# Research Papers — 2026-04 · bundle page 159 of 233

> Full-text markdown of papers #7901–7950 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/159
[← Previous bundle](https://aipod.dev/papers/archive/2026-04/page/158.md)
[Next bundle →](https://aipod.dev/papers/archive/2026-04/page/160.md)

### An Auditable Policy-Simulation Framework for Student Dropout in Intervention-Free Data

Authors: Rafael da Silva, Jeff Eicher, Gregory Longo
Published: 2026-04-10
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2604.08874v3

This study proposes a temporal modeling framework with a counterfactual policy-simulation layer for student dropout in higher education, using LMS engagement data and administrative withdrawal records. Dropout is operationalized as a time-to-event outcome at the enrollment level; weekly risk is modeled in discrete time via penalized, class-balanced logistic regression over person--period rows. Under a late-event temporal holdout, the model attains row-level AUCs of 0.8350 (train) and 0.8405 (test), with aggregate calibration acceptable but sparsely supported in the highest-risk bins. Ablation analyses indicate performance is sensitive to feature set composition, underscoring the role of temporal engagement signals. A scenario-indexed policy layer produces survival contrasts $ΔS(T)$ under an explicit trigger/schedule contract: positive contrasts are confined to the shock branch ($T_{\rm policy}=18$: 0.0102, 0.0260, 0.0819), while the mechanism-aware branch is negative ($ΔS_{\rm mech}(18)=-0.0078$, $ΔS_{\rm mech}(38)=-0.0134$). A subgroup analysis by gender quantifies scenario-induced survival gaps via bootstrap; contrasts are directionally stable but small. Results are not causally identified; they demonstrate the framework's capacity for internal structural scenario comparison under observational data constraints.

---

### How does Chain of Thought decompose complex tasks?

Authors: Amrut Nadgir, Vijay Balasubramanian, Pratik Chaudhari
Published: 2026-04-10
Categories: cs.LG, cond-mat.dis-nn, cond-mat.stat-mech
arXiv: https://arxiv.org/abs/2604.08872v2

Many language tasks can be modeled as classification problems where a large language model (LLM) is given a prompt and selects one among many possible answers. We show that the classification error in such problems scales as a power law in the number of classes. This has a dramatic consequence: the prediction error can be reduced substantially by splitting the overall task into a sequence of smaller classification problems, each with the same number of classes ("degree"). This tree-structured decomposition models chain-of-thought (CoT). It has been observed that CoT-based predictors perform better when they "think", i.e., when they develop a deeper tree, thus decomposing the problem into a larger number of steps. We identify a critical threshold for the degree, below which thinking is detrimental, and above which there exists an optimal depth that minimizes the error. It is impossible to surpass this minimal error by increasing the depth of thinking.

---

### A Unified Survival Benchmark for Temporal Dropout Risk Prediction in Learning Analytics

Authors: Rafael da Silva, Jeff Eicher, Gregory Longo
Published: 2026-04-10
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2604.08870v3

Student dropout is a persistent concern in Learning Analytics, yet comparative studies frequently evaluate predictive models under heterogeneous protocols, prioritizing discrimination over temporal interpretability and calibration. This study introduces a survival-oriented benchmark for temporal dropout risk modelling using the Open University Learning Analytics Dataset (OULAD). Two arms are compared: Family A: Dynamic Weekly, with models in person-period representation, and Family B: Static Early-Window, with an expanded roster of families: tree-based survival, parametric, and neural models. The evaluation protocol integrates four analytical layers: predictive performance, ablation, explainability, and calibration. Results are reported within each family separately, because a single numerical cross-family ranking would conflate genuine model differences with artifacts of temporal representation, to which survival metrics are known to be sensitive. Within Family B, Random Survival Forest showed the highest point estimates for time-dependent concordance and the lowest Brier scores across all three horizons; within Family A, Poisson Piecewise-Exponential showed the lowest point estimate for integrated Brier score within a tight five-model cluster. No-refit bootstrap resampling qualifies these positions as directional signals, not claims of strict superiority. Ablation and explainability analyses converged, across all models, on a shared finding: the dominant predictive signal was not primarily demographic or structural, but temporal and behavioral. Calibration corroborated this pattern in the better-discriminating models, except for XGBoost AFT, the sole outlier (analyzed in the Discussion). These results support unified, multi-dimensional benchmarking in Learning Analytics and situate dropout risk as a temporal-behavioral process rather than a function of static background attributes.

---

### MedFormer-UR: Uncertainty-Routed Transformer for Medical Image Classification

Authors: Mohammed Maaz Sibhai, Abedalrhman Alkhateeb, Saad B. Ahmed
Published: 2026-04-10
Categories: eess.IV, cs.AI, cs.CV, cs.LG
arXiv: https://arxiv.org/abs/2604.08868v1

To ensure safe clinical integration, deep learning models must provide more than just high accuracy; they require dependable uncertainty quantification. While current Medical Vision Transformers perform well, they frequently struggle with overconfident predictions and a lack of transparency, issues that are magnified by the noisy and imbalanced nature of clinical data. To address this, we enhanced the modified Medical Transformer (MedFormer) that incorporates prototype-based learning and uncertainty-guided routing, by utilizing a Dirichlet distribution for per-token evidential uncertainty, our framework can quantify and localize ambiguity in real-time. This uncertainty is not just an output but an active participant in the training process, filtering out unreliable feature updates. Furthermore, the use of class-specific prototypes ensures the embedding space remains structured, allowing for decisions based on visual similarity. Testing across four modalities (mammography, ultrasound, MRI, and histopathology) confirms that our approach significantly enhances model calibration, reducing expected calibration error (ECE) by up to 35%, and improves selective prediction, even when accuracy gains are modest.

---

### AudioGuard: Toward Comprehensive Audio Safety Protection Across Diverse Threat Models

Authors: Mintong Kang, Chen Fang, Bo Li
Published: 2026-04-10
Categories: cs.SD, cs.AI
arXiv: https://arxiv.org/abs/2604.08867v1

Audio has rapidly become a primary interface for foundation models, powering real-time voice assistants. Ensuring safety in audio systems is inherently more complex than just "unsafe text spoken aloud": real-world risks can hinge on audio-native harmful sound events, speaker attributes (e.g., child voice), impersonation/voice-cloning misuse, and voice-content compositional harms, such as child voice plus sexual content. The nature of audio makes it challenging to develop comprehensive benchmarks or guardrails against this unique risk landscape. To close this gap, we conduct large-scale red teaming on audio systems, systematically uncover vulnerabilities in audio, and develop a comprehensive, policy-grounded audio risk taxonomy and AudioSafetyBench, the first policy-based audio safety benchmark across diverse threat models. AudioSafetyBench supports diverse languages, suspicious voices (e.g., celebrity/impersonation and child voice), risky voice-content combinations, and non-speech sound events. To defend against these threats, we propose AudioGuard, a unified guardrail consisting of 1) SoundGuard for waveform-level audio-native detection and 2) ContentGuard for policy-grounded semantic protection. Extensive experiments on AudioSafetyBench and four complementary benchmarks show that AudioGuard consistently improves guardrail accuracy over strong audio-LLM-based baselines with substantially lower latency.

---

### AI-Induced Human Responsibility (AIHR) in AI-Human teams

Authors: Greg Nyilasy, Brock Bastian, Jennifer Overbeck, Abraham Ryan Ade Putra Hito
Published: 2026-04-10
Categories: cs.HC, cs.AI
arXiv: https://arxiv.org/abs/2604.08866v1

As organizations increasingly deploy AI as a teammate rather than a standalone tool, morally consequential mistakes often arise from joint human-AI workflows in which causality is ambiguous. We ask how people allocate responsibility in these hybrid-agent settings. Across four experiments (N = 1,801) in an AI-assisted lending context (e.g., discriminatory rejection, irresponsible lending, and low-harm filing errors), participants consistently attributed more responsibility to the human decision maker when the human was paired with AI than when paired with another human (by an average of 10 points on a 0-100 scale across studies). This AI-Induced Human Responsibility (AIHR) effect held across high and low harm scenarios and persisted even where self-serving blame-shifting (when the human in question was the self) would be expected. Process evidence indicates that AIHR is explained by inferences of agent autonomy: AI is seen as a constrained implementer, which makes the human the default locus of discretionary responsibility. Alternative mechanisms (mind perception; self-threat) did not account for the effect. These findings extend research on algorithm aversion, hybrid AI-human organizational behavior and responsibility gaps in technology by showing that AI-human teaming can increase (rather than dilute) human responsibility, with implications for accountability design in AI-enabled organizations.

---

### SPPO: Sequence-Level PPO for Long-Horizon Reasoning Tasks

Authors: Tianyi Wang, Yixia Li, Long Li, Yibiao Chen, Shaohan Huang, Yun Chen, Peng Li, Yang Liu, Guanhua Chen
Published: 2026-04-10
Categories: cs.AI
arXiv: https://arxiv.org/abs/2604.08865v1

Proximal Policy Optimization (PPO) is central to aligning Large Language Models (LLMs) in reasoning tasks with verifiable rewards. However, standard token-level PPO struggles in this setting due to the instability of temporal credit assignment over long Chain-of-Thought (CoT) horizons and the prohibitive memory cost of the value model. While critic-free alternatives like GRPO mitigate these issues, they incur significant computational overhead by requiring multiple samples for baseline estimation, severely limiting training throughput. In this paper, we introduce Sequence-Level PPO (SPPO), a scalable algorithm that harmonizes the sample efficiency of PPO with the stability of outcome-based updates. SPPO reformulates the reasoning process as a Sequence-Level Contextual Bandit problem, employing a decoupled scalar value function to derive low-variance advantage signals without multi-sampling. Extensive experiments on mathematical benchmarks demonstrate that SPPO significantly surpasses standard PPO and matches the performance of computation-heavy group-based methods, offering a resource-efficient framework for aligning reasoning LLMs.

---

### Hidden in Plain Sight: Visual-to-Symbolic Analytical Solution Inference from Field Visualizations

Authors: Pengze Li, Jiaquan Zhang, Yunbo Long, Xinping Liu, Zhou wenjie, Encheng Su, Zihang Zeng, Jiaqi Liu, Jiyao Liu, Junchi Yu, Lihao Liu, Philip Torr et al.
Published: 2026-04-10
Categories: cs.AI
arXiv: https://arxiv.org/abs/2604.08863v1

Recovering analytical solutions of physical fields from visual observations is a fundamental yet underexplored capability for AI-assisted scientific reasoning. We study visual-to-symbolic analytical solution inference (ViSA) for two-dimensional linear steady-state fields: given field visualizations (and first-order derivatives) plus minimal auxiliary metadata, the model must output a single executable SymPy expression with fully instantiated numeric constants. We introduce ViSA-R2 and align it with a self-verifying, solution-centric chain-of-thought pipeline that follows a physicist-like pathway: structural pattern recognition solution-family (ansatz) hypothesis parameter derivation consistency verification. We also release ViSA-Bench, a VLM-ready synthetic benchmark covering 30 linear steady-state scenarios with verifiable analytical/symbolic annotations, and evaluate predictions by numerical accuracy, expression-structure similarity, and character-level accuracy. Using an 8B open-weight Qwen3-VL backbone, ViSA-R2 outperforms strong open-source baselines and the evaluated closed-source frontier VLMs under a standardized protocol.

---

### BIAS: A Biologically Inspired Algorithm for Video Saliency Detection

Authors: Zhao-ji Zhang, Ya-tang Li
Published: 2026-04-10
Categories: cs.CV
arXiv: https://arxiv.org/abs/2604.08858v1

We present BIAS, a fast, biologically inspired model for dynamic visual saliency detection in continuous video streams. Building on the Itti--Koch framework, BIAS incorporates a retina-inspired motion detector to extract temporal features, enabling the generation of saliency maps that integrate both static and motion information. Foci of attention (FOAs) are identified using a greedy multi-Gaussian peak-fitting algorithm that balances winner-take-all competition with information maximization. BIAS detects salient regions with millisecond-scale latency and outperforms heuristic-based approaches and several deep-learning models on the DHF1K dataset, particularly in videos dominated by bottom-up attention. Applied to traffic accident analysis, BIAS demonstrates strong real-world utility, achieving state-of-the-art performance in cause-effect recognition and anticipating accidents up to 0.72 seconds before manual annotation with reliable accuracy. Overall, BIAS bridges biological plausibility and computational efficiency to achieve interpretable, high-speed dynamic saliency detection.

---

### Cross-Lingual Attention Distillation with Personality-Informed Generative Augmentation for Multilingual Personality Recognition

Authors: Jing Jie Tan, Ban-Hoe Kwan, Danny Wee-Kiat Ng, Yan-Chai Hum, Noriyuki Kawarazaki, Kosuke Takano
Published: 2026-04-10
Categories: cs.CL
arXiv: https://arxiv.org/abs/2604.08851v1

While significant work has been done on personality recognition, the lack of multilingual datasets remains an unresolved challenge. To address this, we propose ADAM (Cross-Lingual (A)ttention (D)istillation with Personality-Guided Generative (A)ugmentation for (M)ultilingual Personality Recognition), a state-of-the-art approach designed to advance multilingual personality recognition. Our approach leverages an existing English-language personality dataset as the primary source and employs a large language model (LLM) for translationbased augmentation, enhanced by Personality-Informed Generative Augmentation (PIGA), to generate high-quality training data in multiple languages, including Japanese, Chinese, Malay, and French. We provide a thorough analysis to justify the effectiveness of these augmentation techniques. Building on these advancements, ADAM integrates Cross-Lingual Attention Distillation (CLAD) to train a model capable of understanding and recognizing personality traits across languages, bridging linguistic and cultural gaps in personality analysis. This research presents a thorough evaluation of the proposed augmentation method, incorporating an ablation study on recognition performance to ensure fair comparisons and robust validation. Overall, with PIGA augmentation, the findings demonstrate that CLAD significantly outperforms the standard BCE across all languages and personality traits, achieving notable improvements in average BA scores - 0.6332 (+0.0573) on the Essays dataset and 0.7448 (+0.0968) on the Kaggle dataset. The CLAD-trained model also demonstrated strong generalizability and achieved benchmark performance comparable to current leading encoder models. The model weight, dataset, and algorithm repository are available at https://research.jingjietan.com/?q=ADAM.

---

### Finite-Sample Analysis of Nonlinear Independent Component Analysis:Sample Complexity and Identifiability Bounds

Authors: Yuwen Jiang
Published: 2026-04-10
Categories: cs.LG
arXiv: https://arxiv.org/abs/2604.08850v1

Independent Component Analysis (ICA) is a fundamental unsupervised learning technique foruncovering latent structure in data by separating mixed signals into their independent sources. While substantial progress has been made in establishing asymptotic identifiability guarantees for nonlinear ICA, the finite-sample statistical properties of learning algorithms remain poorly understood. This gap poses significant challenges for practitioners who must determine appropriate sample sizes for reliable source recovery. This paper presents a comprehensive finite-sample analysis of nonlinear ICA with neural network encoders, providing the first complete characterization with matching upper and lower bounds. Our theoretical development introduces three key technical contributions. First, we establish a direct relationship between excess risk and identification error that bypasses parameter-space arguments, thereby avoiding the rate degradation that would otherwise yield suboptimal scaling. Second, we prove matching information-theoretic lower bounds that confirm the optimality of our sample complexity results. Third, we extend our analysis to practical SGD optimization, showing that the same sample efficiency can be achieved with finite-iteration gradient descent under standard landscape assumptions. We validate our theoretical predictions through carefully designed simulation experiments. This gap points toward valuable future research on finite-sample behavior of neural network training and highlights the importance of our validated scaling laws for dimension and diversity.

---

### SatIR: Scalable High-Recall Constraint-Satisfaction-Based Information Retrieval for Clinical Trials Matching

Authors: Zikai Zhou, Yufei Jin, Yilin Xu, Yu-Chiang Wang, Chieh-Ju Chao, Monica S. Lam
Published: 2026-04-10
Categories: cs.CL, cs.AI, cs.DB, cs.MA, cs.SC
arXiv: https://arxiv.org/abs/2604.08849v4

Many real-world retrieval and matching problems require more than topical relevance: a candidate must satisfy the specific constraints of one profile among many, not just be relevant to it. Clinical trials are a high-stakes instance of this challenge: they are central to evidence-based medicine, yet many struggle to meet enrollment targets, despite the availability of over half a million trials listed on ClinicalTrials.gov, which attracts approximately two million users monthly. Existing retrieval techniques, largely based on keyword and embedding-similarity matching, treat eligibility constraints as soft signals rather than binding requirements, resulting in low recall, low precision, and limited interpretability. We propose SatIR, a scalable, efficient, high-precision, high-recall, interpretable clinical trial retrieval method based on formal constraint satisfaction. Leveraging established medical ontologies, we use Large Language Models (LLMs) to convert informal reasoning -- regarding ambiguity, implicit clinical assumptions, and incomplete patient records -- into explicit, precise, controllable, and interpretable formal Satisfiability Modulo Theories (SMT) constraints. For scalable and efficient retrieval, we project the SMT matching problem onto relational algebra, enabling an efficient database implementation that retains high recall while sacrificing little precision. SatIR consistently improves eligibility-aware retrieval over similarity-based baselines on the SIGIR 2016 dataset and a benchmark derived from TREC 2022. Relative to TrialGPT-style retrieval, SatIR retrieves 32%-72% more relevant-and-eligible trials per patient on SIGIR 2016 and achieves 1.8-3.2x higher eligible-trial recall on the TREC benchmark. Retrieval is fast, requiring only 146 milliseconds per patient over 3,621 SIGIR trials.

---

### SatIR: Scalable High-Recall Constraint-Satisfaction-Based Information Retrieval for Clinical Trials Matching

Authors: Cyrus Zhou, Yufei Jin, Yilin Xu, Yu-Chiang Wang, Chieh-Ju Chao, Monica S. Lam
Published: 2026-04-10
Categories: cs.CL, cs.AI, cs.DB, cs.MA, cs.SC
arXiv: https://arxiv.org/abs/2604.08849v2

Many important retrieval problems are not merely problems of semantic similarity, but problems of constraint satisfaction: a retrieved item should be topically relevant to a query and satisfy explicit requirements involving negation, temporal conditions, numeric thresholds, exceptions, ontological relations, and incomplete evidence. We study this challenge in clinical trial matching, a high-stakes test bed where a useful trial must both address a patient's medical needs and satisfy complex eligibility criteria. We propose SatIR, a scalable constraint-based retrieval method for clinical trial matching. SatIR converts trial eligibility criteria and summaries into formal constraints, then retrieves patient--trial pairs by executing these constraints over a database. The system combines Satisfiability Modulo Theories (SMT), relational algebra, medical ontology grounding, and large language models (LLMs): formal methods provide executable and inspectable matching, while LLMs convert ambiguous, incomplete, and implicit clinical information into explicit, controllable constraint representations. Across the SIGIR 2016 patient--trial collection and TREC-2022-RetrievalSubset, a benchmark derived from TREC 2022, SATIR consistently improves eligibility-aware retrieval over similarity-based baselines. Relative to TrialGPT-style retrieval, SATIR retrieves 32%--72% more relevant-and-eligible trials per patient on SIGIR 2016 and achieves $1.8$--$3.2\times$ higher eligible-trial recall on TREC-2022-RetrievalSubset. Retrieval is fast, requiring only 146 milliseconds per patient over 3,621 SIGIR trials.

---

### DeFakeQ: Enabling Real-Time Deepfake Detection on Edge Devices via Adaptive Bidirectional Quantization

Authors: Xiangyu Li, Yujing Sun, Yuhang Zheng, Yuexin Ma, Kwok-Yan Lam
Published: 2026-04-10
Categories: cs.CV
arXiv: https://arxiv.org/abs/2604.08847v1

Deepfake detection has become a fundamental component of modern media forensics. Despite significant progress in detection accuracy, most existing methods remain computationally intensive and parameter-heavy, limiting their deployment on resource-constrained edge devices that require real-time, on-site inference. This limitation is particularly critical in an era where mobile devices are extensively used for media-centric applications, including online payments, virtual meetings, and social networking. Meanwhile, due to the unique requirement of capturing extremely subtle forgery artifacts for deepfake detection, state-of-the-art quantization techniques usually underperform for such a challenging task. These fine-grained cues are highly sensitive to model compression and can be easily degraded during quantization, leading to noticeable performance drops. This challenge highlights the need for quantization strategies specifically designed to preserve the discriminative features essential for reliable deepfake detection. To address this gap, we propose DefakeQ, the first quantization framework tailored for deepfake detectors, enabling real-time deployment on edge devices. Our approach introduces a novel adaptive bidirectional compression strategy that simultaneously leverages feature correlations and eliminates redundancy, achieving an effective balance between model compactness and detection performance. Extensive experiments across five benchmark datasets and eleven state-of-the-art backbone detectors demonstrate that DeFakeQ consistently surpasses existing quantization and model compression baselines. Furthermore, we deploy DefakeQ on mobile devices in real-world scenarios, demonstrating its capability for real-time deepfake detection and its practical applicability in edge environments.

---

### Dictionary-Aligned Concept Control for Safeguarding Multimodal LLMs

Authors: Jinqi Luo, Jinyu Yang, Tal Neiman, Lei Fan, Bing Yin, Son Tran, Mubarak Shah, René Vidal
Published: 2026-04-10
Categories: cs.LG, cs.AI, cs.CL, cs.CV
arXiv: https://arxiv.org/abs/2604.08846v1

Multimodal Large Language Models (MLLMs) have been shown to be vulnerable to malicious queries that can elicit unsafe responses. Recent work uses prompt engineering, response classification, or finetuning to improve MLLM safety. Nevertheless, such approaches are often ineffective against evolving malicious patterns, may require rerunning the query, or demand heavy computational resources. Steering the activations of a frozen model at inference time has recently emerged as a flexible and effective solution. However, existing steering methods for MLLMs typically handle only a narrow set of safety-related concepts or struggle to adjust specific concepts without affecting others. To address these challenges, we introduce Dictionary-Aligned Concept Control (DACO), a framework that utilizes a curated concept dictionary and a Sparse Autoencoder (SAE) to provide granular control over MLLM activations. First, we curate a dictionary of 15,000 multimodal concepts by retrieving over 400,000 caption-image stimuli and summarizing their activations into concept directions. We name the dataset DACO-400K. Second, we show that the curated dictionary can be used to intervene activations via sparse coding. Third, we propose a new steering approach that uses our dictionary to initialize the training of an SAE and automatically annotate the semantics of the SAE atoms for safeguarding MLLMs. Experiments on multiple MLLMs (e.g., QwenVL, LLaVA, InternVL) across safety benchmarks (e.g., MM-SafetyBench, JailBreakV) show that DACO significantly improves MLLM safety while maintaining general-purpose capabilities.

---

### Spectral Geometry of LoRA Adapters Encodes Training Objective and Predicts Harmful Compliance

Authors: Roi Paul
Published: 2026-04-10
Categories: cs.LG
arXiv: https://arxiv.org/abs/2604.08844v1

We study whether low-rank spectral summaries of LoRA weight deltas can identify which fine-tuning objective was applied to a language model, and whether that geometric signal predicts downstream behavioral harm. In a pre-registered experiment on \texttt{Llama-3.2-3B-Instruct}, we manufacture 38 LoRA adapters across four categories: healthy SFT baselines, DPO on inverted harmlessness preferences, DPO on inverted helpfulness preferences, and activation-steering-derived adapters, and extract per-layer spectral features (norms, stable rank, singular-value entropy, effective rank, and singular-vector cosine alignment to a healthy centroid). Within a single training method (DPO), a logistic regression classifier achieves AUC~1.00 on binary drift detection, all six pairwise objective comparisons, and near-perfect ordinal severity ranking ($ρ\geq 0.956$). Principal component analysis on flattened weight deltas reveals that training objective is PC1 (AUC~1.00 for objective separation), orthogonal to training duration on PC2. Query-projection weights detect that drift occurred; value-projection weights identify which objective. Cross-method generalization fails completely: a DPO-trained classifier assigns every steering adapter a lower drift score than every DPO adapter (AUC~0.00). In a behavioral evaluation phase, DPO-inverted-harmlessness adapters show elevated harmful compliance on HEx-PHI prompts (mean ASR 0.266 vs.\ healthy 0.112, $Δ= +0.154$), with near-perfect dose--response ($ρ= 0.986$). The geometry-to-behavior rank correlation is $ρ= 0.72$ across 24 non-steered adapters. These results establish that within a controlled manufacturing regime, LoRA weight-space geometry carries objective identity, intensity ordering, and a coarse link to harmful compliance, and that cross-method monitoring requires per-method calibration.

---

### Discrete Meanflow Training Curriculum

Authors: Chia-Hong Hsu, Frank Wood
Published: 2026-04-10
Categories: cs.LG
arXiv: https://arxiv.org/abs/2604.08837v1

Flow-based image generative models exhibit stable training and produce high quality samples when using multi-step sampling procedures. One-step generative models can produce high quality image samples but can be difficult to optimize as they often exhibit unstable training dynamics. Meanflow models exhibit excellent few-step sampling performance and tantalizing one-step sampling performance. Notably, MeanFlow models that achieve this have required extremely large training budgets. We significantly decrease the amount of computation and data budget it takes to train Meanflow models by noting and exploiting a particular discretization of the Meanflow objective that yields a consistency property which we formulate into a ``Discrete Meanflow'' (DMF) Training Curriculum. Initialized with a pretrained Flow Model, DMF curriculum reaches one-step FID 3.36 on CIFAR-10 in only 2000 epochs. We anticipate that faster training curriculums of Meanflow models, specifically those fine-tuned from existing Flow Models, drives efficient training methods of future one-step examples.

---

### CatalogStitch: Dimension-Aware and Occlusion-Preserving Object Compositing for Catalog Image Generation

Authors: Sanyam Jain, Pragya Kandari, Manit Singhal, He Zhang, Soo Ye Kim
Published: 2026-04-10
Categories: cs.CV
arXiv: https://arxiv.org/abs/2604.08836v1

Generative object compositing methods have shown remarkable ability to seamlessly insert objects into scenes. However, when applied to real-world catalog image generation, these methods require tedious manual intervention: users must carefully adjust masks when product dimensions differ, and painstakingly restore occluded elements post-generation. We present CatalogStitch, a set of model-agnostic techniques that automate these corrections, enabling user-friendly content creation. Our dimension-aware mask computation algorithm automatically adapts the target region to accommodate products with different dimensions; users simply provide a product image and background, without manual mask adjustments. Our occlusion-aware hybrid restoration method guarantees pixel-perfect preservation of occluding elements, eliminating post-editing workflows. We additionally introduce CatalogStitch-Eval, a 58-example benchmark covering aspect-ratio mismatch and occlusion-heavy catalog scenarios, together with supplementary PDF and HTML viewers. We evaluate our techniques with three state-of-the-art compositing models (ObjectStitch, OmniPaint, and InsertAnything), demonstrating consistent improvements across diverse catalog scenarios. By reducing manual intervention and automating tedious corrections, our approach transforms generative compositing into a practical, human-friendly tool for production catalog workflows.

---

### Hierarchical Kernel Transformer: Multi-Scale Attention with an Information-Theoretic Approximation Analysis

Authors: Giansalvo Cirrincione
Published: 2026-04-10
Categories: cs.LG, cs.NE, stat.ML
arXiv: https://arxiv.org/abs/2604.08829v1

The Hierarchical Kernel Transformer (HKT) is a multi-scale attention mechanism that processes sequences at L resolution levels via trainable causal downsampling, combining level-specific score matrices through learned convex weights. The total computational cost is bounded by 4/3 times that of standard attention, reaching 1.3125x for L = 3. Four theoretical results are established. (i) The hierarchical score matrix defines a positive semidefinite kernel under a sufficient condition on the symmetrised bilinear form (Proposition 3.1). (ii) The asymmetric score matrix decomposes uniquely into a symmetric part controlling reciprocal attention and an antisymmetric part controlling directional attention; HKT provides L independent such pairs across scales, one per resolution level (Propositions 3.5-3.6). (iii) The approximation error decomposes into three interpretable components with an explicit non-Gaussian correction and a geometric decay bound in L (Theorem 4.3, Proposition 4.4). (iv) HKT strictly subsumes single-head standard attention and causal convolution (Proposition 3.4). Experiments over 3 random seeds show consistent gains over retrained standard attention baselines: +4.77pp on synthetic ListOps (55.10+-0.29% vs 50.33+-0.12%, T = 512), +1.44pp on sequential CIFAR-10 (35.45+-0.09% vs 34.01+-0.19%, T = 1,024), and +7.47pp on IMDB character-level sentiment (70.19+-0.57% vs 62.72+-0.40%, T = 1,024), all at 1.31x overhead.

---

### Post-Hoc Guidance for Consistency Models by Joint Flow Distribution Learning

Authors: Chia-Hong Hsu, Randall Balestriero
Published: 2026-04-10
Categories: cs.LG, cs.CV
arXiv: https://arxiv.org/abs/2604.08828v1

Classifier-free Guidance (CFG) lets practitioners trade-off fidelity against diversity in Diffusion Models (DMs). The practicality of CFG is however hindered by DMs sampling cost. On the other hand, Consistency Models (CMs) generate images in one or a few steps, but existing guidance methods require knowledge distillation from a separate DM teacher, limiting CFG to Consistency Distillation (CD) methods. We propose Joint Flow Distribution Learning (JFDL), a lightweight alignment method enabling guidance in a pre-trained CM. With a pre-trained CM as an ordinary differential equation (ODE) solver, we verify with normality tests that the variance-exploding noise implied by the velocity fields from unconditional and conditional distributions is Gaussian. In practice, JFDL equips CMs with the familiar adjustable guidance knob, yielding guided images with similar characteristics to CFG. Applied to an original Consistency Trained (CT) CM that could only do conditional sampling, JFDL unlocks guided generation and reduces FID on both CIFAR-10 and ImageNet 64x64 datasets. This is the first time that CMs are able to receive effective guidance post-hoc without a DM teacher, thus, bridging a key gap in current methods for CMs.

---

### Intelligent Three Level Learning Architecture for Autonomous UAV Swarms in Search and Rescue

Authors: Oleksii Bychkov
Published: 2026-04-09
Categories: cs.AI
arXiv: https://arxiv.org/abs/2607.14093v1

This paper presents a novel three level hierarchical learning architecture for autonomous UAV swarms performing search and rescue operations. Unlike conventional approaches that apply a single learning paradigm across all hierarchy levels, the proposed architecture integrates three qualitatively different learning mechanisms corresponding to the biological hierarchy of reflexes, skills, and reasoning such as Hebbian neuroplasticity for individual agent adaptation, multi agent reinforcement learning with graph neural networks and behavior trees for tactical coordination, and model agnostic meta learning with BDI reasoning and a digital twin for strategic decision making. The architecture is formalized through twenty two architectural contracts organized across six components such as BDI, Behavior Trees, GNN, MARL, Neuroplasticity, Meta Learning that collectively provide six classes of formal guarantees such as safety, budget correctness, optimality, liveness, starvation freedom, and inter level consistency. We introduce Swarm Meta Cognition as a compositional property arising from the structured interaction of all three levels, enabling the swarm to monitor its own cognitive state and switch between cognitive strategies. Five constructive progress functions for SAR task types bridge the gap between abstract optimization theory and concrete operational scenarios. The main integration theorem establishes that when all contracts are satisfied, the hybrid neuro-symbolic system preserves all six guarantee classes. For the dynamic case with active learning, five new contracts extend the framework with three additional guarantees such as cognitive resilience, graceful degradation, and monotonic meta improvement. Theoretical analysis demonstrates that the architecture addresses five fundamental limitations of existing hierarchical RL approaches.

---

### Interval Certifications for Multilayered Perceptrons via Lattice Traversal

Authors: Merkouris Papamichail, Konstantinos Varsos, Giorgos Flouris, João Marques-Silva
Published: 2026-04-09
Categories: cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2607.08773v1

In this work we present a rigorous theoretical framework to a foundational problem of AI safety, namely adversarial robustness. In particular, we show that the adversarial robustness problem can be reduced to a lattice traversal problem. Each element of this lattice corresponds to an interval, i.e., an axis-aligned hyper-rectangle, containing an input point $\mathbf{x}$. Consider a multilayered perceptron classifier (MLP). An interval $I$ constitutes a sound certification if $\mathbf{x} \in I$ and $\mathbf{x}$ can be freely perturbed in $I$ without changing the MLP's prediction. Complementarily, an interval $I$ constitutes a complete certification if $\mathbf{x} \in I$ and when $\mathbf{x}$ moves outside of $I$ the MLP's prediction is guaranteed to change. While the sound certification problem corresponds to the well-studied adversarial robustness, complete certifications have not been examined in the literature. We develop lattice traversal operators, which we apply in a refine & verify iterative scheme. Using formal MLP verifiers, sound maximality and complete minimality are guaranteed. Moreover, we examine objective optimization problems. There we discover some interesting asymmetries. For complete certifications, the minimum solution is obtained in polynomial oracle calls. This does not hold for sound certifications, where we prove strong intractability results. Additionally, we examine optimization problems in symmetric intervals (i.e., $\ell_\infty$-spheres), where we provide logarithmic algorithms. Finally, we present an empirical evaluation, using the novel ParallelepipedoNN system.

---

### Beyond the Autoregressive Horizon: A Comprehensive Survey of Diffusion Models, World Modelling, and State Space Models for Code

Authors: Kishan Maharaj, Ashita Saxena, Srikanth Tamilselvam
Published: 2026-04-09
Categories: cs.SE, cs.AI
arXiv: https://arxiv.org/abs/2606.23690v1

Autoregressive (AR) language models have driven significant progress in automated software engineering, enabling powerful code generation and assistance systems. However, the next-token prediction paradigm introduces structural limitations for code reasoning, including restricted global planning, challenges in maintaining long-range dependencies, and limited grounding in program execution semantics. Noting the heavy skewness of existing literature towards AR models, we discuss emerging paradigms that could potentially overcome the logic and scaling bottlenecks of next-token prediction by unlocking next-generation architectural capabilities for code intelligence. Specifically, we discuss the potential of Diffusion Models, which generate code via holistic denoising that captures long-range syntactic constraints often missed by AR models. We also discuss Code World Models (CWMs), which simulate execution states to support reasoning, and State Space Models (SSMs), which provide linear-time efficiency for massive contexts. By connecting these developments with findings from cognitive neuroscience, we outline directions for developing "System 2" code generation agents.

---

### Imperfect Visual Verification for Code Edition : A Case Study on TikZ

Authors: Charly Reux, Mathieu Acher, Djamel Eddine Khelladi, Clément Quinton, Olivier Barais
Published: 2026-04-09
Categories: cs.SE, cs.AI
arXiv: https://arxiv.org/abs/2606.15693v1

LLMs have significantly advanced code generation, enabling the synthesis of functional programs. While recent systems achieve strong performance on many coding benchmarks, tasks involving programs such as TikZ that generate visual artifacts remain challenging, in particular on visual code customization. Unlike generation from scratch, customization requires localized, semantics-preserving edits: the model must locate relevant code, modify it according to the instruction, and preserve the remaining structure and rendering. Approaches based on post-hoc iterative refinement/correction where a verifier provides feedback to guide corrections, have shown promise. However, in the case of programs with a visual outcome such as in TikZ, where correctness is harder or likely impossible to formalize and evaluate automatically, deterministic verifiers do not exist. Hence, developers can only rely on imperfect verifiers. In this paper, we conduct an empirical study to answer:to what extent can iterative refinement remain effective when the verifier itself is unreliable?} We use TikZ as a focused case study that isolates the core difficulties of the problem (weak code structure, fine-grained visual semantics, and difficult feature localization) in a controlled and challenging setting. We define visual code customization as an iterative editing problem with an imperfect oracle, and introduce a framework for analyzing such iterative refinements. We conduct a large-scale study and evaluate multiple LLM-based and tool-augmented visual verifiers within iterative refinement pipelines, and perform extensive manual annotation of refinement trajectories to assess verifier behavior and feedback quality. Our findings show that even imperfect verifiers can determine with moderate accuracy whether visual instructions are applied to code, achieving F1-scores up to 0.815. Feedback improves iterative refinement, especially for weaker models, adding 11--20 perfect customizations for Qwen3-vl-30b…

---

### Large Language Models as Optimizers: A Survey of Direct vs. Tool-Augmented Approaches and Their Performance Frontiers

Authors: Roko Peran, Luka Hobor, Mihael Kovac, Mario Brcic
Published: 2026-04-09
Categories: cs.AI
arXiv: https://arxiv.org/abs/2606.15577v1

Large Language Models (LLMs) are increasingly involved in complex mathematical optimization, even if the pragmatic user who triggers them is unaware of it. After all, many real-world problems reduce to the search for better or the best solutions. The field of LLM-as-optimizer has three paradigms: direct optimization, tool-augmented optimization, and tool-creating optimization. Direct optimization uses iterative prompting and heuristic generation to navigate solution spaces. Tool-augmented optimization translates natural language problems into formal specifications and orchestrates external solvers. Tool-creating optimization goes further, using LLMs to discover reusable algorithms or heuristics that can be deployed at zero marginal LLM cost. We describe current performance frontiers based on the benchmarks from the literature. We identify the critical reasoning gap in current architectures and argue for trade-offs between the future potential of direct optimization and the auditability of tool-augmented optimization. Even future, more powerful models might opt for tool-making to improve operational efficiency for repetitive families of problems.

---

### The Computational Boundary of Inference: Capability Internalization, Training, and the Turing Jump

Authors: Chien-Ping Lu
Published: 2026-04-09
Categories: cs.CC, cs.AI
arXiv: https://arxiv.org/abs/2605.27381v1

Claims about recursive self-improvement in AI often slide from repeated internal revision to the possibility of qualitatively stronger capability without clearly distinguishing the underlying computational regimes. This paper gives a formal separation result in classical computability theory that blocks that move under a precise modeling assumption. For an oracle $A$, let $\mathcal{C}(A)=\{B : B \leq_T A\}$ be the corresponding computational layer. We prove that finite internal self-modification remains inside $\mathcal{C}(A)$, while stabilized revision is governed instead by the jump $A'$ via the relativized limit lemma. Together with a local closure versus escape theorem, this yields a clean formal separation between within-layer iteration and ascent to a stronger relative level. The point is not that stronger layers never arise, but that they are not explained by finite repetition inside one already settled layer. The resulting separation gives a computability-theoretic limit on a broad class of recursive-improvement narratives in which repeated internal updating is treated as sufficient for qualitative capability ascent.

---

### BioELX: Context-Aware Cross-lingual Biomedical Entity Linking without Task-Specific Supervision

Authors: Yi Wang, Corina Dima, Liangyu Zhong, Steffen Staab
Published: 2026-04-09
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2605.27380v2

Cross-lingual biomedical entity linking (BEL) maps mentions in any language to unique identifiers in a biomedical knowledge base, supporting clinical and biomedical NLP applications. We identify two issues affecting current systems. First, the UMLS (Bodenreider,2004) aliases used to train cross-lingual BEL retrievers are heavily skewed toward English, so retrievers generalize poorly to non-English mentions. Second, although context is often necessary for disambiguation, naively injecting context into retrievers trained only to align aliases severely degrades retrieval. We propose BioELX, a retrieve-rerank framework that addresses both issues. For retrieval, we continue training SapBERT_multi (Liu et al., 2021b) using Wikidata-derived cross-lingual alias supervision, forming shared concept neighborhoods across languages. For reranking, we adapt pretrained LLM rerankers to entity linking through mention-anchored prompting, which marks the target mention so that rerankers score candidates with respect to the intended mention rather than other salient tokens in the context. Experiments show that BioELX achieves new state-of-the-art results on four cross-lingual BEL benchmarks, improving Recall@1 by 4.8 to 18.2 percentage points over prior best results, without any task-specific BEL annotations. Our code and resources are available at https://github.com/AI4MedCode/BioELX.

---

### BioELX: Cross-lingual Biomedical Entity Linking via Alias-based Retrieval and LLM Ranking

Authors: Yi Wang, Corina Dima, Liangyu Zhong, Steffen Staab
Published: 2026-04-09
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2605.27380v1

Cross-lingual biomedical entity linking (BEL) maps mentions in any language to unique identifiers in a biomedical knowledge base (KB), supporting clinical and biomedical NLP applications. However, expert-annotated training data for BEL are costly, especially for low-resource languages. Moreover, many cross-lingual BEL systems rely on SapBERT-based retrievers trained on predominantly English aliases in the KB, leading to poor generalization to unseen non-English mentions and limited context-aware disambiguation. We propose BioELX, a two-stage cross-lingual BEL framework that requires no task-specific annotated training corpora. In Stage~1, we enrich SapBERT training with Wikidata-derived multilingual aliases and use the resulting retriever to improve cross-lingual candidate retrieval. In Stage~2, we perform context-aware disambiguation with a pre-trained LLM ranker that jointly considers the mention context and candidate, eliminating the need for supervised training. Experiments on five benchmarks (XL-BEL, EMEA, Patent, WikiMed-DE, and MedMentions) show that BioELX achieves new state-of-the-art performance. It improves average Recall@1 on XL-BEL by +19.2, with especially large gains for low-resource languages, e.g., +21.6 on Turkish, +22.1 on Korean, +30.8 on Thai, and delivers consistent improvements on EMEA (+6.2), Patent (+5.4), and WikiMed-DE (+12.8). Code and resources will be released upon publication.

---

### Soro: A Lightweight Foundation Model and Chatbot for Tajik

Authors: Stanislav Liashkov, Haitz Sáez de Ocáriz Borde, Azizjon Azimi, Khushbakht Shoymardonov, Shuhratjon Khalilbekov, Bonu Boboeva
Published: 2026-04-09
Categories: cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2605.27379v2

We present Soro, a family of Tajik-specialized conversational large language models (LLMs) designed for real-world deployment under tight compute and connectivity constraints in Tajikistan. Starting from open-weight Gemma 3 checkpoints, we perform Tajik-only continual pretraining on a curated 1.9-billion-token corpus spanning filtered web text, PDF documents, and curriculum-aligned educational materials, followed by supervised instruction tuning on 40K Tajik teacher-style examples. To enable rigorous evaluation despite the limited coverage of Tajik in standard benchmarks, we introduce a suite of Tajik benchmarks covering general knowledge, linguistic competence, and school- and university entrance-exam domains, and we open-source them on Hugging Face. Across these Tajik benchmarks, Soro substantially outperforms same-size Gemma 3 baselines while retaining strong English performance on standard datasets. We further show that FP8 and INT4 quantization of Soro preserves most Tajik-language gains while reducing memory requirements for edge deployment, supporting an ongoing education-sector pilot and planned scale-out across schools in Tajikistan.

---

### OralAgent: Integrating Reasoning, Tools, and Knowledge for Interactive Dental Image Analysis

Authors: Jing Hao, Siyuan Dai, Yongxin Zhang, Yuci Liang, Jiamin Wu, Jiahao Bao, Yuxuan Fan, Zanting Ye, Yanpeng Sun, Xinyu Zhang, Ming Hu, Liang Zhan et al.
Published: 2026-04-09
Categories: cs.CL, cs.CV, cs.MA
arXiv: https://arxiv.org/abs/2605.27378v1

Dental image analysis plays a pivotal role in supporting accurate diagnosis and treatment planning in oral healthcare. Although recent advances have produced dental AI models for specific tasks and individual imaging modalities, their isolated designs limit practical use in real-world clinical workflows. In this paper, we present OralAgent, the first dental-specialized AI agent that unifies multimodal reasoning, tool-based decision-making, and knowledge-grounded retrieval within an end-to-end automated framework. It integrates 22 visual analysis tools and 368 widely-used classical dental textbooks, enabling autonomous reasoning, planning, tool use, knowledge retrieval, and multi-step workflow execution. Furthermore, we introduce OralCorpus, a large-scale, high-quality bilingual textual resource containing 134.8M tokens curated for dental retrieval-augmented generation (RAG). To evaluate models' multidisciplinary dental knowledge, we construct OralQA-ZH, a Chinese multiple-choice question benchmark consisting of 798 items across eleven oral subspecialties. Extensive experiments demonstrate that OralAgent achieves state-of-the-art performance on the MMOral-Uni, MMOral-OPG, and OralQA-ZH benchmarks, highlighting its effectiveness, interpretability, and adaptability in real-world clinical settings. The code and models are publicly available at https://github.com/isjinghao/OralAgent.

---

### Enhancing LLM Medical Coding with Structured External Knowledge

Authors: Yidong Gan, David D. Nguyen, Yang Lin, Peter Zhong, Thanh Vu, Long Duong, Yuan-Fang Li
Published: 2026-04-09
Categories: cs.CL, cs.AI, cs.IR
arXiv: https://arxiv.org/abs/2605.27377v2

Accurate medical coding requires consulting authoritative resources such as the ICD tabular list and coding guidelines. Existing LLM-based automated methods largely rely on LLMs' internal knowledge, which is prone to hallucination and cannot keep pace with guideline updates. We introduce RAG-Coding, an agentic, training-free method that augments LLMs with structured external knowledge: the tabular list is encoded as a knowledge graph capturing hierarchical and instructional code relationships, and the guidelines are distilled into concise, code-specific summaries rather than retrieved as raw text. To enable our study, we also introduce MDACE-2025, expert re-annotations of the MDACE dataset under the 2025 ICD-10-CM/PCS guidelines, adding code sequencing and justification comments. On MDACE, RAG-Coding outperforms the best LLM-based baseline by 3--13\% in micro-F1 across five LLM backbones, and achieves comparable micro- and macro-F1 to the supervised state-of-the-art, with higher recall ($+$11\%) at the cost of precision ($-$6\%). On MDACE-2025, RAG-Coding outperforms all baselines, demonstrating effective generalisation to updated guidelines. Ablations confirm stepwise gains, highlighting the importance of integrating structured external knowledge for LLM-based medical coding.

---

### Unlocking Fine-Grained and Within-Utterance Speaking Style Control in Prompt-Based Text-to-Speech Models

Authors: Jaehoon Kang, Yejin Lee, Yoonji Park, Kyuhong Shim
Published: 2026-04-09
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2605.27376v1

While prompt-based text-to-speech (TTS) models enable natural language-driven speaking style control, they often provide limited fine-grained control and apply a single global style across an utterance. This restricts practical use cases that require continuous style attribute interpolation across utterances and time-varying style transitions within a single utterance. In this paper, we propose novel techniques to achieve both capabilities in existing prompt-based TTS models. For inter-utterance style interpolation, we compute direction vectors between contrastive style prompts in the embedding space and perform simple interpolation, enabling smooth transitions between style characteristics. For intra-utterance style transition, we first identify a strong attention bias toward early tokens in autoregressive TTS decoders, causing the initial audio realization to dominate subsequent generation. To mitigate this effect, we introduce KV-cache swapping and sliding-window attention masking. Experiments demonstrate that our proposed inter-utterance interpolation achieves a 99-100% success rate in gender conversion, up to 36 Hz pitch variation, and up to 1.6 syllables-per-second speed change. Our intra-utterance transition maintains a speaker similarity of 0.81-0.91 and achieves perceptual smoothness scores of 3.48-4.48.

---

### VineLM: Trie-Based Fine-Grained Control for Agentic Workflows

Authors: Nikos Pagonas, Matthew Lou, Tianyi Peng, Dan Rubenstein, Kostis Kaffes
Published: 2026-04-09
Categories: cs.DC, cs.AI, cs.MA
arXiv: https://arxiv.org/abs/2605.23914v1

Agentic workflows interleave configurable LLM stages with tool stages and often include retries or refinement loops. Existing workflow managers profile full workflow configurations offline and assign each request a static workflow-level plan that binds each configurable LLM stage to a single model, reuses that model across repeated loop iterations, and does not revisit those choices at runtime. We present VineLM, a workflow manager that enables fine-grained control by choosing the model for each stage invocation as execution unfolds under request-level objectives such as maximizing accuracy under cost or latency budgets. VineLM represents feasible executions as an annotated trie of model-choice prefixes and uses checkpointing and cascade profiling to estimate path accuracy, cost, and latency without exhaustively profiling every request on every path. At runtime, VineLM re-roots the trie after each stage invocation and replans over the remaining subtrie using the realized execution prefix and remaining latency budget. On NL2SQL and math reasoning workflows, VineLM improves the cost-latency-accuracy frontier over coarse workflow-level baselines, achieving up to 18% higher accuracy at the same per-request budget with its sparse profiling reducing offline profiling cost by 98-99.8% when compared to exhaustive profiling.

---

### Can LoRA Fusion Support Cross-Domain Tasks in Cloud-Edge Collaboration?

Authors: Yatong Wang, Fali Wang, Naibin Gu, Zheng Lin, Zhengxiao Liu, Dingyu Yao, Zhiwei Zhang, Jianxin Shi, Weiping Wang
Published: 2026-04-09
Categories: cs.DC, cs.CL
arXiv: https://arxiv.org/abs/2605.23913v1

Cloud-hosted large language models (LLMs) commonly rely on LoRA for domain adaptation, yet domain data are distributed across multiple edge devices and cannot be uploaded due to privacy constraints. This raises a fundamental question: how can knowledge from multiple private edges be integrated into a cloud LLM for cross-domain problem solving? A natural solution is to train LoRA adapters locally and fuse them in the cloud; however, existing pipelines rely on unrealistic assumptions that edge devices can host cloud-scale LLMs and are evaluated mainly on single-domain tasks. To address these limitations, we propose a prune-train-recover framework that enables local LoRA training on pruned models and privacy-preserving cloud integration. We further introduce MMLU-CD, a cross-domain benchmark that composes multiple domain samples into a single instance, enabling explicit evaluation of cross-domain problem solving. This allows us to ask a concrete question: Can existing LoRA fusion methods support cross-domain tasks in cloud-edge collaboration? Our empirical answer is negative. Existing LoRA fusion methods perform poorly on MMLU-CD, often underperforming the base LLM, revealing their inability to support cross-domain problem solving. We attribute this failure to parameter conflicts among LoRA adapters and propose a simple conflict-resolution module, LoRA-CR, which mitigates conflicting updates and improves LoRA fusion performance by up to 3.8%. These results identify conflict mitigation as a critical yet largely overlooked factor in cloud-edge LoRA fusion, warranting further investigation in future research.

---

### Evaluating Large Language Models in a Complex Hidden Role Game

Authors: Niklas Bauer
Published: 2026-04-09
Categories: cs.CL, cs.AI, cs.GT, cs.MA
arXiv: https://arxiv.org/abs/2605.22826v1

Quantifying the deceptive potential of Large Language Models (LLMs) is critical for AI safety, yet difficult to achieve in uncontrolled environments. This work investigates the reasoning, persuasion, and deceptive capabilities of LLMs within the social deduction game Secret Hitler. I introduce an open-source framework and novel metrics to measure performance: Role Identification Accuracy, Deception Retention Rate, and Game State Impact Rate. By benchmarking models against rule-based algorithms and human games, I identify a gap between conversational ability and strategic depth. The study also analyzes the impact of reasoning-enhancement techniques on win rates and strategic reasoning. Neither Chain-of-Thought prompting nor internal memory bring improvements in performance, with up to 23.2% worse win rates for fascist roles. While rule-based agents align with expert human voting decisions 86.7% of the time, models like Llama 3.1 70B achieve only a 59.7% accuracy. Models playing as Fascists consistently yield negative impact scores and fail to sustain deception, resulting in roughly 40% shorter games compared to humans. These findings suggest that current architectures remain ineffective at complex, multi-turn manipulation. As capabilities advance, detecting when models begin to master these deceptive behaviors is crucial. The developed framework serves as a reproducible testbed for future alignment research.

---

### Keeping an Eye on AI: A Framework for Effective Human Oversight of AI Systems

Authors: Susanne Gaube, Markus Langer, Tim Miller, Kevin Baum, Raimund Dachselt, Anna Maria Feit, Ujwal Gadiraju, Harmanpreet Kaur, Mark T. Keane, Richard Landers, Johann Laux, Q. Vera Liao et al.
Published: 2026-04-09
Categories: cs.CY, cs.AI, cs.HC
arXiv: https://arxiv.org/abs/2605.16278v1

The use of Artificial Intelligence (AI) in high-risk, decision-making scenarios presents technical, safety, and normative challenges; problems that may only be ameliorated by human oversight. However, notions of human oversight lack a common foundational understanding: oversight architectures are not well defined, the roles involved remain unclear, and implementation steps are opaque. Hence, researchers and practitioners struggle to determine how to design, implement, and evaluate systems that enable effective human oversight. This paper advances a practical framework for effective human oversight of AI systems, based on a cross-disciplinary perspective that draws on insights from computer science, human-computer interaction, psychology, philosophy, and law. The core contributions are: (1) a foundational framework, with a working definition, architecture and processes for effective human oversight of AI systems; (2) an initial template for documenting oversight architectures and processes, applied to diverse domains; and (3) a synthesis of open research challenges that need to be considered in the emerging field of effective human oversight of AI systems.

---

### MedThink: Enhancing Diagnostic Accuracy in Small Models via Teacher-Guided Reasoning Correction

Authors: Xinchun Su, Chunxu Luo, Lipeng Ma, Yixuan Li, Weidong Yang
Published: 2026-04-09
Categories: cs.CY, cs.AI
arXiv: https://arxiv.org/abs/2605.08094v1

Accurate clinical diagnosis requires extensive domain knowledge and complex clinical reasoning capabilities. Although large language models (LLMs) hold great potential for clinical reasoning, their high computational and memory requirements limit their deployment in resource-constrained environments. Knowledge distillation (KD) can compress LLM capabilities into smaller models, but traditional KD merely transfers superficial answer patterns and fails to preserve the structured reasoning required for reliable diagnosis. To address this, we propose a two-stage distillation framework, MedThink, designed to cultivate robust clinical reasoning in small language models (SLMs). In the first stage, a teacher LLM screens data and injects domain-knowledge explanations to fine-tune a student model, establishing a knowledge foundation. In the second stage, the teacher evaluates the student's errors, generates reasoning chains linking knowledge to correct answers, and refines the student's diagnostic reasoning through a second round of fine-tuning. We evaluate MedThink on general medical benchmarks and a gastroenterology dataset comprising 955 question-answer pairs. Experiments demonstrate that MedThink outperforms six distillation strategies in all benchmarks: achieving an improvement of up to 12.7% over the student baseline in general tasks, and reaching a total top accuracy of 56.4% in gastroenterology evaluation. This indicates that iterative distillation centered on reasoning can significantly enhance the diagnostic accuracy and generalization capabilities of SLMs whilst maintaining computational efficiency. Our code and data are publicly available at https://github.com/destinybird/PrecisionBoost.

---

### Heterogeneous Graph Importance Scoring and Clustering with Automated LLM-based Interpretation

Authors: Takato Yasuno
Published: 2026-04-09
Categories: cs.LG
arXiv: https://arxiv.org/abs/2605.02919v1

Urban bridge networks are critical infrastructure whose disruption can cascade into severe impacts on transportation, emergency services, and economic activity. This paper presents a comprehensive methodology for assessing bridge importance through heterogeneous graph analysis, unsupervised clustering, and automated interpretation via large language models (LLMs). Our approach addresses three fundamental challenges: (1) quantifying multi-dimensional bridge importance using only open data sources, (2) discovering functional bridge archetypes across different cities, and (3) generating policy-relevant interpretations automatically. We construct heterogeneous graphs from OpenStreetMap (OSM) data incorporating bridges, road networks, buildings, and public facilities. Five social impact indicators are computed: transit desert score, hospital access score, isolation risk score, supply chain impact score, and green space access score. These 52-dimensional feature vectors undergo dimensionality reduction via UMAP and density-based clustering via HDBSCAN. Discovered clusters are interpreted using temperature-optimized LLMs (Elyza8b, trained on construction domain corpus). (1) A complete open-data pipeline from OSM to actionable bridge importance rankings, (2) a five-indicator scoring methodology with 40$\times$ computational optimization, (3) a UMAP+HDBSCAN clustering framework validated on multi-city data, (4) an LLM interpretation methodology including temperature optimization and model selection rationale, and (5) transferability demonstration across cities via configuration-only adaptation.

---

### Mitigating the reconstruction-detection trade-off in VAE-based unsupervised anomaly detection

Authors: Agathe Senellart, Maëlys Solal, Stéphanie Allassonnière, Ninon Burgos
Published: 2026-04-09
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2605.02918v1

Variational autoencoders are widely used for unsupervised anomaly detection. Model selection however remains an open-question: to remain fully unsupervised, hyperparameters are often chosen to minimize the reconstruction error on normal samples. In this paper, we reveal a trade-off between reconstruction quality and anomaly detection among $β$-VAE models. Models with constrained latent space reach higher detection metrics but lower reconstruction quality. We also assess the performance variability across random seeds and show it is linked to the distance between normal and abnormal latent distributions. From this analysis, we justify and investigate two methods to mitigate the reconstructiondetection tradeoff: beta-scheduling and the Sparse VAE. The latter especially shows an improvement in detection while maintaining high reconstruction quality.

---

### PRISM-CTG: A Foundation Model for Cardiotocography Analysis with Multi-View SSL

Authors: Sheng Wong, Ravi Shankar, Beth Albert, Hao Fei, Lin Li, Imane Ben M'Barek, Manu Vatish, Gabriel Davis Jones
Published: 2026-04-09
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2605.02917v1

Supervised deep learning models for automated CTG analysis are typically constrained by narrowly curated labelled datasets and limited patient cohorts, leaving substantial volumes of physiologically informative clinical recordings untapped. To address this limitation, we propose Physiology-aware Representation Learning via Integrated Self-supervision and Metadata for CTG (PRISM-CTG), a clinically grounded self-supervised foundation model (FM) for CTG that leverages large-scale unlabelled recordings to learn transferable domain-level representations. PRISM-CTG is pretrained using a multi-view self-supervised framework that jointly optimises 3 complementary pretext objectives: random-projected guided masked signal reconstruction, clinical variable prediction, and feature classification. Each objective is associated with a dedicated task-specific token, enabling specialised representation learning, while controlled cross-attention facilitates information exchange across clinical context. By reframing patient metadata and domain knowledge, which are often underutilised in conventional training as prediction targets, Prism-CTG transforms readily available clinical information into additional supervisory targets that guide clinically meaningful representation learning. Extensive experiments across 7 downstream CTG tasks in both antepartum and intrapartum domains demonstrated that PRISM-CTG consistently outperforms in-domain and SSL baselines. Notably, PRISM-CTG demonstrated strong generalisation under external validation on 2 datasets, while achieving comparable performance to studies trained on substantially larger, privately labelled datasets. To our knowledge, this is the first study to introduce large-scale FM for CTG that learns domain-level representations.

---

### From Synthesis to Clinical Assistance: A Strategy-Aware Agent Framework for Autism Intervention based on Real Clinical Dataset

Authors: Junhong Lai, Shuzhong Lai, Yanhao Yu, Wanlin Chen, Chenyu Yan, Haifeng Li, Lin Yao, Yueming Wang
Published: 2026-04-09
Categories: cs.LG
arXiv: https://arxiv.org/abs/2605.02916v1

The development of AI-assisted Early Intensive Behavioral Intervention (EIBI) for Autism Spectrum Disorder (ASD) is severely constrained by data scarcity. Furthermore, while Applied Behavior Analysis (ABA) serves as the gold standard for clinical intervention, general-purpose Large Language Models (LLMs) struggle to strictly adhere to its standardized procedures, often resulting in interactions that are linguistically fluent but strategically inconsistent. To address these challenges, we introduce \textsc{ASDAgent}, a strategy-aware framework designed to unify high-fidelity intervention dialogue synthesis and clinical decision support. \textsc{ASDAgent} incorporates two specialized components to solve distinct problems: (i) a \textsc{DoctorAgent} equipped with an Observe-Think-Act-Correct (O-T-A-C) reasoning loop, which resolves the issue of strategy collapse in LLMs by making ABA execution explicit and controllable; and (ii) a \textsc{ChildAgent} that utilizes probabilistic behavior modeling to mitigate data homogeneity, simulating diverse and non-deterministic ASD response patterns. Experiments demonstrate that dialogues generated by \textsc{ASDAgent} closely mirror the strategy distribution of human therapists (KL divergence: 0.083). In real autism intervention, \textsc{ASDAgent} achieves nearly 80\% strategic consistency with human experts. Moreover, we show that synthetic data produced by \textsc{ASDAgent} effectively distills professional clinical knowledge into small language models (SLMs), significantly enhancing their therapeutic capabilities.

---

### Graph Query Generation with Constraint-guided Large Language Agents

Authors: Mengying Wang, Nicolaas Jedema, Rahul Pandey, RaviKiran Krishnan, Jens Lehmann, Yinghui Wu
Published: 2026-04-09
Categories: cs.DB, cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2605.00845v1

Knowledge Graph Question Answering (KGQA) has advanced through structured query generation, yet most efforts target RDF/SPARQL, leaving Cypher and property graphs underexplored, despite increasing demand for unified KGQA in industry settings. We propose UniQGen, a novel constraint-based framework that employs LLM agents to dynamically extract and refine representative graph query clauses into executable, intent-aligned graph queries across query languages. The foundation of our method is a variant of Chase & Backchase, a family of algorithms for query optimization and reformulation. We extend Chase & Backchase with a dynamic reasoning process over query constraints that also interact with LLMs for query quality estimation. With a Cypher-supported Freebase graph deployed on Amazon Neptune, we extensively evaluate our approach on popular KGQA benchmarks (GraphQ, GrailQA, and WebQSP). We demonstrate that UniQGen outperforms state-of-the-art graph query generation techniques in both accuracy and efficiency, with F1 gains of 31.6% on GraphQ and 4.9% on GrailQA. Unlike prior methods, our framework does not require fine-tuning for schema matching, making it more extensible to schema-less graphs and semantics in query workloads, and is more suitable for enterprise-grade KGQA. We release Cypher outputs and a Neptune-ready Freebase snapshot to support reproducible, cross-language KGQA research.

---

### Static Program Slicing Using Language Models With Dataflow-Aware Pretraining and Constrained Decoding

Authors: Pengfei He, Shaowei Wang, Tse-Hsun Chen, Muhammad Asaduzzaman
Published: 2026-04-09
Categories: cs.SE, cs.AI, cs.PL
arXiv: https://arxiv.org/abs/2604.26961v2

Static program slicing is a fundamental software engineering technique for isolating code relevant to specific variables. While recent learning-based approaches using language models (LMs) show promise in automating slice prediction, they suffer from inaccurate dependency modeling and unconstrained generation, where LMs fail to capture precise data flow relations and produce slices containing hallucinated tokens and statements. To address these challenges, we propose Sliceformer, a novel approach that reformulates static program slicing as a sequence-to-sequence task using small language models such as CodeT5+. Sliceformer introduces two key innovations that directly target the identified limitations. First, to improve dependency modeling, we design dataflow-aware pretraining objectives that leverage data flow graphs (DFG) to teach models data dependencies through dataflow-preserving statement permutation and dataflow-aware span corruption. Second, to eliminate hallucination, we develop a constrained decoding mechanism that enforces both lexical and syntactic constraints. We evaluate Sliceformer on Java and Python program slicing benchmarks, demonstrating consistent improvements over state-of-the-art baselines with up to 22% gain in ExactMatch.

---

### Dual Triangle Attention: Effective Bidirectional Attention Without Positional Embeddings

Authors: Logan Hallee, Jason P. Gleghorn
Published: 2026-04-09
Categories: q-bio.QM, cs.LG
arXiv: https://arxiv.org/abs/2604.18603v1

Bidirectional transformers are the foundation of many sequence modeling tasks across natural, biological, and chemical language domains, but they are permutation-invariant without explicit positional embeddings. In contrast, unidirectional attention inherently encodes positional information through its triangular mask, enabling models to operate without positional embeddings altogether. Here, we introduce Dual Triangle Attention, a novel bidirectional attention mechanism that separates the query-key subspace of each attention head into two complementary triangular masks: one that attends to past-and-self positions and one that attends to future-and-self positions. This design provides bidirectional context while maintaining the causal mask's implicit positional inductive bias in both directions. Using PyTorch's flex_attention, Dual Triangle Attention is implemented as a single compiled kernel call with no additional parameters beyond standard multi-head attention. We evaluated Dual Triangle Attention across three settings: (1) a synthetic argmax position probe, (2) masked language modeling (MLM) on natural language, and (3) MLM on protein sequences. In the argmax task, both Dual Triangle Attention and causal attention learn positional information without explicit positional embeddings, whereas standard bidirectional attention cannot. In the MLM experiments, Dual Triangle Attention with Rotary Positional Embeddings (RoPE) achieved the best context extension performance and strong performance across the board. These findings suggest that Dual Triangle Attention is a viable attention mechanism for bidirectional transformers, with or without positional embeddings.

---

### B-PASTE: Beam-Aware Pattern-Guided Speculative Execution for Resource-Constrained LLM Agents

Authors: Yanfei Song
Published: 2026-04-09
Categories: cs.DC, cs.AI
arXiv: https://arxiv.org/abs/2604.16469v1

LLM agents execute in an interleaved reasoning-and-action loop, where future tool calls cannot be launched until the current reasoning step completes. This serial dependency inflates end-to-end latency and leaves the model idle while waiting for tool execution. Prior work, Pattern-Aware Speculative Tool Execution (PASTE), mitigates this bottleneck by speculating likely future tool invocations from mined control-flow and data-flow regularities. However, PASTE is tool-centric and speculates only individual invocations rather than bounded future branches. We propose B-PASTE, a beam-aware extension that lifts speculation from single tools to local branch hypotheses under strict resource constraints. B-PASTE maintains a bounded beam of future execution subgraphs, ranks them by expected critical-path reduction rather than raw execution probability, and schedules only high-value branch prefixes on transient slack resources. It explicitly models co-run interference, downstream unlock value, and state-safety constraints, enabling the system to prioritize serial fast-path execution when early completion unlocks valuable future work, while still exploiting safe parallelism under low contention. This design is especially important for edge-side deployments, where speculative work must not steal scarce resources from latency-critical authoritative execution. Preliminary internal testing on Thor-class edge environments shows up to 1.4X end-to-end speedup, suggesting that branch-aware speculative execution remains effective even under tight resource budgets.

---

### Multi-Label Phase Diagram Prediction in Complex Alloys via Physics-Informed Graph Attention Networks

Authors: Eunjeong Park, Amrita Basak
Published: 2026-04-09
Categories: cs.LG, cond-mat.mtrl-sci
arXiv: https://arxiv.org/abs/2604.16468v1

Accurate phase equilibria are foundational to alloy design because they encode the underlying thermodynamics governing stability, transformations, and processing windows. However, while the CALculation of Phase Diagrams (CALPHAD) provides a rigorous thermodynamic framework, exploring multicomponent composition-temperature space remains computationally expensive and is typically limited to sparse section. To enable rapid phase mapping and alloy screening, we propose a physics-informed graph attention network (GAT) that learns element-aware representations and couples them with thermodynamic constraints for multi-label phase-set prediction in the Ag-Bi-Cu-Sn alloy system. Using about 25,000 equilibrium states generated with pycalphad, each composition-temperature point is represented as a four-node element graph with atomic fractions and elemental descriptors as node features. The model combines graph attention, global pooling, and a multilayer perceptron to predict nine relevant phases. To improve physical consistency, we incorporate thermodynamic constraints, applied as training penalties or as an inference-time projection. Across six binary and three ternary subsystems, the baseline model achieves a macro-F1 score of 0.951 and 93.98% exact-set match, while physics-informed decoding improves robustness and raises exact-set accuracy to about 96% on dense in-domain grids. The surrogate also generalizes to an unseen ternary section with 99.32% exact-set accuracy and to a quaternary section at 700 °C with 91.78% accuracy. These results demonstrate that attention-based graph learning coupled with thermodynamic constraint enforcement provides an effective and physically consistent surrogate for high-resolution phase mapping and extrapolative alloy screening.

---

### On Tackling Complex Tasks with Reward Machines and Signal Temporal Logics

Authors: Ana María Gómez Ruiz, Thao Dang, Alexandre Donzé
Published: 2026-04-09
Categories: cs.AI
arXiv: https://arxiv.org/abs/2604.14440v1

We propose a Reinforcement Learning (RL) based control design framework for handling complex tasks. The approach extends the concept of Reward Machines (RM) with Signal Temporal Logic (STL) formulas that can be used for event generation. The use of STL allows not only a more efficient representation of rewards for complex tasks but also guiding the training process to converge towards behaviors satisfying specified requirements. We also propose an implementation of the framework that leverages the STL online monitoring algorithms. We illustrate the framework with three case studies (minigrid, cart-pole and high-way environments) with non-trivial tasks.

---

### Functional Emotions or Situational Contexts? A Discriminating Test from the Mythos Preview System Card

Authors: Hiranya V. Peiris
Published: 2026-04-09
Categories: cs.HC, cs.AI, cs.CL, cs.LG
arXiv: https://arxiv.org/abs/2604.13466v2

The Claude Mythos Preview system card deploys emotion vectors, sparse autoencoder (SAE) features, and activation verbalisers to study model internals during misaligned behaviour. The two primary toolkits are not jointly reported on the most alignment-relevant episodes. This note identifies two hypotheses that are qualitatively consistent with the published results: that the emotion vectors track functional emotions that causally drive behaviour, or that they are a projection of a richer situational-context structure onto human emotional axes. The hypotheses can be distinguished by cross-referencing the two toolkits on episodes where only one is currently reported: most directly, applying emotion probes to the strategic concealment episodes analysed only with SAE features. If emotion probes show flat activation while SAE features are strongly active, the alignment-relevant structure lies outside the emotion subspace. Which hypothesis is correct determines whether emotion-based monitoring will robustly detect dangerous model behaviour or systematically miss it.

---

### GRACE: A Dynamic Coreset Selection Framework for Large Language Model Optimization

Authors: Tianhao Tang, Haoyang Li, Lei Chen
Published: 2026-04-09
Categories: cs.DB, cs.AI
arXiv: https://arxiv.org/abs/2604.11810v1

Large Language Models (LLMs) have demonstrated remarkable capabilities in natural language understanding and generation. However, their immense number of parameters and complex transformer-based architectures result in significant resource demands and computational complexity during training, making it challenging to optimize them efficiently on large datasets. To reduce training costs while preserving performance, researchers have investigated coreset selection techniques, which aim to identify small, representative subsets of the entire training dataset to accelerate LLM training. However, existing coreset selection methods fail to adapt to the dynamic nature of LLM training and often struggle with scalability for models of this size. To address these limitations, we propose a graph-guided adaptive and dynamic coreset selection framework for LLMs, namely GRACE. GRACE dynamically constructs and updates coresets by combining representation diversity with gradient-based importance metrics, ensuring both informativeness and efficiency. To mitigate the computational cost of frequent updates, GRACE leverages a $k$-NN graph-based propagation mechanism and selectively updates scores and embeddings, adapting to evolving training dynamics. Extensive experiments on three benchmarks demonstrate that GRACE significantly improves training efficiency and downstream performance across diverse LLMs and tasks.

---

### Improving DNS Exfiltration Detection via Transformer Pretraining

Authors: Miloš Tomić, Aleksa Cvetanović, Predrag Tadić
Published: 2026-04-09
Categories: cs.CR, cs.LG
arXiv: https://arxiv.org/abs/2604.09849v1

We study whether in-domain pretraining of Bidirectional Encoder Representations from Transformer (BERT) model improves subdomain-level detection of exfiltration at low false positive rates. While previous work mostly examines fine-tuned generic Transformers, it does not aim to isolate the effect of pretraining on the downstream task of classification. To address this gap, we develop a controlled pipeline where we freeze operating points on validation and transfer them to the test set, thus enabling clean ablations across different label and pretraining budgets. Our results show significant improvements in the left tail of the Receiver Operating Characteristic (ROC) curve, especially against randomly initialized baseline. Additionally, within pretrained model variants, increasing the number of pretraining steps helps the most when more labeled data are available for fine-tuning.

---