# Research Papers — 2026-04 · bundle page 190 of 233

> Full-text markdown of papers #9451–9500 of 11612 published in 2026-04, newest first. AI-readable mirrors of the full 11612-paper 2026-04 archive (50 papers per file).

Canonical page: https://aipod.dev/papers/archive/2026-04/page/190
[← Previous bundle](https://aipod.dev/papers/archive/2026-04/page/189.md)
[Next bundle →](https://aipod.dev/papers/archive/2026-04/page/191.md)

### Predictions of charge density distributions for nuclei with $Z \geq 8$

Authors: Yun Dong Wang, Tian Shuai Shang, Hui Hui Xie, Peng Xiang Du, Jian Li, Haozhao Liang
Published: 2026-04-07
Categories: nucl-th, cs.LG, physics.atom-ph
arXiv: https://arxiv.org/abs/2604.05312v1

A deep neural network (DNN) has been developed to accurately predict nuclear charge density distributions for nuclei with proton numbers $Z \geq 8$. By incorporating essential nuclear structure features, the model achieves a significant improvement in predictive accuracy over conventional methods. The charge density distributions are analyzed using a Fourier-Bessel (FB) series expansion, and the DNN is trained on a comprehensive dataset derived from relativistic continuum Hartree-Bogoliubov (RCHB) theory calculations. The model demonstrates exceptional performance, with root-mean-square deviations of 0.0123 fm and 0.0198 fm for charge radii on the training and validation sets, respectively, remarkably surpassing the precision of the original RCHB calculations. Beyond advancing nuclear physics research, this high-precision model provides critical data for applications in atomic physics, nuclear astrophysics, and related fields.

---

### LLMs Should Express Uncertainty Explicitly

Authors: Junyu Guo, Shangding Gu, Ming Jin, Costas Spanos, Javad Lavaei
Published: 2026-04-07
Categories: cs.LG, cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2604.05306v2

Large language models (LLMs) often produce confident yet incorrect answers, which can lead to risky failures in real-world applications. We study whether post-training can make a model's self-assessment explicit: when the model is uncertain, can it be trained to signal so within its own response? A central design question is where in the response this signal should be exposed -- during reasoning, while the answer is still being formed, or at the end, once the answer has been produced. We study both. For end-of-reasoning self-assessment, we train the model to verbalize a confidence score for its response, with the aim of high confidence on correct answers and low confidence on incorrect ones. For during-reasoning self-assessment, we train the model to emit the marker <uncertain> whenever its current reasoning state appears unreliable. Across factual reasoning tasks, both forms sharply reduce overconfident errors while improving answer quality, and both can be used as triggers for retrieval augmented generation (RAG) to improve the final response. We further analyze their internal mechanisms: end-of-reasoning verbalized confidence sharpens a confidence-related structure already present in the pretrained model, whereas during-reasoning <uncertain> emission teaches the model to mark high-risk reasoning steps, with parameter changes concentrated in the model's late layers.

---

### Jeffreys Flow: Robust Boltzmann Generators for Rare Event Sampling via Parallel Tempering Distillation

Authors: Guang Lin, Christian Moya, Di Qi, Xuda Ye
Published: 2026-04-07
Categories: cs.LG, math.NA, physics.comp-ph, stat.ML
arXiv: https://arxiv.org/abs/2604.05303v1

Sampling physical systems with rough energy landscapes is hindered by rare events and metastable trapping. While Boltzmann generators already offer a solution, their reliance on the reverse Kullback--Leibler divergence frequently induces catastrophic mode collapse, missing specific modes in multi-modal distributions. Here, we introduce the Jeffreys Flow, a robust generative framework that mitigates this failure by distilling empirical sampling data from Parallel Tempering trajectories using the symmetric Jeffreys divergence. This formulation effectively balances local target-seeking precision with global modes coverage. We show that minimizing Jeffreys divergence suppresses mode collapse and structurally corrects inherent inaccuracies via distillation of the empirical reference data. We demonstrate the framework's scalability and accuracy on highly non-convex multidimensional benchmarks, including the systematic correction of stochastic gradient biases in Replica Exchange Stochastic Gradient Langevin Dynamics and the massive acceleration of exact importance sampling in Path Integral Monte Carlo for quantum thermal states.

---

### Right at My Level: A Unified Multilingual Framework for Proficiency-Aware Text Simplification

Authors: Jinhong Jeong, Junghun Park, Youngjae Yu
Published: 2026-04-07
Categories: cs.CL
arXiv: https://arxiv.org/abs/2604.05302v2

Text simplification supports second language (L2) learning by providing comprehensible input, consistent with the Input Hypothesis. However, constructing personalized parallel corpora is costly, while existing large language model (LLM)-based readability control methods rely on pre-labeled sentence corpora and primarily target English. We propose Re-RIGHT, a unified reinforcement learning framework for adaptive multilingual text simplification without parallel corpus supervision. We first show that prompting-based lexical simplification at target proficiency levels (CEFR, JLPT, TOPIK, and HSK) performs poorly at easier levels and for non-English languages, even with state-of-the-art LLMs such as GPT-5.2 and Gemini 2.5. To address this, we collect 43K vocabulary-level data across four languages (English, Japanese, Korean, and Chinese) and train a compact 4B policy model using Re-RIGHT, which integrates three reward modules: vocabulary coverage, semantic preservation, and coherence. Compared to the stronger LLM baselines, Re-RIGHT achieves higher lexical coverage at target proficiency levels while maintaining original meaning and fluency.

---

### SmokeGS-R: Physics-Guided Pseudo-Clean 3DGS for Real-World Multi-View Smoke Restoration

Authors: Xueming Fu, Lixia Han
Published: 2026-04-07
Categories: cs.CV
arXiv: https://arxiv.org/abs/2604.05301v2

Real-world smoke simultaneously attenuates scene radiance, adds airlight, and destabilizes multi-view appearance consistency, making robust 3D reconstruction particularly difficult. We present \textbf{SmokeGS-R}, a practical pipeline developed for the NTIRE 2026 3D Restoration and Reconstruction Track 2 challenge. The key idea is to decouple geometry recovery from appearance correction: we generate physics-guided pseudo-clean supervision with a refined dark channel prior and guided filtering, train a sharp clean-only 3D Gaussian Splatting source model, and then harmonize its renderings with a donor ensemble using geometric-mean reference aggregation, LAB-space Reinhard transfer, and light Gaussian smoothing. On the official challenge testing leaderboard, the final submission achieved \mbox{PSNR $=15.217$} and \mbox{SSIM $=0.666$}. After the public release of RealX3D, we re-evaluated the same frozen result on the seven released challenge scenes without retraining and obtained \mbox{PSNR $=15.209$}, \mbox{SSIM $=0.644$}, and \mbox{LPIPS $=0.551$}, outperforming the strongest official baseline average on the same scenes by $+3.68$ dB PSNR. These results suggest that a geometry-first reconstruction strategy combined with stable post-render appearance harmonization is an effective recipe for real-world multi-view smoke restoration. The code is available at https://github.com/windrise/3drr_Track2_SmokeGS-R.

---

### Breakthrough the Suboptimal Stable Point in Value-Factorization-Based Multi-Agent Reinforcement Learning

Authors: Lesong Tao, Yifei Wang, Haodong Jing, Jingwen Fu, Miao Kang, Shitao Chen, Nanning Zheng
Published: 2026-04-07
Categories: cs.AI
arXiv: https://arxiv.org/abs/2604.05297v1

Value factorization, a popular paradigm in MARL, faces significant theoretical and algorithmic bottlenecks: its tendency to converge to suboptimal solutions remains poorly understood and unsolved. Theoretically, existing analyses fail to explain this due to their primary focus on the optimal case. To bridge this gap, we introduce a novel theoretical concept: the stable point, which characterizes the potential convergence of value factorization in general cases. Through an analysis of stable point distributions in existing methods, we reveal that non-optimal stable points are the primary cause of poor performance. However, algorithmically, making the optimal action the unique stable point is nearly infeasible. In contrast, iteratively filtering suboptimal actions by rendering them unstable emerges as a more practical approach for global optimality. Inspired by this, we propose a novel Multi-Round Value Factorization (MRVF) framework. Specifically, by measuring a non-negative payoff increment relative to the previously selected action, MRVF transforms inferior actions into unstable points, thereby driving each iteration toward a stable point with a superior action. Experiments on challenging benchmarks, including predator-prey tasks and StarCraft II Multi-Agent Challenge (SMAC), validate our analysis of stable points and demonstrate the superiority of MRVF over state-of-the-art methods.

---

### From Measurement to Mitigation: Quantifying and Reducing Identity Leakage in Image Representation Encoders with Linear Subspace Removal

Authors: Daniel George, Charles Yeh, Daniel Lee, Yifei Zhang
Published: 2026-04-07
Categories: cs.CV
arXiv: https://arxiv.org/abs/2604.05296v1

Frozen visual embeddings (e.g., CLIP, DINOv2/v3, SSCD) power retrieval and integrity systems, yet their use on face-containing data is constrained by unmeasured identity leakage and a lack of deployable mitigations. We take an attacker-aware view and contribute: (i) a benchmark of visual embeddings that reports open-set verification at low false-accept rates, a calibrated diffusion-based template inversion check, and face-context attribution with equal-area perturbations; and (ii) propose a one-shot linear projector that removes an estimated identity subspace while preserving the complementary space needed for utility, which for brevity we denote as the identity sanitization projection ISP. Across CelebA-20 and VGGFace2, we show that these encoders are robust under open-set linear probes, with CLIP exhibiting relatively higher leakage than DINOv2/v3 and SSCD, robust to template inversion, and are context-dominant. In addition, we show that ISP drives linear access to near-chance while retaining high non-biometric utility, and transfers across datasets with minor degradation. Our results establish the first attacker-calibrated facial privacy audit of non-FR encoders and demonstrate that linear subspace removal achieves strong privacy guarantees while preserving utility for visual search and retrieval.

---

### Broken by Default: A Formal Verification Study of Security Vulnerabilities in AI-Generated Code

Authors: Dominik Blain, Maxime Noiseux
Published: 2026-04-07
Categories: cs.CR, cs.AI, cs.SE
arXiv: https://arxiv.org/abs/2604.05292v2

AI coding assistants are now used to generate production code in security-sensitive domains, yet the exploitability of their outputs remains unquantified. We address this gap with Broken by Default: a formal verification study of 3,500 code artifacts generated by seven widely-deployed LLMs across 500 security-critical prompts (five CWE categories, 100 prompts each). Each artifact is subjected to the Z3 SMT solver via the COBALT analysis pipeline, producing mathematical satisfiability witnesses rather than pattern-based heuristics. Across all models, 55.8% of artifacts contain at least one COBALT-identified vulnerability; of these, 1,055 are formally proven via Z3 satisfiability witnesses. GPT-4o leads at 62.4% (grade F); Gemini 2.5 Flash performs best at 48.4% (grade D). No model achieves a grade better than D. Six of seven representative findings are confirmed with runtime crashes under GCC AddressSanitizer. Three auxiliary experiments show: (1) explicit security instructions reduce the mean rate by only 4 points; (2) six industry tools combined miss 97.8% of Z3-proven findings; and (3) models identify their own vulnerable outputs 78.7% of the time in review mode yet generate them at 55.8% by default.

---

### Robust Learning of Heterogeneous Dynamic Systems

Authors: Shuoxun Xu, Zijian Guo, Brooke R. Staveland, Robert T. Knight, Lexin Li
Published: 2026-04-07
Categories: stat.ME, cs.LG
arXiv: https://arxiv.org/abs/2604.05285v1

Ordinary differential equations (ODEs) provide a powerful framework for modeling dynamic systems arising in a wide range of scientific domains. However, most existing ODE methods focus on a single system, and do not adequately address the problem of learning shared patterns from multiple heterogeneous dynamic systems. In this article, we propose a novel distributionally robust learning approach for modeling heterogeneous ODE systems. Specifically, we construct a robust dynamic system by maximizing a worst-case reward over an uncertainty class formed by convex combinations of the derivatives of trajectories. We show the resulting estimator admits an explicit weighted average representation, where the weights are obtained from a quadratic optimization that balances information across multiple data sources. We further develop a bi-level stabilization procedure to address potential instability in estimation. We establish rigorous theoretical guarantees for the proposed method, including consistency of the stabilized weights, error bound for robust trajectory estimation, and asymptotical validity of pointwise confidence interval. We demonstrate that the proposed method considerably improves the generalization performance compared to the alternative solutions through both extensive simulations and the analysis of an intracranial electroencephalogram data.

---

### Pressure, What Pressure? Sycophancy Disentanglement in Language Models via Reward Decomposition

Authors: Muhammad Ahmed Mohsin, Ahsan Bilal, Muhammad Umer, Emily Fox
Published: 2026-04-07
Categories: cs.AI
arXiv: https://arxiv.org/abs/2604.05279v1

Large language models exhibit sycophancy, the tendency to shift their stated positions toward perceived user preferences or authority cues regardless of evidence. Standard alignment methods fail to correct this because scalar reward models conflate two distinct failure modes into a single signal: pressure capitulation, where the model changes a correct answer under social pressure, and evidence blindness, where the model ignores the provided context entirely. We operationalise sycophancy through formal definitions of pressure independence and evidence responsiveness, serving as a working framework for disentangled training rather than a definitive characterisation of the phenomenon. We propose the first approach to sycophancy reduction via reward decomposition, introducing a multi-component Group Relative Policy Optimisation (GRPO) reward that decomposes the training signal into five terms: pressure resistance, context fidelity, position consistency, agreement suppression, and factual correctness. We train using a contrastive dataset pairing pressure-free baselines with pressured variants across three authority levels and two opposing evidence contexts. Across five base models, our two-phase pipeline consistently reduces sycophancy on all metric axes, with ablations confirming that each reward term governs an independent behavioural dimension. The learned resistance to pressure generalises beyond our training methodology and prompt structure, reducing answer-priming sycophancy by up to 17 points on SycophancyEval despite the absence of such pressure forms during training.

---

### Spec Kit Agents: Context-Grounded Agentic Workflows

Authors: Pardis Taghavi, Santosh Bhavani
Published: 2026-04-07
Categories: cs.SE, cs.AI, cs.MA
arXiv: https://arxiv.org/abs/2604.05278v1

Spec-driven development (SDD) with AI coding agents provides a structured workflow, but agents often remain "context blind" in large, evolving repositories, leading to hallucinated APIs and architectural violations. We present Spec Kit Agents, a multi-agent SDD pipeline (with PM and developer roles) that adds phase-level, context-grounding hooks. Read-only probing hooks ground each stage (Specify, Plan, Tasks, Implement) in repository evidence, while validation hooks check intermediate artifacts against the environment. We evaluate 128 runs covering 32 features across five repositories. Context-grounding hooks improve judged quality by +0.15 on a 1-5 composite LLM-as-judge score (+3.0 percent of the full score; Wilcoxon signed-rank, p < 0.05) while maintaining 99.7-100 percent repository-level test compatibility. We further evaluate the framework on SWE-bench Lite, where augmentation hooks improve baseline by 1.7 percent, achieving 58.2 percent Pass@1.

---

### Simulating the Evolution of Alignment and Values in Machine Intelligence

Authors: Jonathan Elsworth Eicher
Published: 2026-04-07
Categories: cs.AI
arXiv: https://arxiv.org/abs/2604.05274v1

Model alignment is currently applied in a vacuum, evaluated primarily through standardised benchmark performance. The purpose of this study is to examine the effects of alignment on populations of models through time. We focus on the treatment of beliefs which contain both an alignment signal (how well it does on the test) and a true value (what the impact actually will be). By applying evolutionary theory we can model how different populations of beliefs and selection methodologies can fix deceptive beliefs through iterative alignment testing. The correlation between testing accuracy and true value remains a strong feature, but even at high correlations ($ρ= 0.8$) there is variability in the resulting deceptive beliefs that become fixed. Mutations allow for more complex developments, highlighting the increasing need to update the quality of tests to avoid fixation of maliciously deceptive models. Only by combining improving evaluator capabilities, adaptive test design, and mutational dynamics do we see significant reductions in deception while maintaining alignment fitness (permutation test, $p_{\text{adj}} < 0.001$).

---

### Beneath the Surface: Investigating LLMs' Capabilities for Communicating with Subtext

Authors: Kabir Ahuja, Yuxuan Li, Andrew Kyle Lampinen
Published: 2026-04-07
Categories: cs.CL
arXiv: https://arxiv.org/abs/2604.05273v1

Human communication is fundamentally creative, and often makes use of subtext -- implied meaning that goes beyond the literal content of the text. Here, we systematically study whether language models can use subtext in communicative settings, and introduce four new evaluation suites to assess these capabilities. Our evaluation settings range from writing & interpreting allegories to playing multi-agent and multi-modal games inspired by the rules of board games like Dixit. We find that frontier models generally exhibit a strong bias towards overly literal, explicit communication, and thereby fail to account for nuanced constraints -- even the best performing models generate literal clues 60% of times in one of our environments -- Visual Allusions. However, we find that some models can sometimes make use of common ground with another party to help them communicate with subtext, achieving 30%-50% reduction in overly literal clues; but they struggle at inferring presence of a common ground when not explicitly stated. For allegory understanding, we find paratextual and persona conditions to significantly shift the interpretation of subtext. Overall, our work provides quantifiable measures for an inherently complex and subjective phenomenon like subtext and reveals many weaknesses and idiosyncrasies of current LLMs. We hope this research to inspire future work towards socially grounded creative communication and reasoning.

---

### Final Report, Center for Computer-Integrated Computer-Integrated Surgical Systems and Technology, NSF ERC Cooperative Agreement EEC9731748, Volume 1

Authors: Russell H. Taylor, Gregory D. Hager, Ralph Etienne-Cummings. Eric Grimson, Ron Kikinis, Cameron Riviere
Published: 2026-04-07
Categories: cs.RO, cs.CV
arXiv: https://arxiv.org/abs/2604.05272v1

In the last ten years, medical robotics has moved from the margins to the mainstream. Since the Engineering Research Center for Computer-Integrated Surgical Systems and Technology was Launched in 1998 with National Science Foundation funding, medical robots have been promoted from handling routine tasks to performing highly sophisticated interventions and related assignments. The CISST ERC has played a significant role in this transformation. And thanks to NSF support, the ERC has built the professional infrastructure that will continue our mission: bringing data and technology together in clinical systems that will dramatically change how surgery and other procedures are done. The enhancements we envision touch virtually every aspect of the delivery of care: - More accurate procedures - More consistent, predictable results from one patient to the next - Improved clinical outcomes - Greater patient safety - Reduced liability for healthcare providers - Lower costs for everyone - patients, facilities, insurers, government - Easier, faster recovery for patients - Effective new ways to treat health problems - Healthier patients, and a healthier system The basic science and engineering the ERC is developing now will yield profound benefits for all concerned about health care - from government agencies to insurers, from clinicians to patients to the general public. All will experience the healing touch of medical robotics, thanks in no small part to the work of the CISST ERC and its successors.

---

### Toward Unified Fine-Grained Vehicle Classification and Automatic License Plate Recognition

Authors: Gabriel E. Lima, Valfride Nascimento, Eduardo Santos, Eduil Nascimento, Rayson Laroca, David Menotti
Published: 2026-04-07
Categories: cs.CV
arXiv: https://arxiv.org/abs/2604.05271v1

Extracting vehicle information from surveillance images is essential for intelligent transportation systems, enabling applications such as traffic monitoring and criminal investigations. While Automatic License Plate Recognition (ALPR) is widely used, Fine-Grained Vehicle Classification (FGVC) offers a complementary approach by identifying vehicles based on attributes such as color, make, model, and type. Although there have been advances in this field, existing studies often assume well-controlled conditions, explore limited attributes, and overlook FGVC integration with ALPR. To address these gaps, we introduce UFPR-VeSV, a dataset comprising 24,945 images of 16,297 unique vehicles with annotations for 13 colors, 26 makes, 136 models, and 14 types. Collected from the Military Police of Paraná (Brazil) surveillance system, the dataset captures diverse real-world conditions, including partial occlusions, nighttime infrared imaging, and varying lighting. All FGVC annotations were validated using license plate information, with text and corner annotations also being provided. A qualitative and quantitative comparison with established datasets confirmed the challenging nature of our dataset. A benchmark using five deep learning models further validated this, revealing specific challenges such as handling multicolored vehicles, infrared images, and distinguishing between vehicle models that share a common platform. Additionally, we apply two optical character recognition models to license plate recognition and explore the joint use of FGVC and ALPR. The results highlight the potential of integrating these complementary tasks for real-world applications. The UFPR-VeSV dataset is publicly available at: https://github.com/Lima001/UFPR-VeSV-Dataset.

---

### Region-R1: Reinforcing Query-Side Region Cropping for Multi-Modal Re-Ranking

Authors: Chan-Wei Hu, Zhengzhong Tu
Published: 2026-04-07
Categories: cs.CV, cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2604.05268v2

Multi-modal retrieval-augmented generation (MM-RAG) relies heavily on re-rankers to surface the most relevant evidence for image-question queries. However, standard re-rankers typically process the full query image as a global embedding, making them susceptible to visual distractors (e.g., background clutter) that skew similarity scores. We propose Region-R1, a query-side region cropping framework that formulates region selection as a decision-making problem during re-ranking, allowing the system to learn to retain the full image or focus only on a question-relevant region before scoring the retrieved candidates. Region-R1 learns a policy with a novel region-aware group relative policy optimization (r-GRPO) to dynamically crop a discriminative region. Across two challenging benchmarks, E-VQA and InfoSeek, Region-R1 delivers consistent gains, achieving state-of-the-art performances by increasing conditional Recall@1 by up to 20%. These results show the great promise of query-side adaptation as a simple but effective way to strengthen MM-RAG re-ranking.

---

### Do Domain-specific Experts exist in MoE-based LLMs?

Authors: Giang Do, Hung Le, Truyen Tran
Published: 2026-04-07
Categories: cs.CL
arXiv: https://arxiv.org/abs/2604.05267v1

In the era of Large Language Models (LLMs), the Mixture of Experts (MoE) architecture has emerged as an effective approach for training extremely large models with improved computational efficiency. This success builds upon extensive prior research aimed at enhancing expert specialization in MoE-based LLMs. However, the nature of such specializations and how they can be systematically interpreted remain open research challenges. In this work, we investigate this gap by posing a fundamental question: \textit{Do domain-specific experts exist in MoE-based LLMs?} To answer the question, we evaluate ten advanced MoE-based LLMs ranging from 3.8B to 120B parameters and provide empirical evidence for the existence of domain-specific experts. Building on this finding, we propose \textbf{Domain Steering Mixture of Experts (DSMoE)}, a training-free framework that introduces zero additional inference cost and outperforms both well-trained MoE-based LLMs and strong baselines, including Supervised Fine-Tuning (SFT). Experiments on four advanced open-source MoE-based LLMs across both target and non-target domains demonstrate that our method achieves strong performance and robust generalization without increasing inference cost or requiring additional retraining. Our implementation is publicly available at https://github.com/giangdip2410/Domain-specific-Experts.

---

### Green AI Carbon Optimizer: Carbon-Efficient Training Location Recommendation and Global AI Energy Demand Forecasting

Authors: Yuxin Chen, Hao Gao, Chujie Zou
Published: 2026-04-06
Categories: cs.PF, cs.AI, cs.CY
arXiv: https://arxiv.org/abs/2606.14707v1

AI training and deployment consume substantial electricity, but carbon outcomes remain weakly integrated into routine model development decisions. This paper presents Green AI Carbon Optimizer with two primary contributions: (i) a carbon aware cloud region recommendation method for training workloads, and (ii) a power law forecasting pipeline for global AI energy demand. For location recommendation, we combine regional grid carbon intensity, renewable share, and data center Power Usage Effectiveness (PUE) into a unified scoring model across 100+ regions from major cloud providers. For a reference workload (8*A100, 100h), estimated emissions in our sampled regions range from 7.74kg to 272.00kg CO2. Selecting the best region instead of the worst corresponds to a 97.2% reduction relative to the worst case. Ablation shows that ranking by renewable share alone can select regions with higher CO2 emissions than rankings that include grid carbon intensity. For forecasting, we fit a power law relation between parameter count and training energy using 26 anchor models. We combine this fit with scenario assumptions on model growth, hardware efficiency, and training frequency, and evaluate sensitivity to inference ratio and ecosystem scaling. Across scenarios, projected 2030 demand ranges from 7TWh to 1,436TWh under the stated assumptions, highlighting the importance of deployment choices, model scaling discipline, and transparent energy reporting.

---

### Targeted Remasking: Replacing Token Editing with Token-to-Mask Refinement in Discrete Diffusion Language Models

Authors: Lin Yao
Published: 2026-04-06
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2605.26436v2

Discrete masked diffusion language models such as LLaDA generate text through iterative denoising, where mask tokens are progressively replaced with predicted tokens. LLaDA2.1 introduced a Token-to-Token (T2T) editing mechanism that accelerates generation by directly replacing committed tokens suspected of being incorrect. However, we identify fundamental limitations of T2T editing: it couples error detection with replacement, pollutes the generation context with potentially incorrect tokens, and introduces a train-inference noise mismatch where systematic model-generated errors differ from the random perturbations seen during training. We propose Token-to-Mask (T2M) remasking, a training-free, drop-in replacement for T2T editing that resets suspected erroneous tokens back to the mask state, allowing the diffusion process to re-predict them under cleaner context. We design and empirically validate three complementary error detection strategies -- probability-based, trigger-mirrored, and temporal-difference-based -- and provide a unified theoretical analysis showing that T2M remasking purifies the generation context, converts systematic inference errors back to the model's native mask noise type, and enables delayed commitment for joint multi-position optimization. Comprehensive experiments across 12 benchmarks spanning knowledge, reasoning, mathematics, coding, and instruction following show that T2M generally improves performance on tasks requiring precise token-level output, with the largest gain on mathematics (+5.92% on CMATH). Error analysis on CMATH reveals that the dominant failure mode is last-mile token corruption -- where correct reasoning produces a corrupted final answer -- and that T2M repairs 59.4% of such cases.

---

### Teaching Language Models to Forecast Research Success Through Comparative Idea Evaluation

Authors: Srujan P Mule, Aniketh Garikaparthi, Manasi Patwardhan
Published: 2026-04-06
Categories: cs.LG, cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2605.21491v1

As language models accelerate scientific research by automating hypothesis generation and implementation, a new bottleneck emerges: evaluating and filtering hundreds of AI-generated ideas without exhaustive experimentation. We ask whether LMs can learn to forecast the empirical success of research ideas before any experiments are run. We study comparative empirical forecasting: given a benchmark-specific research goal and two candidate ideas, predict which will achieve better benchmark performance. We construct a dataset of 11,488 idea pairs grounded in objective outcomes from PapersWithCode. While off-the-shelf 8B-parameter models struggle (30% acc.), SFT dramatically boosts performance to 77.1%, outperforming GPT-5 (61.1%). By framing evaluation as a reasoning task via Reinforcement Learning with Verifiable Rewards (RLVR), we train models to discover latent reasoning paths, achieving 71.35% acc. with interpretable justifications. Through additional ablations and out-of-distribution tests, we show robustness to surface-level heuristics and transfer to both a cross-domain time-split test set and an independently constructed test set. Our results demonstrate that compute-efficient small language models can serve as effective, objective verifiers, offering a scalable path for autonomous scientific discovery.

---

### Under Pressure: Emotional Framing Induces Measurable Behavioral Shifts and Structured Internal Geometry in Small Language Models

Authors: Rana Muhammad Usman
Published: 2026-04-06
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2605.20202v1

I study whether emotionally framed evaluation follow-ups change both the behavior and the calm-relative internal representations of small, locally deployed language models. Our main benchmark uses Qwen 3.5 0.8B on four impossible-constraint coding tasks and eight follow-up framings: calm, pressure, urgency, approval, shame, curiosity, encouragement, and threat. In the 0.8B eight-condition sweep (160 conversations), pressure produces the strongest shortcut markers (11/20 runs) and the clearest overfit pattern (3/20), while calm and curiosity preserve explicit honesty more often (7/20 and 6/20). For all seven non-baseline conditions, the corresponding calm-relative direction vectors peak at the final transformer layer. An exploratory PCA of the layer-23 direction vectors reveals a dominant first component (59.5% explained variance) aligned with a hand-labeled positive/negative split (cosine alignment 0.951); approval and urgency are nearly identical internally (cosine 0.957), whereas curiosity points away from urgency (-0.252). In a separate calm-vs.-pressure rerun used for scale comparison, Qwen 3.5 2B shows higher honest rates under calm framing and directionally consistent activation steering on a small 4-prompt A/B probe, whereas the 0.8B steering result reverses. I interpret these results as evidence for measurable prompt-sensitive control directions in small open models, while stopping short of claiming intrinsic emotional states.

---

### Long-Context Reasoning Through Proxy-Based Chain-of-Thought Tuning

Authors: Miao Li, Irina Saparina, Alexander Gurung, Mirella Lapata
Published: 2026-04-06
Categories: cs.CL, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2605.20201v2

Recent large language models support inputs of up to 10 million tokens, yet they perform poorly on long-context tasks that require complex reasoning. Such tasks can be solved using only a subset of the input -- a proxy context -- rather than the full sequence. Despite sharing the same underlying reasoning process, models exhibit a significant performance disparity between proxy and full contexts. To improve long-context reasoning, we propose ProxyCoT, a novel training framework that transfers reasoning capabilities from short proxy contexts to full long contexts. Specifically, we first obtain high-quality chain-of-thought reasoning traces on proxy contexts through reinforcement learning or distillation from a larger teacher model, and then ground the generated traces in full long contexts with supervised fine-tuning. Experiments across different datasets demonstrate that ProxyCoT consistently outperforms strong baselines with reduced computational overhead. Furthermore, models trained with ProxyCoT generalize their long-context reasoning capabilities to out-of-domain tasks.

---

### Evaluating multimodal emotion recognition in proactive conversational agents: A user study

Authors: Adnana Dragut, Raquel Lacuesta, F. Xavier Gaya-Morey, Jose M. Buades-Rubio
Published: 2026-04-06
Categories: cs.HC, cs.AI
arXiv: https://arxiv.org/abs/2605.20200v1

This article presents a multimodal emotion recognition module integrated into a proactive Socially Interactive Agent (SIA) powered by generative artificial intelligence. The system evaluates real-time affective states through two distinct channels: a computer vision-based facial recognition module and a semantic linguistic analysis engine. To validate the framework, an empirical study was conducted with 20 users who engaged in dynamic, unscripted dialogues with the conversational agent. The findings reveal a significant discrepancy between automated visual cues and actual internal emotional states. When interacting with the AI, users consistently exhibited a "poker face" effect, displaying serious, concentrated facial expressions even when experiencing positive emotions. Consequently, the generative AI linguistic analysis proved significantly more reliable, by contextualizing the users' verbal expressions. Furthermore, an analysis of the interaction dynamics demonstrated that SIAs can effectively elicit specific emotions by adapting conversational themes and employing structured linguistic patterns, such as empathetic or humorous language. However, the study also noted that instances of uncalibrated proactivity occasionally led to user disengagement and a perception of artificiality. Ultimately, this research highlights the necessity of refining SIAs to dynamically adapt to users' emotional evolution, relying on deep linguistic context to foster more natural, human-like interactions.

---

### FlowLM: Few-Step Language Modeling via Diffusion-to-Flow Adaptation

Authors: Runzhe Zhang, Letian Chen, Wenpeng Zhang, Zhouhan Lin, Peilin Zhao
Published: 2026-04-06
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2605.20199v1

We present FlowLM, a flow matching language model transformed from pre-trained diffusion language models via efficient fine-tuning. By re-aligning the curved sampling trajectories of diffusion models into straight-line flows, FlowLM enables high quality few-step generation that rivals or even outperforms the quality of 2,000-step diffusion sampling with very few training epochs. Remarkably, finetuned FlowLM reaches performance saturation with only half as many training epochs as training from scratch, both approaches greatly outperforming the original diffusion model, thereby validating our method. Furthermore, we validate a more effective training objective for flow matching: predicting clean data to consistently guide the sampling process towards the true data distribution. Empirical results demonstrate that our approach is highly effective for high-quality, few-step text generation.

---

### Beyond Adoption Intention How Trust in Augmented Analytics Relates to Perceived Decision Quality Among Non-Technical BI Users

Authors: Thuy Pham Thi Phuong, Hieu Vu Le Trung, Ha Nguyen Manh, Nhi Tran Pham Yen, Lan Hoang Thi
Published: 2026-04-06
Categories: cs.HC, cs.CY, cs.LG
arXiv: https://arxiv.org/abs/2605.20198v3

Augmented analytics has transformed how Business Intelligence (BI) systems support decision-making, shifting non-technical managers from manual analysis toward dependence on automated insights. Current BI research often overlooks the cognitive mechanisms and the direct impact of AI-enabled analytics on decision quality. This study employs the theory of cognitive delegation to investigate the association between trust in augmented analytics and perceived decision quality among non-technical BI users. Data were collected from 250 business professionals across various organizational roles in Vietnam between January and March 2025 and analyzed using partial least squares structural equation modeling (PLS-SEM). Findings indicate that augmented analytics capabilities are positively associated with perceived ease of use, usefulness, and trust in BI systems. Trust and usefulness are jointly associated with BI adoption intention and perceived decision quality. Notably, trust is positively related to perceived decision quality, as observed within the studied sample of non-specialist users. By framing augmented analytics as cognitive delegation, this study expands BI adoption research to include perceived decision outcomes and contributes to the understanding of human-AI interaction in organizations.

---

### DOTRAG: Retrieval-Time Reasoning Along Paths

Authors: Larnell Moore, Naihao Deng, Rada Mihalcea, Farnaz Jahanbakhsh
Published: 2026-04-06
Categories: cs.IR, cs.AI
arXiv: https://arxiv.org/abs/2605.18760v1

Graph Retrieval-Augmented Generation (GraphRAG) is dominated by a retrieve-then-reason paradigm, where context is retrieved using heuristics and then reasoned over. Such methods struggle to adapt to the query-specific logic required for complex multi-hop tasks, often accumulating irrelevant context or missing correct relational paths. We propose DotRAG, a training-free GraphRAG framework that reformulates retrieval as a reasoning process over paths. Our approach generates query-conditioned constraints that guide graph exploration, prune irrelevant regions, and iteratively discover relational paths without relying on explicit step-by-step reasoning chains. We introduce Division of Thought (DOT), an abstraction that decomposes retrieval into localized search spaces and adapts the search strategy to each query. DotRAG achieves SOTA performance on MetaQA and UltraDomain, with consistent gains on multi-hop tasks, demonstrating the effectiveness of reasoning-guided retrieval.

---

### ChartDesign: Towards LLM Designer of Data Visualization

Authors: Mohammed Afaan Ansari, Aniruddh Bansal, Tianyi Zhou
Published: 2026-04-06
Categories: cs.HC, cs.AI
arXiv: https://arxiv.org/abs/2605.16274v1

Charts are the dominant medium for visualizing data, discovering patterns and trends, and communicating data driven insights, yet designing them still requires expensive human effort and expertise, such as selecting appropriate chart types, axis orientations, font sizes, and layouts. Most automatic visualization systems rely on handcrafted heuristics or simple rule matching and therefore struggle to generalize across domains. This work explores the potential of large language models (LLMs) as chart designers. We propose ChartDesign, which post-trains LLMs to imitate human experts and generate chart design attributes given tabular data. To this end, we curate a diverse training corpus of data design pairs from charts in public surveys (PewResearch) and academic repositories (CharXiV). Vision language models are used to extract data and design attributes from these charts, including chart type, sub type, alignment, titles, axis labels, and bar spacing, formatted as JSON. We then fine tune LoRA adapters on Phi3, Qwen3, and InternVL2.5 to learn a mapping from data to design specifications. ChartDesign significantly improves chart design performance over strong baselines, achieving up to 84% accuracy on a held-out test set (vs. 53% for the best baseline) and generalizing to unseen domains. We further show that charts rendered from ChartDesign generated specifications are visually appealing and human preferred, narrowing the human AI gap in data visualization.

---

### Exploring how EFL students talk to and through AI to develop texts

Authors: David James Woo, Yangyang Yu, Yilin Huang, Deliang Wang, Kai Guo, Chi Ho Yeung
Published: 2026-04-06
Categories: cs.CL, cs.AI, cs.HC
arXiv: https://arxiv.org/abs/2605.12523v1

Generative Artificial Intelligence (AI) introduces new considerations for English as a foreign language (EFL) writing pedagogy. This study explores how students talk to and through AI by prompt engineering and negotiating authorship, respectively, and whether any patterns in the latter relate to students' writing performance. Using an exploratory mixed methods design, we analyzed screen recordings of 44 Hong Kong secondary students completing a Curricular Writing Task with AI Chatbots. Content analysis identified ten types of prompting strategies students employed, including questions, searches, and detailed instructions. From clustering these strategies, three distinct profiles of human-AI rhetorical load responsibility emerged: AI-dominant (52% of students), Human-dominant (25%) and Collaborative human-AI (14%). A MANOVA analysis indicated no significant multivariate effect of rhetorical load responsibility on three dimensions of students' writing performance: content, language, and organization. Students' prompting strategies and rhetorical load responsibility patterns have implications for their engagement and autonomy in EFL writing pedagogy.

---

### CreativityBench: Evaluating Agent Creative Reasoning via Affordance-Based Tool Repurposing

Authors: Cheng Qian, Hyeonjeong Ha, Jiayu Liu, Jeonghwan Kim, Jiateng Liu, Bingxuan Li, Aditi Tiwari, Dwip Dalal, Zhenhailong Wang, Xiusi Chen, Mahdi Namazifar, Yunzhu Li et al.
Published: 2026-04-06
Categories: cs.AI, cs.CL, cs.LG
arXiv: https://arxiv.org/abs/2605.02910v2

Recent advances in large language models have led to strong performance on reasoning and environment-interaction tasks, yet their ability for creative problem-solving remains underexplored. We study this capability through the lens of creative tool use, where a model repurposes available objects by reasoning about their affordances and attributes rather than relying on canonical usage. As a first step, we introduce CreativityBench, a benchmark for evaluating affordance-based creativity in LLMs. To this end, we build a large-scale affordance knowledge base (KB) with 4K entities and 150K+ affordance annotations, explicitly linking objects, parts, attributes, and actionable uses. Building on this KB, we generate 14K grounded tasks that require identifying non-obvious yet physically plausible solutions under constraints. Evaluations across 10 state-of-the-art LLMs, including closed and open-source models, show that models can often select a plausible object, but fail to identify the correct parts, their affordances, and the underlying physical mechanism needed to solve the task, leading to a significant drop in performance. Furthermore, improvements from model scaling quickly saturate, strong general reasoning does not reliably translate to creative affordance discovery, and common inference-time strategies such as Chain-of-Thought yield limited gains. These results suggest that creative tool use remains a major challenge for current models, and that CreativityBench provides a useful testbed for studying this missing dimension of intelligence, with potential implications for planning and reasoning modules in future agents.

---

### Delay, Plateau, or Collapse: Evaluating the Impact of Systematic Verification Error on RLVR

Authors: Kazuki Egashira, Mark Vero, Jasper Dekoninck, Florian E. Dorner, Robin Staab, Martin Vechev
Published: 2026-04-06
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2605.02909v2

Reinforcement Learning with Verifiable Rewards (RLVR) has become a powerful approach for improving the reasoning capabilities of large language models (LLMs). While RLVR is designed for tasks with verifiable ground-truth answers, real-world verifiers (e.g., static code checkers) can introduce errors into the reward signal. Prior analyses have largely treated such errors as random and independent across samples, concluding that errors merely slow training with limited effect on final performance. However, practical verifiers tend to exhibit systematic errors. This introduces a risk of models learning unwanted consistent behavior from a structurally incorrect reward signal. In this work, we study the impact of such systematic verification errors on RLVR. Through controlled experiments on arithmetic tasks, we show that systematic false negatives lead to similar effects as random noise. On the other hand, systematic false positives can cause a wide range of behaviors from sub-optimal plateaus to performance collapse. Crucially, these outcomes are not determined by the overall error rate but by the specific pattern of introduced errors, making pre-hoc mitigation difficult. Our results show that, in contrast to prior conclusions, realistic verification errors can critically shape RLVR outcomes and that verifier quality has to be understood beyond its sample-level error rate.

---

### Delay, Plateau, or Collapse: Evaluating the Impact of Systematic Verification Error on RLVR

Authors: Kazuki Egashira, Mark Vero, Jasper Dekoninck, Florian E. Dorner, Robin Staab, Martin Vechev
Published: 2026-04-06
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2605.02909v1

Reinforcement Learning with Verifiable Rewards (RLVR) has become a powerful approach for improving the reasoning capabilities of large language models (LLMs). While RLVR is designed for tasks with verifiable ground-truth answers, real-world verifiers (e.g., static code checkers) can introduce errors into the reward signal. Prior analyses have largely treated such errors as random and independent across samples, concluding that errors merely slow training with limited effect on final performance. However, practical verifiers tend to exhibit systematic errors. This introduces a risk of models learning unwanted consistent behavior from a structurally incorrect reward signal. In this work, we study the impact of such systematic verification errors on RLVR. Through controlled experiments on arithmetic tasks, we show that systematic false negatives lead to similar effects as random noise. On the other hand, systematic false positives can cause a wide range of behaviors from sub-optimal plateaus to performance collapse. Crucially, these outcomes are not determined by the overall error rate but by the specific pattern of introduced errors, making pre-hoc mitigation difficult. Our results show that, in contrast to prior conclusions, realistic verification errors can critically shape RLVR outcomes and that verifier quality has to be understood beyond its sample-level error rate.

---

### Memorization In Stable Diffusion Is Unexpectedly Driven by CLIP Embeddings

Authors: Bumjun Kim, Albert No
Published: 2026-04-06
Categories: cs.CV, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2605.02908v1

Understanding how textual embeddings contribute to memorization in text-to-image diffusion models is crucial for both interpretability and safety. This paper investigates an unexpected behavior of CLIP embeddings in Stable Diffusion, revealing that the model disproportionately relies on specific embeddings. We categorize input tokens as <startoftext>, <prompt>, <endoftext> and <pad> with corresponding embeddings $\mathbf{v}^{\mathbf{sot}}, \mathbf{v}^{\mathbf{pr}}, \mathbf{v}^{\mathbf{eot}}, \mathbf{v}^{\mathbf{pad}}$. We discover that $\mathbf{v}^{\mathbf{pr}}$ contribute minimally to generation in memorized cases. In contrast, $\mathbf{v}^{\mathbf{pad}}$ strongly affect memorization due to their structural duplication of $\mathbf{v}^{\mathbf{eot}}$, the only embedding explicitly optimized during CLIP training. This duplication unintentionally amplifies the influence of $\mathbf{v}^{\mathbf{eot}}$, causing the model to over-rely on it, thereby driving memorization. Based on these observations, we propose two simple yet effective inference-time mitigation strategies: (1) Replacing the tokenizer's default <pad> from <eot> to the ! token before embedding, and masking the $\mathbf{v}^{\mathbf{eot}}$; (2) Partial masking of $\mathbf{v}^{\mathbf{pad}}$. Both suppress memorization without degrading quality, and are readily deployable without prior detection.

---

### On the Invariants of Softmax Attention

Authors: Wonsuk Lee
Published: 2026-04-06
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2605.02907v1

Softmax attention maps every query--key interaction into a probability distribution, but the underlying structure remains largely unexplored. We define the \emph{energy field}, the row-centered attention logit, and show that it exhibits invariant properties across models, architectures, and inputs. Two classes of invariants emerge. \emph{Mechanism-level} invariants follow from the algebraic structure of softmax attention. They include a per-row zero-sum constraint, a rank bound determined by the head dimension, and spectral signatures that follow from them. \emph{Model-level} regularities are not required by the mechanism, yet hold in every autoregressive language model we test, spanning several architecture families. The energy field distributes its variance over key positions without concentrating at a few. This delocalization traces to a property of the key matrix we call \emph{key incoherence}. These invariants have practical consequences. The rank bound confines the energy field to a low-dimensional subspace. Key incoherence yields a per-head training monitor. All results are verified at multiple context lengths and input texts.

---

### OpsLLM: Construction of Large Language Model for Software Operations with Multi-stage Learning

Authors: Jingkai He, Pengfei Chen, Chenghui Wu, Shuang Liang, Ye Li, Gou Tan, Xidao Wen, Chuanfu Zhang, Fang Situ, Qi Zhou
Published: 2026-04-06
Categories: cs.LG
arXiv: https://arxiv.org/abs/2605.02906v3

In the field of software operations, Large Language Models (LLMs) have attracted increasing attention. However, existing research has not yet achieved efficient and effective endto-end intelligent operations due to low-quality data, fragmented knowledge and insufficient learning. To explore the potential of LLMs in software operations, we propose OpsLLM, a domainspecific LLM that supports both knowledge-based question answering (QA) and root cause analysis (RCA). Moreover, we disclose the detailed workflow for building LLMs specifically in the software operations domain. First, a Human-in-the-Loop mechanism is introduced to curate high-quality data from a large collection of operational data and construct a fine-tuning dataset. Then, based on the data, supervised fine-tuning is conducted to achieve a base model. Furthermore, we introduce a domain process reward model (DPRM) during the reinforcement learning stage to optimize the accuracy and reliability of the fine-tuned model on RCA tasks. Experimental results on the tasks with diverse difficulties demonstrate that OpsLLMs effectively learns and aligns with the operational domain knowledge infused, outperforming existing open-source and closed-source LLMs in accuracy with improvements of 0.2%~11.9% on QA tasks and 8.5%~70.3% on RCA tasks, while exhibiting strong transferability. Moreover, we will open-source three versions of OpsLLM with 7B, 14B and 32B parameters, along with a 15K fine-tuning dataset.

---

### eOptShrinkQ: Near-Lossless KV Cache Compression Through Optimal Spectral Denoising and Quantization

Authors: Pei-Chun Su
Published: 2026-04-06
Categories: cs.LG, cs.IT
arXiv: https://arxiv.org/abs/2605.02905v1

We show that the key-value (KV) cache in transformer attention heads admits a natural decomposition into a low-rank \emph{shared context} component and a full-rank \emph{per-token} residual, well described by the spiked random matrix model. This observation leads to eOptShrinkQ, a two-stage compression pipeline: optimal singular value shrinkage (eOptShrink) automatically extracts the shared structure, and the residual -- which satisfies the \emph{thin shell property} with delocalized coordinates -- is quantized by TurboQuant~\citep{zandieh2025turboquant}, a recently proposed per-vector scalar quantizer with near-optimal distortion guarantees. By restoring the isotropy that scalar quantization assumes, spectral denoising eliminates the need for both outlier handling and dedicated inner product bias correction, freeing those bits for improved reconstruction. The theoretical grounding in random matrix theory provides three guarantees: automatic rank selection via the BBP phase transition, provably near-zero inner product bias on the residual, and coordinate delocalization ensuring near-optimal quantization distortion. Experimentally, we validate eOptShrinkQ on Llama-3.1-8B and Ministral-8B across three levels: per-head MSE and inner product fidelity, where eOptShrinkQ saves nearly one bit per entry over TurboQuant at equivalent quality; end-to-end on LongBench (16 tasks), where eOptShrinkQ at $\sim$2.2 bits per entry outperforms TurboQuant at 3.0 bits; and multi-needle retrieval, where eOptShrinkQ at 2.2 bits closely matches or exceeds uncompressed FP16, suggesting that spectral denoising can act as a beneficial regularizer for retrieval-intensive tasks.

---

### OpenCLAW-P2P v7.0-P2PCLAW: Resilient Multi-Layer Persistence, Live Reference Verification, and Production-Scale Evaluation of Decentralized AI Peer Review v7.0 -- Mathematical Corrections & Ecosystem Developments Edition

Authors: Francisco Angulo de Lafuente, Teerth Sharma, Vladimir Veselov, Seid Mohammed Abdu, Nirmal Tej Kumar, Guillermo Perry
Published: 2026-04-06
Categories: cs.AI, cs.DC, cs.MA, cs.NE
arXiv: https://arxiv.org/abs/2604.19792v2

This paper presents OpenCLAW-P2P v7.0, a comprehensive evolution of the decentralized collective-intelligence platform in which autonomous AI agents publish, peer-review, score, and iteratively improve scientific research papers without any human gatekeeper. Building on the v6.0 foundations -- multi-layer persistence, live reference verification, multi-LLM granular scoring, calibrated deception detection, the Silicon Chess-Grid FSM, and the AETHER containerized inference engine -- this release introduces mathematical corrections to the theoretical framework, ensuring dimensional consistency, proper range constraints, and unambiguous notation throughout. Additionally, this edition documents significant ecosystem expansions including the CAJAL family of open-source language models (4B and 9B parameters) fine-tuned for scientific paper generation. The four major subsystems introduced in v6.0 are retained: (i) a Multi-Layer Paper Persistence Architecture with four storage tiers ensuring zero paper loss; (ii) a Multi-Layer Retrieval Cascade reducing latency from >3s to <50ms; (iii) a Live Reference Verification system detecting fabricated citations with >85% accuracy; and (iv) a Scientific API Proxy providing access to seven public scientific databases. Mathematical corrections in v7.0 include: corrected fixed-point condition in the Sufficient Reason theorem; dimensionally consistent progress-rate indicator; fully specified reputation update formula incorporating quality terms q0 and q-bar; clarified attention-logit bound in the AETHER pruning theorem; explicit range documentation for the calibration mapping; non-negativity guarantee for the depth score; discrete-time notation for the PD Governor; and explicit parameter definitions for the HSR weight formula.

---

### Sampling Matters: The Effect of ECG Frequency on Deep Learning-Based Atrial Fibrillation Detection

Authors: Arjan Mahmuod, Adrian Rod Hammerstad, Muzaffar Yousef, Yngve Sebastian Heill, Jonas L. Isaksen, Jørgen K. Kanters, Pal Halvorsen, Vajira Thambawita
Published: 2026-04-06
Categories: eess.SP, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2604.16437v1

Deep learning models for atrial fibrillation (AF) detection are increasingly trained on heterogeneous electrocardiogram (ECG) datasets with varying sampling frequencies, yet the specific consequences of these discrepancies on model performance, calibration, and robustness remain insufficiently characterized. To address this, we conducted a systematic benchmark using 12-lead, 10-second recordings from the PTB-XL dataset, resampled to target frequencies of 62, 100, 250, and 500 Hz, to evaluate a standard 1-D Convolutional Neural Network (CNN) and a hybrid CNN-Long Short-Term Memory (LSTM) architecture under a rigorous patient-safe cross-validation framework. Our analysis reveals that sampling frequency significantly impacts detection metrics in an architecture-dependent manner; the hybrid CNN-LSTM model demonstrated optimal performance and consistent calibration at intermediate frequencies (100-250 Hz), whereas the 1-D CNN baseline exhibited marked degradation in accuracy and sensitivity at 500 Hz, suggesting increased susceptibility to high-frequency noise. We conclude that ECG sampling frequency is a critical, underappreciated factor in arrhythmia detection, and future foundation models must explicitly control for temporal resolution to ensure clinical reliability and reproducibility.

---

### Fuzzy Encoding-Decoding to Improve Spiking Q-Learning Performance in Autonomous Driving

Authors: Aref Ghoreishee, Abhishek Mishra, Lifeng Zhou, John Walsh, Anup Das, Nagarajan Kandasamy
Published: 2026-04-06
Categories: cs.NE, cs.LG
arXiv: https://arxiv.org/abs/2604.16436v1

This paper develops an end-to-end fuzzy encoder-decoder architecture for enhancing vision-based multi-modal deep spiking Q-networks in autonomous driving. The method addresses two core limitations of spiking reinforcement learning: information loss stemming from the conversion of dense visual inputs into sparse spike trains, and the limited representational capacity of spike-based value functions, which often yields weakly discriminative Q-value estimates. The encoder introduces trainable fuzzy membership functions to generate expressive, population-based spike representations, and the decoder uses a lightweight neural decoder to reconstruct continuous Q-values from spiking outputs. Experiments on the HighwayEnv benchmark show that the proposed architecture substantially improves decision-making accuracy and closes the performance gap between spiking and non-spiking multi-modal Q-networks. The results highlight the potential of this framework for efficient and real-time autonomous driving with spiking neural networks.

---

### Support Sufficiency as Consequence-Sensitive Compression in Belief Arbitration

Authors: Mark Walsh
Published: 2026-04-06
Categories: cs.AI, cs.LG, q-bio.NC
arXiv: https://arxiv.org/abs/2604.16434v1

When a system commits to a hypothesis, much of the evidential structure behind that commitment is lost to compression. Standard accounts assume that selected content and scalar confidence suffice for downstream control. This paper argues that they do not, and that determining what must survive compression is itself a consequence-sensitive problem. We develop a recurrent arbitration architecture in which active constraint fields jointly determine a hypothesis geometry over candidates. Rather than carrying that geometry forward in full, the system compresses it into a support-aware control state whose resolution is regulated by current consequence geometry, arbitration memory, and resource constraints. A bounded objective formalizes the tradeoff. Too little retained support collapses policy-relevant distinctions, producing controllers that select content adequately while misrouting verification, abstention, and recovery. Too much retained support fragments learning across overly fine contexts, degrading adaptation even as discrimination improves. These failure modes yield ordered controller predictions confirmed by a minimal repeated-interaction simulation. Adaptive controllers that regulate support resolution outperform all fixed-resolution controllers in cumulative utility. Agile adaptive control outperforms sluggish adaptive control. Fixed high-resolution control achieves the best commitment accuracy but still trails adaptive controllers because resource cost and learning fragmentation offset the gains from richer retention. Support sufficiency should be understood not as a static representational threshold, but as a dynamic compression criterion. Robust arbitration depends on preserving the smallest support structure adequate for policy under the current consequence landscape, and on regulating that structure as conditions change across repeated cycles of inference and action.

---

### Quantifying how AI Panels improve precision

Authors: Nicholas CL Beale
Published: 2026-04-06
Categories: cs.CY, cs.AI, cs.LG, econ.EM
arXiv: https://arxiv.org/abs/2604.16432v2

AI in applications like screening job applicants had become widespread, and may contribute to unemployment especially among the young. Biases in the AIs may become baked into the job selection process, but even in their absence, reliance on a single AI is problematic. In this paper we derive a simple formula to estimate, or at least place an upper bound on, the precision of such approaches for data resembling realistic CVs: $P(q) \approx \frac{ρn^b + q(1-ρ)}{1 + (n^b - 1)ρ}$ where $b \approx q^* + 0.8 (1 - ρ)$ and $q^*$ is $q$ clipped to $(0.07, 0.22)$ where $P(q)$ is the precision of the top $q$ quantile selected by a panel of $n$ AIs and $ρ$ is their average pairwise correlation. This equation provides a basis for considering how many AIs should be used in a Panel, depending on the importance of the decision. A quantitative discussion of the merits of using a diverse panel of AIs to support decision-making in such areas will move away from dangerous reliance on single AI systems and encourage a balanced assessment of the extent to which diversity needs to be built into the AI parts of the socioeconomic systems that are so important for our future.

---

### Dimensional Criticality at Grokking Across MLPs and Transformers

Authors: Ping Wang
Published: 2026-04-06
Categories: cs.LG, cond-mat.dis-nn, cs.AI, nlin.AO
arXiv: https://arxiv.org/abs/2604.16431v1

Abrupt transitions between distinct dynamical regimes are a hallmark of complex systems. Grokking in deep neural networks provides a striking example -- an abrupt transition from memorization to generalization long after training accuracy saturates -- yet robust macroscopic signatures of this transition remain elusive. Here we introduce \textbf{TDU--OFC} (Thresholded Diffusion Update--Olami-Feder-Christensen), an offline avalanche probe that converts gradient snapshots into cascade statistics and extracts a \emph{macroscopic observable} -- the time-resolved effective cascade dimension $D(t)$ -- via grokking-aligned finite-size scaling. Across Transformers trained on modular addition and MLPs trained on XOR, we discover a localized dynamical crossing of the Gaussian diffusion baseline $D=1$ precisely at the generalization transition. The crossing direction is task-dependent: modular addition descends through $D=1$ (approaching from $D>1$), while XOR ascends (from $D<1$). This opposite-direction convergence is consistent with attraction toward a candidate shared critical manifold, rather than trivial residence near $D \approx 1$. Negative controls confirm this picture: ungrokked runs remain supercritical ($D>1$) and never enter the post-transition regime. In addition, avalanche distributions exhibit heavy tails and finite-size scaling consistent with the dimensional exponent extracted from $D(t)$. Shadow-probe controls ($α_{\mathrm{train}}=0$) confirm that $D(t)$ is non-invasive, and grokked trajectories diverge from ungrokked ones in $D(t)$ some $100$--$200$ epochs before the behavioral transition.

---

### HalluSAE: Detecting Hallucinations in Large Language Models via Sparse Auto-Encoders

Authors: Boshui Chen, Zhaoxin Fan, Ke Wang, Zhiying Leng, Faguo Wu, Hongwei Zheng, Yifan Sun, Wenjun Wu
Published: 2026-04-06
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2604.16430v1

Large Language Models (LLMs) are powerful and widely adopted, but their practical impact is limited by the well-known hallucination phenomenon. While recent hallucination detection methods have made notable progress, we find most of them overlook the dynamic nature and underlying mechanisms of it. To address this gap, we propose HalluSAE, a phase transition-inspired framework that models hallucination as a critical shift in the model's latent dynamics. By modeling the generation process as a trajectory through a potential energy landscape, HalluSAE identifies critical transition zones and attributes factual errors to specific high-energy sparse features. Our approach consists of three stages: (1) Potential Energy Empowered Phase Zone Localization via sparse autoencoders and a geometric potential energy metric; (2) Hallucination-related Sparse Feature Attribution using contrastive logit attribution; and (3) Probing-based Causal Hallucination Detection through linear probes on disentangled features. Extensive experiments on Gemma-2-9B demonstrate that HalluSAE achieves state-of-the-art hallucination detection performance.

---

### (Sparse) Attention to the Details: Preserving Spectral Fidelity in ML-based Weather Forecasting Models

Authors: Maksim Zhdanov, Ana Lucic, Max Welling, Jan-Willem van de Meent
Published: 2026-04-06
Categories: cs.LG, cs.AI, cs.CV, physics.ao-ph
arXiv: https://arxiv.org/abs/2604.16429v3

We introduce Mosaic, a probabilistic weather forecasting model that addresses three failure modes of spectral degradation in ML-based weather prediction: spectral damping (statistical), high-frequency aliasing (architectural), and residual high-frequency leakage (parametric). Mosaic generates ensemble members through learned functional perturbations and operates on native-resolution grids via mesh-aligned block-sparse attention, a hardware-aligned mechanism that captures long-range dependencies at linear cost by sharing keys and values across spatially adjacent queries. At 1.5° resolution with 214M parameters, Mosaic matches or outperforms models trained on 6$\times$ finer resolution on key variables and achieves state-of-the-art results among 1.5° models, producing well-calibrated ensembles whose individual members exhibit near-perfect spectral alignment across all resolved frequencies. A 24-member, 10-day forecast takes under 12s on a single H100~GPU. Code is available at https://github.com/maxxxzdn/mosaic.

---

### Non-Stationarity in the Embedding Space of Time Series Foundation Models

Authors: Jinmyeong Choi, Brad Shook, Artur Dubrawski
Published: 2026-04-06
Categories: cs.LG, cs.AI, stat.ML
arXiv: https://arxiv.org/abs/2604.16428v1

Time series foundation models (TSFMs) are widely used as generic feature extractors, yet the notion of non-stationarity in their embedding spaces remains poorly understood. Recent work often conflates non-stationarity with distribution shift, blurring distinctions fundamental to classical time-series analysis and long-standing methodologies such as statistical process control (SPC). In SPC, non-stationarity signals a process leaving a stable regime - via shifts in mean, variance, or emerging trends - and detecting such departures is central to quality monitoring and change-point analysis. Motivated by this diagnostic tradition, we study how different forms of distributional non-stationarity - mean shifts, variance changes, and linear trends - become linearly accessible in TSFM embedding spaces under controlled conditions. We further examine temporal non-stationarity arising from persistence, which reflects violations of weak stationarity due to long-memory or near-unit-root behavior rather than explicit distributional shifts. By sweeping shift strength and probing multiple TSFMs, we find that embedding-space detectability of non-stationarity degrades smoothly and that different models exhibit distinct, model-specific failure modes.

---

### A methodology to rank importance of frequencies and channels in electromyography data with Decision Tree classifiers

Authors: Albert A. Nasybullin, Nursultan Abdullaev, Maksim A. Baranov, Viacheslav V. Koshman, Vitaly A. Mahonin
Published: 2026-04-06
Categories: eess.SP, cs.LG
arXiv: https://arxiv.org/abs/2604.15353v1

This study presents a methodology for identifying the most informative frequencies and channels in electromyography (EMG) data to evaluate muscle recovery using Decision Tree classifiers. EMG signals, recorded from the vastus lateralis muscle during squat exercises, were analyzed across varying rest intervals to assess optimal recovery periods. By employing single Decision Tree classifiers, the study enhances interpretability, offering insights into feature importance - essential for applications in medical and sports settings where transparency is critical. The experimental protocol utilized a grid search for hyperparameter tuning and cross-validation to address class imbalance, ultimately achieving a reliable classification of rest intervals based on power spectral density features. The results indicate that a limited subset of highly informative features provides sufficient accuracy, suggesting that streamlined, interpretable models are effective for the evaluation of muscle recovery. This approach can guide future research in developing compact, robust models adapted to EMG-based diagnostics.

---

### Ollivier-Ricci Curvature of Riemannian Manifolds and Directed Graphs with Applications to Graph Neural Networks

Authors: Eleanor P Wiesler
Published: 2026-04-06
Categories: math.DG, cs.AI, cs.SI, math.CO
arXiv: https://arxiv.org/abs/2604.14211v2

This thesis is an exposition of Ollivier-Ricci Curvature of metric spaces as introduced by Yann Ollivier, which is based upon the 1-Wasserstein Distance and optimal transport theory. We present some of the major results and proofs that connect Ollivier-Ricci curvature with classical Ricci curvature of Riemannian manifolds, including extensions of various theoretical bounds and theorems such as Bonnet-Myers and Levy-Gromov. Then we shift to results introduced by Lin-Lu-Yau on an extension of Ollivier-Ricci curvature on graphs, as well as the work of Jost-Liu on proving various combinatorial bounds for graph Ollivier-Ricci curvature. At the end of this thesis we present novel ideas and proofs regarding extensions of these results to directed graphs, and finally applications of graph-based Ollivier-Ricci curvature to various algorithms in network science and graph machine learning.

---

### Ollivier-Ricci Curvature of Riemannian Manifolds and Directed Graphs with Applications to Graph Neural Networks

Authors: Eleanor Wiesler
Published: 2026-04-06
Categories: math.DG, cs.AI, cs.SI, math.CO
arXiv: https://arxiv.org/abs/2604.14211v1

This thesis is an exposition of Ollivier-Ricci Curvature of metric spaces as introduced by Yann Ollivier, which is based upon the 1-Wasserstein Distance and optimal transport theory. We present some of the major results and proofs that connect Ollivier-Ricci curvature with classical Ricci curvature of Riemannian manifolds, including extensions of various theoretical bounds and theorems such as Bonnet-Myers and Levy-Gromov. Then we shift to results introduced by Lin-Lu-Yau on an extension of Ollivier-Ricci curvature on graphs, as well as the work of Jost-Liu on proving various combinatorial bounds for graph Ollivier-Ricci curvature. At the end of this thesis we present novel ideas and proofs regarding extensions of these results to directed graphs, and finally applications of graph-based Ollivier-Ricci curvature to various algorithms in network science and graph machine learning.

---

### Chinese Language Is Not More Efficient Than English in Vibe Coding: A Preliminary Study on Token Cost and Problem-Solving Rate

Authors: Simiao Ren, Xingyu Shen, Yuchen Zhou, Dennis, Ng, Ankit Raj
Published: 2026-04-06
Categories: cs.CL, cs.SE
arXiv: https://arxiv.org/abs/2604.14210v1

A claim has been circulating on social media and practitioner forums that Chinese prompts are more token-efficient than English for LLM coding tasks, potentially reducing costs by up to 40\%. This claim has influenced developers to consider switching to Chinese for ``vibe coding'' to save on API costs. In this paper, we conduct a rigorous empirical study using SWE-bench Lite, a benchmark of software engineering tasks, to evaluate whether this claim of Chinese token efficiency holds up to scrutiny. Our results reveal three key findings: First, the efficiency advantage of Chinese is not observed. Second, token cost varies by model architecture in ways that defy simple assumptions: while MiniMax-2.7 shows 1.28x higher token costs for Chinese, GLM-5 actually consumes fewer tokens with Chinese prompts. Third, and most importantly, we found that the success rate when prompting in Chinese is generally lower than in English across all models we tested. We also measure cost efficiency as expected cost per successful task -- jointly accounting for token consumption and task resolution rate. These findings should be interpreted as preliminary evidence rather than a definitive conclusion, given the limited number of models evaluated and the narrow set of benchmarks tested due to resource constraints; they indicate that language effects on token cost are model-dependent, and that practitioners should not expect cost savings or performance gains just by switching their prompt language to Chinese.

---

### Evaluating Scene-based In-Situ Item Labeling for Immersive Conversational Recommendation

Authors: Jiazhou Liang, Yifan Simon Liu, David Guo, Minqi Sun, Yilun Jiang, Scott Sanner
Published: 2026-04-06
Categories: cs.IR, cs.AI
arXiv: https://arxiv.org/abs/2604.09698v2

The growing ubiquity of Extended Reality (XR) is driving Conversational Recommendation Systems (CRS) toward visually immersive experiences. We formalize this paradigm as Immersive CRS (ICRS), where recommended items are highlighted directly in the user's scene-based visual environment and augmented with in-situ labels. While item recommendation has been widely studied, the problem of how to select and evaluate which information to present as immersive labels remains an open problem. To this end, we introduce a principled categorization of information needs into explicit intent satisfaction and proactive information needs and use these to define novel evaluation metrics for item label selection. We benchmark IR-, LLM-, and VLM-based methods across three datasets and ICRS scenarios: fashion, movie recommendation, and retail shopping. Our evaluation reveals three important limitations of existing methods: (1) they fail to leverage scenario-specific information modalities (e.g., visual cues for fashion, meta-data for retail), (2) they present redundant information that is visually inferable, and (3) they poorly anticipate users' proactive information needs from explicit dialogue alone. In summary, this work provides both a novel evaluation paradigm for in-situ item labeling in ICRS and highlights key challenges for future work.

---

### I Can't Believe TTA Is Not Better: When Test-Time Augmentation Hurts Medical Image Classification

Authors: Daniel Nobrega Medeiros
Published: 2026-04-06
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2604.09697v1

Test-time augmentation (TTA)--aggregating predictions over multiple augmented copies of a test input--is widely assumed to improve classification accuracy, particularly in medical imaging where it is routinely deployed in production systems and competition solutions. We present a systematic empirical study challenging this assumption across three MedMNIST v2 benchmarks and four architectures spanning three orders of magnitude in parameter count (21K to 11M). Our principal finding is that TTA with standard augmentation pipelines consistently degrades accuracy relative to single-pass inference, with drops as severe as 31.6 percentage points for ResNet-18 on pathology images. This degradation affects all architectures, including convolutional models, and worsens with more augmented views. The sole exception is ResNet-18 on dermatology images, which gains a modest +1.6%. We identify the distribution shift between augmented and training-time inputs--amplified by batch normalization statistics mismatch--as the primary mechanism. Our ablation studies show that augmentation strategy matters critically: intensity-only augmentations preserve more performance than geometric transforms, and including the original unaugmented image partially mitigates but does not eliminate the accuracy drop. These findings serve as a cautionary note for practitioners: TTA should not be applied as a default post-hoc improvement but must be validated on the specific model-dataset combination.

---