens at larger fine-tuning scales. Overall, safety failure is best understood as a disruption of a low-rank output-routing mechanism

---

### Does Imitation Learning Preserve Temporal Robustness in Dexterous Manipulation? An Expert-Learner Comparison Across Task Execution Speeds

Authors: Clinton Enwerem, John S. Baras, Calin Belta
Published: 2026-09-01
Categories: cs.RO, cs.LG
arXiv: https://arxiv.org/abs/2609.01453v1

Dexterous manipulation policies learned by imitation are typically evaluated for robustness to variation in scenes, objects, or instructions, but their performance across task execution speeds is less often examined. This leaves open how much temporal robustness a learner retains relative to the expert it imitates. We compare an expert and learner under the same task conditions, initial-condition draws, and speedup factors. We instantiate the evaluation in ParcelStow, a contact-rich task in which the robot acquires, reorients, and inserts a parcel. The demonstrations span the speedup range for the manipulation phases after parcel acquisition. A scripted expert and an Action Chunking with Transformers (ACT) policy trained from the expert's demonstrations both achieve 100 percent task success at nominal speed. Their success rates diverge within the demonstrated range: at its maximum, expert success is 84 percent and ACT success is 53 percent. Two ACT policies with different parameter initializations show similar degradation, decreasing by 34 and 48 percentage points from nominal speed to the maximum demonstrated speed, compared with 16 points for the expert. Stage-level analysis shows that 35 of ACT's 47 failures at the maximum demonstrated speed are insertion misalignments. Under the relative-motion handoff, every ACT acquisition retains the parcel through reorientation and transfer in free space, but only 64 percent complete the overall task, compared with 95 percent after expert acquisition. Across all evaluated policies and speeds, none of the 414 acquisitions without force closure completes the task. Equal nominal task success therefore does not imply preservation of expert performance across execution speeds. Code, data, and evaluation scripts are available at https://github.com/coenwerem/parcelstow.

---

### Diffusion as a Training Curriculum for Timestep-Free Iterative Reasoning

Authors: Mariia Drozdova, Aidan Sirbu, Pietro Miotti, Robert Obryk, Mayalen Etcheverry, Eyvind Niklasson, Blake Richards
Published: 2026-09-01
Categories: cs.LG
arXiv: https://arxiv.org/abs/2609.01449v1

Diffusion models and recursive reasoners are both iterative, but they carry information across iterations differently. We add a persistent hidden state to a diffusion denoiser and remove its timestep conditioning, leaving a single shared update that can be run to arbitrary depth. The result is an anytime solver: accuracy keeps improving with inference depth far beyond the rollout lengths and backpropagation window used in training, reaching 99.90% exact solve on Sudoku-Extreme. We also obtain 98.93% solve rate on Maze-Unique. Surprisingly, progressive denoising is unnecessary at inference: holding corruption at its maximum by replacing every non-clue variable with fresh Gaussian noise at each step retains near-perfect solving and converges to stable solutions. This simple noise-injection mechanism enables a single trajectory to efficiently explore the solution space and settle on the correct answer without parallel rollouts, candidate selection, or external verifiers required by prior reasoning models. Nonetheless, ordered annealed corruption remains critical during training, which suggests that diffusion's primary contribution to our anytime solver is not a sampling procedure at inference, but a denoising training curriculum.

---

### Edge-Girth as a Structural Edge Feature for Graph Neural Networks

Authors: Lilian Marey, Charlotte Laclau
Published: 2026-09-01
Categories: cs.LG
arXiv: https://arxiv.org/abs/2609.01441v1

Graph neural networks (GNN) based on message passing are provably no more powerful than the one-dimensional Weisfeiler--Leman colour-refinement test (1-WL): two graphs it cannot tell apart receive identical representations, however deep or wide the network. A common remedy augments node or edge features with precomputed structural descriptors, most often counts of a fixed small subgraph such as triangles or longer cycles, but such counts require committing in advance to the size of the substructure counted, a choice usually made blind to the data. We study a descriptor that avoids this choice. The edge-girth of an edge is the length of a shortest cycle through it, and its multiplicity is the number of such shortest cycles; together they form a per-edge invariant that reports cycles of arbitrary length, computable exactly by a single breadth-first search per edge. Injected into a gated message-passing architecture, EGAGNN, it reaches a test MAE a factor three below the closest gated comparator on the ZINC-12k regression benchmark at 104k parameters; against bounded cycle-counting descriptors under the same architecture, it matches only a dictionary counting cycles up to length eight, using twice as many channels, while a dictionary capped at length four performs no better than no structural information at all. On graph discrimination we prove a matching limitation: on graphs where every edge sees the same number of shortest cycles of the same length, the descriptor becomes constant and any model built on it collapses back to the 1-WL bound. This holds without exception across all 400 pairs of the BREC benchmark: not one of the 90 such pairs is distinguished.

---

### HarnessDev: Can LLMs Create and Evolve Their Own Agent Harness?

Authors: Yuhao Wu, Jingyuan Zhang, Jiajun Shi, Xinping Lei, Qingshui Gu, Yuxuan Zhang, Zexuan Wang, Chen He, Chen Huang, Maojia Song, Zhiyuan Zeng, Shaowen Wang et al.
Published: 2026-09-01
Categories: cs.SE, cs.CL
arXiv: https://arxiv.org/abs/2609.01437v1

As agents move from research prototypes to deployed tools, their capability increasingly depends on model-external execution infrastructure, commonly termed the agent harness. Changing this harness while holding model weights fixed can substantially alter task performance. Current agent evaluations typically report downstream performance under a chosen harness, leaving a model's ability to develop the harness itself comparatively underexplored. We introduce HarnessDev, a benchmark that shifts the unit of evaluation from task outputs to runnable infrastructure. HarnessDev covers two stages. In Creation, the agent starts from a minimal seed and a small number of cases, then builds a complete execution system. In Evolution, it starts from its own created harness and iteratively revises it using downstream execution feedback, with the goal of improving benchmark performance. We then evaluate each constructed harness on capability (task success on held-out benchmarks) and efficiency (execution-token cost). The reported Creation results cover six creator LLMs, four domains, and five downstream benchmarks totaling 2,207 unique downstream instances, with hidden evaluation tasks withheld from development. We find that generated harnesses remain substantially behind mature human-engineered references on code and on search and research, while matching or exceeding the selected references on writing and machine-learning experimentation, with large variation in execution cost. Evolution produces some performance gains, but they are unstable and transfer only partially to held-out tasks. Experiments with a fixed runtime model further show that the gains depend strongly on the model executing the harness, indicating limited transfer across models.

---

### Gaussian Core LoRA: Distribution-Aware Dynamic Adaptation for Broad Concept Erasure

Authors: Qinghui Gong, Xunlei Chen, Yu-Xuan Zhang, Hua Meng, Zhengchun Zhou
Published: 2026-09-01
Categories: cs.CV
arXiv: https://arxiv.org/abs/2609.01433v1

Concept erasure aims to suppress unsafe, privacy-sensitive, or undesirable generations in text-to-image diffusion models while preserving benign semantics, visual quality, and deployment efficiency. Existing adapter-based methods, such as Low-Rank Adaptation (LoRA), typically freeze the diffusion backbone and learn lightweight parameter updates to steer generation away from target semantics. However, these methods usually assign a static semantic erasure direction to each target concept. This assumption is overly coarse for broad and complex target concepts, since a concept often contains multiple latent semantic prototypes involving different objects, scenes, or relations, and requires different local erasure directions. A single LoRA update averages these heterogeneous erasure demands, leading to under-erasure on difficult prototypes and over-editing of nearby benign semantics. To address this limitation, we propose Gaussian Core LoRA, a distribution-aware low-rank adaptation framework. It fits a Gaussian mixture model in the prompt feature space to estimate latent semantic prototypes within the target concept. During inference, each input prompt is projected into this feature space to compute its Gaussian posterior responsibilities, which condition the core generator to produce a prompt-specific, norm-bounded residual reconfiguration of the shared LoRA rank space. This enables prototype-adaptive erasure with a single lightweight adapter. Compared with the strongest baseline on each metric, Gaussian Core LoRA reduces average Attack Success Rate (ASR) by 7.95%, lowers COCO Fr'echet Inception Distance (FID) by 14.72%, and improves CLIP Score by 4.98%. Further experiments show robustness to adversarial prompts, scalability to multi-identity and multi-style erasure, and compatibility with SDXL and FLUX.

---

### Citing Less Critically: LLMs Reshape the Rhetoric and Reach of Scientific Citation

Authors: Yixuan Liu, Lin Chen, Zhuoqi Liu, Jianglin Lu, Dakota Murray
Published: 2026-09-01
Categories: cs.DL, cs.CL, cs.CY, cs.SI
arXiv: https://arxiv.org/abs/2609.01432v1

Scientific citations carry rhetorical intent. Scholars may cite prior work positively (supporting), negatively (contrasting), or neutrally (mentioning). As large language models (LLMs) increasingly assist scientific writing, whether they reproduce citations with the same rhetorical intent as humans remains unclear. We introduce a masked-citation task to compare human and LLM-generated citation behavior. For each citation context, an LLM generates a replacement citation sentence, producing a counterfactual corpus directly comparable to human citation. We analyze what, whom, and how models cite, using an LLM-as-a-judge to classify citation intent and a 20-million-edge coauthorship network to measure social distance between cited authors. Across six popular LLMs and 1,746 top NLP conference papers (63k+ contexts, 132k+ citations), three patterns emerge: (1) Compared with human citation, LLMs cite significantly less critically; (2) LLMs over-cite popular and older papers, a tendency amplified for contrasting citations where human writing more often draws on recent, niche work; (3) Whereas humans often cite within their close social network, especially for supporting citations, LLMs tend to draw on more socially distant authors. Together, these differences are double-edged: LLM citation reaches beyond a scholar's close collaborators while being less critical and amplifying visibility bias, reshaping the rhetoric and reach of scientific citation.

---

### Efficiently Estimating Optimal Hyperparameter Scaling Laws through Power-Law Entropy Search

Authors: Zhiliang Chen, Sebastian Ament, David Eriksson, Maximilian Balandat, Bryan Kian Hsiang Low, Eytan Bakshy, Jihao Andreas Lin
Published: 2026-09-01
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2609.01431v2

Optimal hyperparameter scaling laws describe how the best hyperparameters for large language model (LLM) training change with model and data scale, enabling practitioners to predict optimal configurations at production scales without expensive large-scale tuning. However, estimating these scaling laws conventionally requires exhaustive grid searches over thousands of training runs, consuming enormous computational resources. We introduce Power-Law Entropy Search (PLES), a computational cost-aware acquisition function built on multi-fidelity Bayesian optimization that efficiently estimates optimal hyperparameter scaling laws through adaptive experimentation. A key innovation in PLES is that it searches for candidates that reduce the overall uncertainty of a scaling law estimate, instead of optimizing a single objective function. At each iteration, PLES selects the candidate configuration that maximally reduces the uncertainty of the scaling law estimates per unit computational cost, naturally favoring informative small-scale experiments. We evaluate PLES on synthetic benchmarks, surrogate models fitted to real LLM training data, and actual LLM pre-training runs. Across all settings, PLES converges to accurate optimal hyperparameter scaling laws using less than one-tenth of the computational budget required by conventional grid search and other baselines.

---

### Efficiently Estimating Optimal Hyperparameter Scaling Laws through Power-Law Entropy Search

Authors: Zhiliang Chen, Sebastian Ament, David Eriksson, Maximilian Balandat, Eytan Bakshy, Jihao Andreas Lin
Published: 2026-09-01
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2609.01431v1

Optimal hyperparameter scaling laws describe how the best hyperparameters for large language model (LLM) training change with model and data scale, enabling practitioners to predict optimal configurations at production scales without expensive large-scale tuning. However, estimating these scaling laws conventionally requires exhaustive grid searches over thousands of training runs, consuming enormous computational resources. We introduce Power-Law Entropy Search (PLES), a computational cost-aware acquisition function built on multi-fidelity Bayesian optimization that efficiently estimates optimal hyperparameter scaling laws through adaptive experimentation. A key innovation in PLES is that it searches for candidates that reduce the overall uncertainty of a scaling law estimate, instead of optimizing a single objective function. At each iteration, PLES selects the candidate configuration that maximally reduces the uncertainty of the scaling law estimates per unit computational cost, naturally favoring informative small-scale experiments. We evaluate PLES on synthetic benchmarks, surrogate models fitted to real LLM training data, and actual LLM pre-training runs. Across all settings, PLES converges to accurate optimal hyperparameter scaling laws using less than one-tenth of the computational budget required by conventional grid search and other baselines.

---

### Learning Sparse Decision Trees via Transformer Variational Auto-Encoders

Authors: Giacomo Fidone, Alessio Cascione, Riccardo Guidotti
Published: 2026-09-01
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2609.01430v1

Decision trees are among the most widely used models in machine learning, largely due to their transparent decision logic, making them well-suited for high-stakes decision-making contexts. However, most existing learning algorithms focus on predictive performance, overlooking the joint optimization of other desirable properties, such as structural sparsity. In this work we propose TREVIS, an approach for learning decision trees with respect to complex objectives, based on the exploration of the latent space of a Tree Transformer Variational Auto-Encoder (TTVAE). By mapping decision trees onto latent representations, TREVIS replaces the discrete search space with a continuous one, enabling gradient-based optimization via a differentiable surrogate model. We experiment with TREVIS for learning decision trees that jointly optimize predictive performance and sparsity. Results show that TREVIS discovers decision trees matching the predictive performance of existing near-optimal algorithms while improving their structural sparsity.

---

### TRIAGE: Three-level Routing and Intelligent Agent Guidance for Efficient Execution

Authors: Ruocan Wei
Published: 2026-09-01
Categories: cs.LG
arXiv: https://arxiv.org/abs/2609.01428v1

Large Language Model (LLM) agents based on the ReAct paradigm have demonstrated remarkable capabilities in tool use and task execution. However, ReAct suffers from a fundamental efficiency problem: every query triggers a complete reasoning loop from scratch, and similar queries repeat identical steps without leveraging historical experience. We propose TRIAGE,a three-level routing framework that reduces token consumption by reusing historical execution trajectories. Its core innovation is TaaS (Trajectory-as-a-Skill), which abstracts historical execution trajectories into reusable skills, realizing 'experience as a service'. TRIAGE classifies queries into three levels: (1) Direct Reuse-identical queries, 0 tokens; (2) Skill Substitution-similar queries, 0 tokens via deterministic parameter substitution; (3) Full ReAct-novel queries, automatically stored for future reuse. In large-scale experiments on 1,007 security monitoring queries, TRIAGE achieves 62.3% token savings, with 56.0% of queries at Level 2 and 5.5% at Level 1, both executing at zero cost. Cross-domain validation on ToolBench (15 domains, 345 queries) achieves 76.3% token reduction, confirming the generalizability of semantic routing. An online learning experiment demonstrates cold-start-to-mature evolution: the L2 hit rate rises from 0% to 57% within the first 100 queries, and the average token cost drops from 198 to 74.7. We also propose an automatic Skill extraction mechanism that distills high-frequency trajectory patterns into deterministic Skills, creating a positive feedback loop of 'the more you use it, the more efficient it becomes'.

---

### Pix2Rep-v2: Data-Efficient Representation Learning for Dense Medical Imaging Applications

Authors: S. Sifaoui, E. Angelini, S. Toupin, T. Pezel, L. Le Folgoc
Published: 2026-09-01
Categories: cs.CV
arXiv: https://arxiv.org/abs/2609.01427v1

Dense self-supervised learning (SSL) is a powerful paradigm for learning without annotations the local descriptors required to solve dense medical imaging tasks. We present Pix2Rep-v2, a framework for SSL of pixel- and voxel-level representations suitable for few-shot downstream applications. Pix2Rep-v2 addresses the main challenges of dense SSL by leveraging a redundancy reduction objective at the pixel-level with a principle of equivariance of dense representations, that scales efficiently to 3D or wide field-of-view applications. We evaluate our method on four datasets, across multiple tasks, multiple modalities and anatomical structures using multiple backbones in 2D and 3D, and under various data regimes. As an alternative to linear probing or full fine-tuning on the downstream task, we also propose an in-context variant, without downstream training, based on a dense prototype approach. Pix2Rep-v2 shows substantially higher data-efficiency in few-shot scenarios compared to fully supervised baselines, and is competitive with the state-of-the-art e.g., +9.3 Dice points in one-shot segmentation on the M&Ms-2 dataset. Our code and pre-trained models are publicly available at https://github.com/BioMedTP/pix2rep-v2.

---

### Semantic-Guided Multimodal Preprocessing for Vision Transformer-Based Clear Cell Renal Cell Carcinoma Grading

Authors: Fatemeh Javadian, Zhu Chen, Zahra Aminparast, Johannes Stegmaier
Published: 2026-09-01
Categories: cs.CV, cs.AI, cs.LG, eess.IV
arXiv: https://arxiv.org/abs/2609.01426v1

Clear cell renal cell carcinoma (CCRCC) grading is essential for treatment planning, yet existing approaches either analyze patch-level images directly or focus solely on nuclei-level classification, without linking to final tumor grading. We propose a semantic-guided multimodal preprocessing method that integrates nuclei classification maps from existing pre-trained models with RGB histopathology images for Vision Transformer (ViT)-based CCRCC grading. Our approach employs classification map channel concatenation and multiplicative modulation, with optimized overlays to leverage nuclei grading information, while preserving RGB textural features. Evaluation of multiple preprocessing strategies demonstrates that semantic-guided enhancement achieves 0.916 balanced accuracy, outperforming RGB-only baseline (0.707) and max-voting aggregation from prior studies (0.427). Sensitivity analysis reveals that this 21 percentage point improvement over baseline persists even under simulated perturbation at rates matching current state-of-the-art nuclei classification model error thresholds, suggesting both effective semantic utilization and practical robustness. These findings show that preprocessing-based multimodal fusion can leverage the diagnostic potential of existing imperfect nuclei classifiers, effectively bridging previously isolated fine-grained nuclear-level analysis with coarse-grained ViT-based patch classification. Per-class recall was consistent across grades (0.93, 0.91, 0.91), indicating that gains are not concentrated in the majority class. Because the sensitivity analysis perturbs ground-truth maps rather than predictions from an actual nuclei model, this result characterizes robustness under simulated error rather than deployment with a real upstream model, which remains for future work.

---

### CATeye: Coupled Attribute-Topology Invariance Learning for Voucher Abuse Detection

Authors: Tian Tian, Shuaicheng Niu, Hao Kuang, Yuanhang Hu, Dong Li, Zhiqi Shen
Published: 2026-09-01
Categories: cs.LG
arXiv: https://arxiv.org/abs/2609.01425v1

Voucher abuse poses a major challenge in e-commerce, where malicious users exploit promotional vouchers for profit. Unfortunately, fraud patterns evolve rapidly over time and across regions, causing distribution shifts that degrade existing detection models unless retrained frequently. To tackle this, we propose the Coupled Attribute-Topology Invariance Learning framework (CATeye). The key challenge arises from coupled attribute-topology shift, where edges built from attribute proximity cause environment-driven attribute shift to induce shifted topology, thereby amplifying variant signals through GNN message passing. CATeye sees through such coupled shifts with two learnable selectors. First, an Attribute Invariance Selector (AIS) learns node-adaptive masks to filter out non-invariant attributes. Then, conditioned on retained invariant attributes, an Edge Invariance Selector (EIS) samples an invariant subgraph and isolates non-invariant edges. Using the resulting invariant and non-invariant components, CATeye constructs multiple views and applies view-specific objectives to emphasize domain-invariant representations while suppressing domain-specific variations. Experiments on both a proprietary dataset from Lazada, a major Southeast Asian e-commerce platform, and a public benchmark show that CATeye consistently outperforms nine strong domain generalization and graph anomaly detection baselines, achieving up to an 8.61% improvement in average F1 score over the strongest baseline. Source code is publicly available at https://github.com/Tian0426/CATeye.

---

### MegaStyle++: Scaling Image Style Space through Hierarchical Style Definition

Authors: Junyao Gao, Sibo Liu, Jiaxing Li, Yanan Sun, Weidong Zhang, Jun Zhang, Cairong Zhao
Published: 2026-09-01
Categories: cs.CV
arXiv: https://arxiv.org/abs/2609.01423v2

Image style is a highly abstract, human-constructed concept shaped by a range of visual factors and intrinsically entangled with content, yet a unified and explicit definition of image style remains lacking. In this work, we first discuss the fundamental question of what is style and then propose a hierarchical style definition that describes image style from an overall style identity to fine-grained visual attributes, providing a more structured, transferable, and interpretable style representation. Based on this definition, we refine the style annotation pipeline of MegaStyle and construct MegaStyle++-8M, a large-scale style dataset containing 150K overall style identities, 1M fine-grained style prompts, and 8M stylized images. Extensive analyses demonstrate that our hierarchical definition substantially expands the style space in both diversity and semantic breadth, while precisely capturing intrinsic visual style of reference images. The dataset and code will be updated at https://github.com/Tencent/MegaStyle, we hope MegaStyle++ provides a scalable foundation for studying and modeling diverse image styles.

---

### MegaStyle++: Scaling Image Style Space through Hierarchical Style Definition

Authors: Junyao Gao, Sibo Liu, Jiaxing Li, Yanan Sun, Weidong Zhang, Cairong Zhao, Jun Zhang
Published: 2026-09-01
Categories: cs.CV
arXiv: https://arxiv.org/abs/2609.01423v1

Image style is a highly abstract, human-constructed concept shaped by a range of visual factors and intrinsically entangled with content, yet a unified and explicit definition of image style remains lacking. In this work, we first discuss the fundamental question of what is style and then propose a hierarchical style definition that describes image style from an overall style identity to fine-grained visual attributes, providing a more structured, transferable, and interpretable style representation. Based on this definition, we refine the style annotation pipeline of MegaStyle and construct MegaStyle++-8M, a large-scale style dataset containing 150K overall style identities, 1M fine-grained style prompts, and 8M stylized images. Extensive analyses demonstrate that our hierarchical definition substantially expands the style space in both diversity and semantic breadth, while precisely capturing intrinsic visual style of reference images. The dataset and code will be updated at https://github.com/Tencent/MegaStyle, we hope MegaStyle++ provides a scalable foundation for studying and modeling diverse image styles.

---

### From Rollouts to Recipes: Self-Contained Post-Training for LLMs

Authors: Yifei Li, Lingling Zhang, Muye Huang, Zihan Ma, Jiashuai Liu, Jun Liu
Published: 2026-09-01
Categories: cs.CL
arXiv: https://arxiv.org/abs/2609.01422v1

Post-training large language models usually applies a single training recipe to all samples, even though the model's own rollouts reveal different sample-level learning states. We propose Self-Routing, a behavior-conditioned post-training framework that uses rollout correctness and confidence to decide how each sample should be optimized. Depending on its behavior state, a sample is routed to GRPO, on-policy self-distillation, regularization, or skipping, allowing training to adapt without external teachers, extra annotations, or additional sampling. Experiments on mathematical reasoning across Qwen3 and Qwen3.5 backbones show that Self-Routing consistently improves over uniform GRPO, uniform OPSD, fixed mixtures, and simpler routing baselines. Further analyses show that the routing distribution changes over training and reduces unnecessary updates on low-signal or already stable samples.

---

### Provably Safe Sim-to-Real Transfer

Authors: Tingting Ni, Maryam Kamgarpour
Published: 2026-09-01
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2609.01418v2

We address safe sim-to-real transfer, in which an agent leverages an imperfect simulator and limited real-world interaction while ensuring safety throughout data collection in the real system. This problem arises in applications such as robotics and healthcare: simulators provide cheap data, but sim-to-real mismatch makes direct transfer unreliable, and collecting real-world data to correct this mismatch must itself be safe. Moreover, deployment objectives may vary across tasks, making it costly to collect new data for each reward function. We therefore formulate safe sim-to-real transfer as a reward-free safe reinforcement learning (RL) problem, in which data are collected once and reused to plan for arbitrary reward functions. We develop a computationally efficient algorithm that identifies where the simulator and real dynamics differ, uses certified simulator transitions where they are reliable, and estimates mismatched transitions from safely collected data. With high probability, every policy deployed during learning is feasible, and the collected data support the computation of a feasible and near-optimal policy for any reward function. When the simulator is uninformative, our algorithm recovers online reward-free safe RL while improving the best-known sample complexity by a factor of \(\widetildeΘ(H/ξ^2)\), where \(ξ\) is the safety margin of a baseline policy. When the simulator is accurate on most transitions, this improvement grows to \(\widetildeΘ(H^2|\mc S||\mc A|/(ξ^2|\mc B|))\), where \(|\mc B|\) denotes the size of the sim-to-real mismatch region.

---

### Provably Safe Sim-to-Real Transfer

Authors: Tingting Ni, Maryam Kamgarpour
Published: 2026-09-01
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2609.01418v1

To mitigate the sample complexity of real-world reinforcement learning (RL), a common practice is to first train a policy in a simulator, where samples are cheap, and then deploy the learned policy in the real world with the hope that it generalizes effectively. Such direct sim-to-real transfer is not guaranteed to succeed: simulator-trained policies can be suboptimal in the real world due to sim-to-real mismatch. Correcting this mismatch requires collecting data from the real system, but in many applications, such as robotics and healthcare, this data-collection process is itself subject to safety constraints. This gives rise to the problem of safe sim-to-real transfer: how can an agent exploit an imperfect simulator while ensuring safe real-world data collection and learning a near-optimal feasible policy for the target system? We address this problem by formulating safe sim-to-real transfer within the framework of reward-free safe RL. We design a computationally efficient algorithm that exploits simulator information to provably reduce real-world interaction while ensuring safe exploration and enabling the computation of a near-optimal feasible policy for any potential reward function. Our real-world sample complexity bound characterizes the benefit of using the simulator in terms of the sim-to-real mismatch.

---

### Predicting Subsurface Abnormalities Growth using Physics-Informed Neural Networks

Authors: Mehrdad Shafiei Dizaji, Hoda Azari
Published: 2026-09-01
Categories: cs.LG
arXiv: https://arxiv.org/abs/2609.01417v1

The research explores the pioneering integration of Physics-Informed Neural Networks (PINNs) into the domain of Ground-Penetrating Radar (GPR) data prediction. This research presents a detailed development framework for a specialized PINN model, proficient at interpreting and forecasting GPR data, much like how medical imaging models predict tumor behavior. By harnessing the synergy between deep learning algorithms and the physical laws governing subsurface structures or in medical terms, human tissues the model effectively embeds the physics of electromagnetic wave propagation into its architecture. This ensures that predictions not only align with fundamental physical principles but also mirror the precision needed in medical diagnostics for detecting and monitoring tumors. The suggested deep learning structure comprises three components: a CNN, a spatial feature channel attention (SFCA) mechanism, and ConvLSTM, along with temporal feature frame attention (TFFA) modules. The attention mechanism computes channel attention and temporal attention weights using self-adaptation, thereby fine tuning the visual and temporal feature responses to extract the most pertinent and significant visual and temporal features. By integrating physics directly into the neural network, our model has shown enhanced accuracy in forecasting GPR data. This improvement is vital for conducting effective assessments of bridge deck conditions and other evaluations related to civil infrastructure. The use of Physics Informed Neural Networks (PINNs) has demonstrated the potential to transform the field of Non-Destructive Evaluation (NDE) by enhancing the precision of infrastructure deterioration predictions. Moreover, it offers a deeper insight into the fundamental mechanisms of deterioration, viewed through the prism of physics-based models.

---

### On the Reliability of Generative Augmentation: A Wasserstein-Based Theoretical and Empirical Study

Authors: Chathurika S Abeykoon, Mathias Nthiani Muia, Mallory Goldstein
Published: 2026-09-01
Categories: stat.ML, cs.LG
arXiv: https://arxiv.org/abs/2609.01410v1

Generative data augmentation is widely used to mitigate class imbalance, yet its theoretical effect on downstream generalization remains poorly understood. In this work, we develop a statistical framework for conditional generative augmentation and analyze its impact on classification risk. We formalize augmentation as a distribution-mixing process and show that the resulting risk distortion is controlled by both the augmentation strength and the class-conditional Wasserstein discrepancy between real and generated distributions. We further derive a capacity-dependent generalization bound based on Rademacher complexity, revealing an explicit trade-off between hypothesis complexity, augmentation intensity, and generative fidelity. Empirically, we evaluate the framework on binary and multiclass imbalanced classification tasks using Conditional GAN and Conditional WGAN-GP augmentation. Across datasets, CWGAN-GP consistently achieves lower Wasserstein discrepancies than CGAN, indicating improved distributional fidelity. However, improved fidelity does not necessarily translate into superior classification performance, with classical oversampling methods often remaining competitive. These findings support the central theoretical prediction that augmentation reliability is governed by distributional approximation error rather than predictive performance alone. Overall, this work establishes generative augmentation as a distributional perturbation process whose reliability can be quantified through Wasserstein-based measures and supported by finite-sample generalization guarantees. The proposed framework provides a principled foundation for evaluating synthetic data quality beyond classification accuracy alone.

---

### EdiTikZ: Scientific Figure Editing from Revision Trajectories

Authors: Christian Greisinger, Zhixue Zhao, Steffen Eger
Published: 2026-09-01
Categories: cs.AI, cs.CL, cs.CV
arXiv: https://arxiv.org/abs/2609.01409v2

Vision-language models (VLMs) have shown strong performance in generating scientific figures from text or images. However, publication-ready figures often require iterative refinement, making scientific figure editing an important yet largely unexplored step toward interactive figure creation. Existing approaches rely on costly proprietary agentic systems, focus primarily on evaluation, or construct training supervision from synthetically generated edits. Instead, we leverage naturally occurring scientific revision and development trajectories as a scalable source of supervision. To this end, we introduce DaEdiTikZ, the first large-scale dataset of revision-derived scientific figure edits, constructed by mining 391K plausible TikZ edit pairs from arXiv, GitHub, and TeX SE and inferring 781K directed edit instructions with a VLM conditioned on rendered figures and TikZ code. We further introduce DaEdiTikZ-Bench, a human-refined benchmark with 690 instances, and train two compact Qwen3.5-based EdiTikZ models (4B and 9B) by jointly learning image-to-TikZ reconstruction and instruction-conditioned editing, followed by reinforcement learning (RL) with complementary rewards for rendered fidelity and edit application. Automatic evaluation places our 9B model above all tested baselines, while human evaluation with 9 annotators and 4,320 ratings places it above GPT-5.6-Sol and on par with Gemini-3.1-Pro. Under severe out-of-distribution shifts, it remains competitive with GPT-5.6-Sol near its 2K training sequence-length regime.

---

### EdiTikZ: Scientific Figure Editing from Revision Trajectories

Authors: Christian Greisinger, Zhixue Zhao, Steffen Eger
Published: 2026-09-01
Categories: cs.AI, cs.CL, cs.CV
arXiv: https://arxiv.org/abs/2609.01409v1

Vision-language models (VLMs) have shown strong performance in generating scientific figures from text or images. However, producing publication-ready figures requires iterative refinement, making scientific figure editing an important yet largely unexplored task. Existing approaches rely on costly proprietary agentic systems, focus primarily on evaluation, or construct training supervision from synthetically generated edits. Instead, we leverage naturally occurring scientific revision and development trajectories as a scalable source of supervision. To this end, we introduce DaEdiTikZ, the first large-scale dataset of revision-derived scientific figure edits, constructed by mining 391K plausible TikZ edit pairs from arXiv, GitHub, and TeX SE and inferring 781K directed edit instructions with a VLM conditioned on rendered figures and TikZ code. We further introduce DaEdiTikZ-Bench, a human-refined benchmark with 790 instances, and train two compact Qwen3.5-based EdiTikZ models (4B and 9B) by jointly learning reconstruction and editing, followed by reinforcement learning (RL) with complementary rewards for rendered fidelity and edit application. Automatic evaluation places our 9B model above all tested baselines, while human evaluation with 9 annotators and 4,320 ratings places it above GPT-5.6-Sol and on par with Gemini-3.1-Pro. Under severe out-of-distribution shifts, it remains competitive with GPT-5.6-Sol near its 2K training sequence-length regime. Models and datasets will be released.

---

### Neuro-Symbolic Geometric Abstraction (NeuSOGA): From Observations to Symbolic Mathematical Representations

Authors: Qingde Li, Qingqi Hong, Zihan Li, Jie Tian
Published: 2026-09-01
Categories: cs.AI, cs.CV, cs.GR
arXiv: https://arxiv.org/abs/2609.01408v2

A fundamental challenge in artificial intelligence is the transformation of observations into explicit symbolic representations suitable for abstraction, interpretation, and reasoning. While modern AI systems achieve remarkable perceptual capabilities through large-scale statistical learning, the resulting knowledge is typically encoded within latent parameters that are difficult to inspect or manipulate analytically. Inspired by Neuro-Symbolic AI and theories of human abstraction, this paper investigates the formation of symbolic mathematical representations from geometric observations. We propose NeuSOGA (Neuro-Symbolic Geometric Abstraction), a framework that progressively transforms observations into topological abstractions, geometric abstractions, and ultimately symbolic mathematical representations. The architecture combines topology-guided structural discovery using Euclidean Distance Transforms, foundation-model perception using Segment Anything, adaptive multi-scale geometric abstraction, and symbolic synthesis through Implicit Area Splines. The resulting representation is an analytical implicit model supporting arbitrary-order smoothness, additive composition, and closed-form evaluation. Unlike neural latent encodings, the generated representation remains interpretable, editable, and mathematically explicit. Experiments on ModelNet40 point clouds, arbitrary-view projections, and segmented optical observations demonstrate that NeuSOGA transforms diverse observations into compact symbolic representations while preserving essential geometric and topological structure across sensing modalities and viewing directions. NeuSOGA provides an interpretable and explainable pathway from observation to symbol and establishes

---

### Neuro-Symbolic Geometric Abstraction (NeuSOGA): From Observations to Symbolic Mathematical Representations

Authors: Qingde Li, Qingqi Hong, Jie Tian
Published: 2026-09-01
Categories: cs.AI, cs.CV, cs.GR
arXiv: https://arxiv.org/abs/2609.01408v1

A fundamental challenge in artificial intelligence is the transformation of observations into explicit symbolic representations suitable for abstraction, interpretation, and reasoning. While modern AI systems achieve remarkable perceptual capabilities through large-scale statistical learning, the resulting knowledge is typically encoded within latent parameters that are difficult to inspect or manipulate analytically. Inspired by Neuro-Symbolic AI and theories of human abstraction, this paper investigates the formation of symbolic mathematical representations from geometric observations. We propose NeuSOGA (Neuro-Symbolic Geometric Abstraction), a framework that progressively transforms observations into topological abstractions, geometric abstractions, and ultimately symbolic mathematical representations. The architecture combines topology-guided structural discovery using Euclidean Distance Transforms, foundation-model perception using Segment Anything, adaptive multi-scale geometric abstraction, and symbolic synthesis through Implicit Area Splines. The resulting representation is an analytical implicit model supporting arbitrary-order smoothness, additive composition, and closed-form evaluation. Unlike neural latent encodings, the generated representation remains interpretable, editable, and mathematically explicit. Experiments on ModelNet40 point clouds, arbitrary-view projections, and segmented optical observations demonstrate that NeuSOGA transforms diverse observations into compact symbolic representations while preserving essential geometric and topological structure across sensing modalities and viewing directions. NeuSOGA provides an interpretable and explainable pathway from observation to symbol and establishes

---

### Contribution-Aware Bandwidth Allocation for Multimodal Split Learning

Authors: Iason Ofeidis, Leandros Tassiulas
Published: 2026-09-01
Categories: cs.LG, cs.DC, cs.NI
arXiv: https://arxiv.org/abs/2609.01406v1

Multimodal models are increasingly the default option for perception at the network edge, yet they are trained almost entirely in the datacenter, because a client holding several sensor streams cannot host an encoder per modality. Split Learning makes such training feasible by keeping only the first layers on the device, at the cost of an uplink that must carry smashed activations for every modality at every step. Existing compression schemes give each modality the same keep-ratio, so the shared budget is divided in proportion to smashed-activation dimension, a quantity unrelated to how much each modality contributes to the fused prediction. We make that division an explicit decision and call it inter-modality allocation: under a fixed uplink budget, every policy transmits the same expected payload and differs only in how that payload is split across modalities. Our allocator, ModalShare, sets each modality's keep-ratio from a Shapley contribution score that the server computes over coalitions of activations it has already received. Measuring this score adds no uplink traffic and no client-side computation, and needs no prior knowledge of which stream is which. ModalShare improves accuracy over equal keep-ratios by 15.4 and 12.4 percentage points on CREMA-D and MVSA at matched payload in 5x compression, with strong performance across three compressors, three datasets, and four budgets. We show that existing compressors underperform in multimodal settings, with ModalShare recovering what gains are left behind.

---

### Evaluating Multimodal LLMs as Generalist Vision-Language-Action Agents for Drone Control: Commanding, Approaching, Tracking and Searching

Authors: Jaewoo Park, Minyoung Lee, Sukmin Seo, Moonbin Yim, Hyunwook Yoon, Dohoon Ryu, Daehee Kim, Myungseo Song, Jihyuk Byun, Seunggyu Chang, Taeho Kil, Jiseob Kim et al.
Published: 2026-09-01
Categories: cs.RO, cs.AI
arXiv: https://arxiv.org/abs/2609.01404v1

Multimodal Large Language Models (MLLMs) are strong perceivers of images and video. We ask how far that reach extends into acting: dropping an MLLM directly into a drone's control loop, with its entire action space declared solely in the prompt. Recent systems approach this setting but increasingly narrow the model's decision-making. We widen it back. We introduce DroneCATS-Agent, an architecture where the MLLM is a swappable component, and DroneCATS, a benchmark treating the model as the independent variable. Beyond merely flying toward a pixel, our agent entrusts the model to yaw and search, deliberate when unsure, and self-declare arrival---all without fine-tuning or function-calling schemas. Evaluating frontier and open models across four core capabilities---approaching a visible target, tracking a moving one, searching outside the initial view, and commanding a multi-drone fleet---reveals that even the simplest embodied settings are far from solved. Crucially, to identify what breaks first at the edge, our roster scales down to 2B parameters. The findings expose a stark paradox: it is not the flying that fails. Small open models often navigate into the success radius more reliably than frontier models, yet lose the episode by declaring arrival prematurely or not at all. Multi-drone commanding amplifies this divide, with small models failing by blindly copying a single coordinate across distinct views. Viewed as vision-language-action agents, the models' spatial perception holds up, but their action protocol does not. What separates a deployable edge model from a frontier model is not navigation, but the discipline to sustain a declared protocol and emit the correct terminating action. The open problem is closing this gap at onboard compute costs---yielding a fast model that plans persistently and knows exactly when it is done---and DroneCATS is built to measure that distance.

---

### Measuring consistency via ensemble margin and local prediction variability: Auditing decision systems in the presence of predictive multiplicity

Authors: Sinjini Banerjee, Tim Marrinan, Anand D. Sarwate
Published: 2026-09-01
Categories: stat.ML, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2609.01397v1

The Rashomon effect is a machine learning phenomenon where equally accurate models produce different predictions for the same inputs (predictive multiplicity). Existing work primarily focuses on multiplicity within individual models, but in more complex decision systems, the impact of the Rashomon effect is less well understood. In this work, we study multiplicity from the perspective of auditing incorrect ensemble predictions, where the decision to divert an instance for human review is based on a consistency criterion that combines the ensemble margin with a measure of local prediction variability for each constituent model. With mild assumptions about stability and smoothness, we show that the consistency scores of finite ensembles converge to the corresponding consistency score of the expected model from the Rashomon set as the ensemble size and the number of samples used to measure local prediction variability increase. To demonstrate the efficacy of the proposed criterion, we evaluate the framework with respect to transformer models applied to natural language understanding tasks and parameter-efficient fine-tuning of large language models used for tabular data classification tasks. Our experiments show that ensembling models from the Rashomon set substantially reduces the risk of incorrect predictions going unchecked compared with auditing a single model, while incurring only a moderate increase in the number of diversions. Moreover, the auditing behavior of the full Rashomon set can be closely approximated by finite ensembles of relatively modest size, with the risk approaching zero for some datasets. We further demonstrate that the proposed measure exhibits stronger agreement with established predictive multiplicity metrics than existing consistency measures, providing a more reliable way to capture multiplicity in the Rashomon set.

---

### Scale-based Approach for Active Wildfire Segmentation on Satellite Imagery

Authors: Matheus F. Kovaleski, Cristiano Premebida, João Ruivo Paulo
Published: 2026-09-01
Categories: cs.CV
arXiv: https://arxiv.org/abs/2609.01392v1

Active wildfire mapping from satellite imagery is challenging due to the sparse and highly imbalanced nature of fire pixels, especially in early-stage or low-density fire observations. This work investigates the use of multispectral Landsat-8 imagery for active-fire segmentation under multi-scale wildfire size conditions. We propose a data-driven protocol to characterize fire-region size distributions through connected-component analysis and an interquartile range criterion, enabling the evaluation of model robustness across different local fire-region densities. Three segmentation architectures, U-Net, DeepLabV3+, and SegFormer, are evaluated under different SWIR-based spectral configurations. Results show that U-Net achieves the strongest robustness across the evaluated conditions, SegFormer provides competitive performance, and DeepLabV3+ tends to produce conservative predictions with reduced recall. Across architectures, SWIR2 consistently achieves the strongest or near-best results, highlighting its importance for active-fire segmentation in Landsat-8 imagery. These findings suggest that both spectral band selection and architectural design are critical for robust satellite-based active wildfire mapping trained on low active fire-pixel density images.

---

### Multimodal RGB-Infrared Combination for UAV-Based Wildfire Segmentation: A Comparative Study on FLAME3

Authors: Matheus F. Kovaleski, Luís Garrote, Cristiano Premebida, Jérôme Mendes, João Ruivo Paulo
Published: 2026-09-01
Categories: cs.CV
arXiv: https://arxiv.org/abs/2609.01390v1

Unmanned Aerial Vehicles (UAVs) have emerged as a promising platform for firefighting operations due to their flexibility, low operational cost, and ability to acquire high-resolution imagery in locations that may be difficult or dangerous to access using conventional methods. Recent advances in deep learning have significantly improved the capabilities of UAV-based wildfire monitoring systems. The present work investigates RGB-infrared fusion for binary wildfire segmentation on the FLAME3 dataset. In this Study, RGB and Infrared baselines are compared with three representative fusion strategies across three segmentation architectures, including U-Net, DeepLabV3+, and SegFormer. The key motivation of this work is to analyze the contribution of each modality, evaluate the impact of fusion timing, and examine how different network architectures exploit multimodal information for UAV wildfire delineation. The findings indicate that thermal information plays a dominant role in UAV segmentation and that feature-level multimodal fusion combined with transformer-based architectures offers the most promising direction for future research.

---