# Research Papers — 2026-04 · bundle page 123 of 233 > Full-text markdown of papers #6101–6150 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/123 [← Previous bundle](https://aipod.dev/papers/archive/2026-04/page/122.md) [Next bundle →](https://aipod.dev/papers/archive/2026-04/page/124.md) ### Pareto-Optimal Offline Reinforcement Learning via Smooth Tchebycheff Scalarization Authors: Aadyot Bhatnagar, Peter Mørch Groth, Sebastian Ibarraran, Ali Madani Published: 2026-04-14 Categories: cs.LG, cs.AI, q-bio.BM, q-bio.QM arXiv: https://arxiv.org/abs/2604.13175v2 Large language models can be aligned with human preferences through offline reinforcement learning (RL) on small labeled datasets. While single-objective alignment is well-studied, many real-world applications demand the simultaneous optimization of multiple conflicting rewards, e.g. activity and specificity for proteins, or helpfulness and harmlessness for chatbots. Prior work has largely relied on linear reward scalarization, which provably fails to recover non-convex regions of the Pareto front. In this paper, instead of scalarizing the rewards directly, we frame multi-objective RL itself as an optimization problem to be scalarized via smooth Tchebycheff scalarization, a recent technique that overcomes the drawbacks of linear scalarization. We use this formulation to derive Smooth Tchebycheff Optimization of Multi-Objective Preferences (STOMP), a novel offline RL algorithm that extends direct preference optimization to the multi-objective setting in a principled way by standardizing the individual rewards based on their observed distributions. We empirically validate STOMP using multiple base models on 3 protein engineering and 3 chatbot alignment datasets. Compared to state-of-the-art baselines, STOMP achieves or ties for the highest hypervolumes on 16/18 protein tasks and 5/6 natural language tasks. We thus demonstrate that STOMP is a powerful, robust multi-objective alignment algorithm that can meaningfully improve post-training in multiple domains. --- ### Pareto-Optimal Offline Reinforcement Learning via Smooth Tchebysheff Scalarization Authors: Aadyot Bhatnagar, Peter Mørch Groth, Ali Madani Published: 2026-04-14 Categories: cs.LG, cs.AI, q-bio.BM, q-bio.QM arXiv: https://arxiv.org/abs/2604.13175v1 Large language models can be aligned with human preferences through offline reinforcement learning (RL) on small labeled datasets. While single-objective alignment is well-studied, many real-world applications demand the simultaneous optimization of multiple conflicting rewards, e.g. optimizing both catalytic activity and specificity in protein engineering, or helpfulness and harmlessness for chatbots. Prior work has largely relied on linear reward scalarization, but this approach provably fails to recover non-convex regions of the Pareto front. In this paper, instead of scalarizing the rewards directly, we frame multi-objective RL itself as an optimization problem to be scalarized via smooth Tchebysheff scalarization, a recent technique that overcomes the shortcomings of linear scalarization. We use this formulation to derive Smooth Tchebysheff Optimization of Multi-Objective Preferences (STOMP), a novel offline RL algorithm that extends direct preference optimization to the multi-objective setting in a principled way by standardizing the individual rewards based on their observed distributions. We empirically validate STOMP on a range of protein engineering tasks by aligning three autoregressive protein language models on three laboratory datasets of protein fitness. Compared to state-of-the-art baselines, STOMP achieves the highest hypervolumes in eight of nine settings according to both offline off-policy and generative evaluations. We thus demonstrate that STOMP is a powerful, robust multi-objective alignment algorithm that can meaningfully improve post-trained models for multi-attribute protein optimization and beyond. --- ### 3DRealHead: Few-Shot Detailed Head Avatar Authors: Jalees Nehvi, Timo Bolkart, Thabo Beeler, Justus Thies Published: 2026-04-14 Categories: cs.CV arXiv: https://arxiv.org/abs/2604.13171v1 The human face is central to communication. For immersive applications, the digital presence of a person should mirror the physical reality, capturing the users idiosyncrasies and detailed facial expressions. However, current 3D head avatar methods often struggle to faithfully reproduce the identity and facial expressions, despite having multi-view data or learned priors. Learning priors that capture the diversity of human appearances, especially, for regions with highly person-specific features, like the mouth and teeth region is challenging as the underlying training data is limited. In addition, many of the avatar methods are purely relying on 3D morphable model-based expression control which strongly limits expressivity. To address these challenges, we are introducing 3DRealHead, a few-shot head avatar reconstruction method with a novel expression control signal that is extracted from a monocular video stream of the subject. Specifically, the subject can take a few pictures of themselves, recover a 3D head avatar and drive it with a consumer-level webcam. The avatar reconstruction is enabled via a novel few-shot inversion process of a 3D human head prior which is represented as a Style U-Net that emits 3D Gaussian primitives which can be rendered under novel views. The prior is learned on the NeRSemble dataset. For animating the avatar, the U-Net is conditioned on 3DMM-based facial expression signals, as well as features of the mouth region extracted from the driving video. These additional mouth features allow us to recover facial expressions that cannot be represented by the 3DMM leading to a higher expressivity and closer resemblance to the physical reality. --- ### PatchPoison: Poisoning Multi-View Datasets to Degrade 3D Reconstruction Authors: Prajas Wadekar, Venkata Sai Pranav Bachina, Kunal Bhosikar, Ankit Gangwal, Charu Sharma Published: 2026-04-14 Categories: cs.CV, cs.CR, cs.LG arXiv: https://arxiv.org/abs/2604.13153v1 3D Gaussian Splatting (3DGS) has recently enabled highly photorealistic 3D reconstruction from casually captured multi-view images. However, this accessibility raises a privacy concern: publicly available images or videos can be exploited to reconstruct detailed 3D models of scenes or objects without the owner's consent. We present PatchPoison, a lightweight dataset-poisoning method that prevents unauthorized 3D reconstruction. Unlike global perturbations, PatchPoison injects a small high-frequency adversarial patch, a structured checkerboard, into the periphery of each image in a multi-view dataset. The patch is designed to corrupt the feature-matching stage of Structure-from-Motion (SfM) pipelines such as COLMAP by introducing spurious correspondences that systematically misalign estimated camera poses. Consequently, downstream 3DGS optimization diverges from the correct scene geometry. On the NeRF-Synthetic benchmark, inserting a 12 X 12 pixel patch increases reconstruction error by 6.8x in LPIPS, while the poisoned images remain unobtrusive to human viewers. PatchPoison requires no pipeline modifications, offering a practical, "drop-in" preprocessing step for content creators to protect their multi-view data. --- ### Exploration and Exploitation Errors Are Measurable for Language Model Agents Authors: Jaden Park, Jungtaek Kim, Jongwon Jeong, Robert D. Nowak, Kangwook Lee, Yong Jae Lee Published: 2026-04-14 Categories: cs.AI arXiv: https://arxiv.org/abs/2604.13151v1 Language Model (LM) agents are increasingly used in complex open-ended decision-making tasks, from AI coding to physical AI. A core requirement in these settings is the ability to both explore the problem space and exploit acquired knowledge effectively. However, systematically distinguishing and quantifying exploration and exploitation from observed actions without access to the agent's internal policy remains challenging. To address this, we design controllable environments inspired by practical embodied AI scenarios. Each environment consists of a partially observable 2D grid map and an unknown task Directed Acyclic Graph (DAG). The map generation can be programmatically adjusted to emphasize exploration or exploitation difficulty. To enable policy-agnostic evaluation, we design a metric to quantify exploration and exploitation errors from agent's actions. We evaluate a variety of frontier LM agents and find that even state-of-the-art models struggle on our task, with different models exhibiting distinct failure modes. We further observe that reasoning models solve the task more effectively and show both exploration and exploitation can be significantly improved through minimal harness engineering. We release our code \href{https://github.com/jjj-madison/measurable-explore-exploit}{here}. --- ### Adaptive Learning via Off-Model Training and Importance Sampling for Fully Non-Markovian Optimal Stochastic Control. Complete version Authors: Dorival Leão, Alberto Ohashi, Simone Scotti, Adolfo M. D da Silva Published: 2026-04-14 Categories: stat.ML, cs.LG, math.PR arXiv: https://arxiv.org/abs/2604.13147v2 This paper studies continuous-time stochastic control problems whose controlled states are fully non-Markovian and depend on unknown model parameters. Such problems arise naturally in path-dependent stochastic differential equations, rough-volatility hedging, and systems driven by fractional Brownian motion. Building on the discrete skeleton approach developed in earlier work, we propose a Monte Carlo learning methodology for the associated embedded backward dynamic programming equation. Our main contribution is twofold. First, we construct explicit dominating training laws and Radon--Nikodym weights for several representative classes of non-Markovian controlled systems. This yields an off-model training architecture in which a fixed synthetic dataset is generated under a reference law, while the dynamic programming operators associated with a target model are recovered by importance sampling. Second, we use this structure to design an adaptive update mechanism under parametric model uncertainty, so that repeated recalibration can be performed by reweighting the same training sample rather than regenerating new trajectories. For fixed parameters, we establish non-asymptotic error bounds for the approximation of the embedded dynamic programming equation via deep neural networks. For adaptive learning, we derive quantitative estimates that separate Monte Carlo approximation error from model-risk error. Numerical experiments illustrate both the off-model training mechanism and the adaptive importance-sampling update in structured linear-quadratic examples. --- ### Multi-modal panoramic 3D outdoor datasets for place categorization Authors: Hojung Jung, Yuki Oto, Oscar M. Mozos, Yumi Iwashita, Ryo Kurazume Published: 2026-04-14 Categories: cs.RO, cs.CV, cs.DB arXiv: https://arxiv.org/abs/2604.13142v1 We present two multi-modal panoramic 3D outdoor (MPO) datasets for semantic place categorization with six categories: forest, coast, residential area, urban area and indoor/outdoor parking lot. The first dataset consists of 650 static panoramic scans of dense (9,000,000 points) 3D color and reflectance point clouds obtained using a FARO laser scanner with synchronized color images. The second dataset consists of 34,200 real-time panoramic scans of sparse (70,000 points) 3D reflectance point clouds obtained using a Velodyne laser scanner while driving a car. The datasets were obtained in the city of Fukuoka, Japan and are publicly available in (1), (2). In addition, we compare several approaches for semantic place categorization with best results of 96.42% (dense) and 89.67% (sparse). --- ### Automated co-design of high-performance thermodynamic cycles via graph-based hierarchical reinforcement learning Authors: Wenqing Li, Xu Feng, Peixue Jiang, Yinhai Zhu Published: 2026-04-14 Categories: cs.LG arXiv: https://arxiv.org/abs/2604.13133v1 Thermodynamic cycles are pivotal in determining the efficacy of energy conversion systems. Traditional design methodologies, which rely on expert knowledge or exhaustive enumeration, are inefficient and lack scalability, thereby constraining the discovery of high-performance cycles. In this study, we introduce a graph-based hierarchical reinforcement learning approach for the co-design of structure parameters in thermodynamic cycles. These cycles are encoded as graphs, with components and connections depicted as nodes and edges, adhering to grammatical constraints. A deep learning-based thermophysical surrogate facilitates stable graph decoding and the simultaneous resolution of global parameters. Building on this foundation, we develop a hierarchical reinforcement learning framework wherein a high-level manager explores structural evolution and proposes candidate configurations, whereas a low-level worker optimizes parameters and provides performance rewards to steer the search towards high-performance regions. By integrating graph representation, thermophysical surrogate, and manager-worker learning, this method establishes a fully automated pipeline for encoding, decoding, and co-optimization. Using heat pump and heat engine cycles as case studies, the results demonstrate that the proposed method not only replicates classical cycle configurations but also identifies 18 and 21 novel heat pump and heat engine cycles, respectively. Relative to classical cycles, the novel configurations exhibit performance improvements of 4.6% and 133.3%, respectively, surpassing the traditional designs. This method effectively balances efficiency with broad applicability, providing a practical and scalable intelligent alternative to expert-driven thermodynamic cycle design. --- ### Depth-Resolved Coral Reef Thermal Fields from Satellite SST and Sparse In-Situ Loggers Using Physics-Informed Neural Networks Authors: Alzayat Saleh, Mostafa Rahimi Azghadi Published: 2026-04-14 Categories: cs.LG, cs.CV arXiv: https://arxiv.org/abs/2604.13131v1 Satellite sea surface temperature (SST) products underpin global coral bleaching monitoring, yet they measure only the ocean skin. Corals inhabit depths from the shallows to beyond 20 metres, where temperatures can be 1-3°C cooler than the surface; applying satellite SST uniformly to all depths therefore overestimates subsurface thermal stress. We present a physics-informed neural network (PINN) that fuses NOAA Coral Reef Watch SST with sparse in-situ temperature loggers within the one-dimensional vertical heat equation, enforcing SST as a hard surface boundary condition and jointly learning effective thermal diffusivity (\k{appa}) and light attenuation (Kd). Validated across four Great Barrier Reef sites (30 holdout experiments), the PINN achieves 0.25-1.38°C RMSE at unseen depths. Under extreme sparsity (three training depths), the PINN maintains 0.27°C RMSE at the 5 metre holdout and 0.32°C at the 9.1 metre holdout, where statistical baselines collapse to >1.8°C; it outperforms a physics-only finite-difference baseline in 90% of experiments. Depth-resolved Degree Heating Day (DHD) profiles show that thermal stress attenuates with depth: at Davies Reef, DHD drops from 0.29 at the surface to zero by 10.7 metres, consistent with logger observations, while satellite DHD remains constant at 0.31 across all depths. However, the PINN underestimates absolute DHD at shallow depths because its smooth predictions attenuate the short-duration peaks that drive threshold exceedances; PINN DHD values should be interpreted as conservative lower bounds on depth-resolved stress. These results demonstrate that physics-constrained fusion of satellite SST with sparse loggers can extend bleaching assessment to the depth dimension using existing observational infrastructure. --- ### Lyra 2.0: Explorable Generative 3D Worlds Authors: Tianchang Shen, Sherwin Bahmani, Kai He, Sangeetha Grama Srinivasan, Tianshi Cao, Jiawei Ren, Ruilong Li, Zian Wang, Nicholas Sharp, Zan Gojcic, Sanja Fidler, Jiahui Huang et al. Published: 2026-04-14 Categories: cs.CV arXiv: https://arxiv.org/abs/2604.13036v1 Recent advances in video generation enable a new paradigm for 3D scene creation: generating camera-controlled videos that simulate scene walkthroughs, then lifting them to 3D via feed-forward reconstruction techniques. This generative reconstruction approach combines the visual fidelity and creative capacity of video models with 3D outputs ready for real-time rendering and simulation. Scaling to large, complex environments requires 3D-consistent video generation over long camera trajectories with large viewpoint changes and location revisits, a setting where current video models degrade quickly. Existing methods for long-horizon generation are fundamentally limited by two forms of degradation: spatial forgetting and temporal drifting. As exploration proceeds, previously observed regions fall outside the model's temporal context, forcing the model to hallucinate structures when revisited. Meanwhile, autoregressive generation accumulates small synthesis errors over time, gradually distorting scene appearance and geometry. We present Lyra 2.0, a framework for generating persistent, explorable 3D worlds at scale. To address spatial forgetting, we maintain per-frame 3D geometry and use it solely for information routing -- retrieving relevant past frames and establishing dense correspondences with the target viewpoints -- while relying on the generative prior for appearance synthesis. To address temporal drifting, we train with self-augmented histories that expose the model to its own degraded outputs, teaching it to correct drift rather than propagate it. Together, these enable substantially longer and 3D-consistent video trajectories, which we leverage to fine-tune feed-forward reconstruction models that reliably recover high-quality 3D scenes. --- ### SceneCritic: A Symbolic Evaluator for 3D Indoor Scene Synthesis Authors: Kathakoli Sengupta, Kai Ao, Paola Cascante-Bonilla Published: 2026-04-14 Categories: cs.CV, cs.CL arXiv: https://arxiv.org/abs/2604.13035v1 Large Language Models (LLMs) and Vision-Language Models (VLMs) increasingly generate indoor scenes through intermediate structures such as layouts and scene graphs, yet evaluation still relies on LLM or VLM judges that score rendered views, making judgments sensitive to viewpoint, prompt phrasing, and hallucination. When the evaluator is unstable, it becomes difficult to determine whether a model has produced a spatially plausible scene or whether the output score reflects the choice of viewpoint, rendering, or prompt. We introduce SceneCritic, a symbolic evaluator for floor-plan-level layouts. SceneCritic's constraints are grounded in SceneOnto, a structured spatial ontology we construct by aggregating indoor scene priors from 3D-FRONT, ScanNet, and Visual Genome. SceneOnto traverses this ontology to jointly verify semantic, orientation, and geometric coherence across object relationships, providing object-level and relationship-level assessments that identify specific violations and successful placements. Furthermore, we pair SceneCritic with an iterative refinement test bed that probes how models build and revise spatial structure under different critic modalities: a rule-based critic using collision constraints as feedback, an LLM critic operating on the layout as text, and a VLM critic operating on rendered observations. Through extensive experiments, we show that (a) SceneCritic aligns substantially better with human judgments than VLM-based evaluators, (b) text-only LLMs can outperform VLMs on semantic layout quality, and (c) image-based VLM refinement is the most effective critic modality for semantic and orientation correction. --- ### Generative Refinement Networks for Visual Synthesis Authors: Jian Han, Jinlai Liu, Jiahuan Wang, Bingyue Peng, Zehuan Yuan Published: 2026-04-14 Categories: cs.CV arXiv: https://arxiv.org/abs/2604.13030v2 While diffusion models dominate the field of visual generation, they are computationally inefficient, applying a uniform computational effort regardless of different complexity. In contrast, autoregressive (AR) models are inherently complexity-aware, as evidenced by their variable likelihoods, but are often hindered by lossy discrete tokenization and error accumulation. In this work, we introduce Generative Refinement Networks (GRN), a next-generation visual synthesis paradigm that addresses these issues. At its core, GRN addresses the discrete tokenization bottleneck through a theoretically near-lossless Hierarchical Binary Quantization (HBQ), achieving a reconstruction quality comparable to continuous counterparts. Built upon HBQ's latent space, GRN fundamentally upgrades AR generation with a global refinement mechanism that progressively perfects and corrects artworks -- like a human artist painting. Besides, GRN integrates an entropy-guided sampling strategy, enabling complexity-aware, adaptive-step generation without compromising visual quality. On the ImageNet benchmark, GRN establishes new records in image reconstruction (0.56 rFID) and class-conditional image generation (1.81 gFID). We also scale GRN to more challenging text-to-image and text-to-video generation, delivering superior performance on an equivalent scale. We release all models and code to foster further research on GRN. --- ### Visual Preference Optimization with Rubric Rewards Authors: Ya-Qi Yu, Fangyu Hong, Xiangyang Qu, Hao Wang, Gaojie Wu, Qiaoyu Luo, Nuo Xu, Huixin Wang, Wuheng Xu, Yongxin Liao, Zihao Chen, Haonan Li et al. Published: 2026-04-14 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2604.13029v1 The effectiveness of Direct Preference Optimization (DPO) depends on preference data that reflect the quality differences that matter in multimodal tasks. Existing pipelines often rely on off-policy perturbations or coarse outcome-based signals, which are not well suited to fine-grained visual reasoning. We propose rDPO, a preference optimization framework based on instance-specific rubrics. For each image-instruction pair, we create a checklist-style rubric of essential and additional criteria to score responses from any possible policies. The instruction-rubric pool is built offline and reused during the construction of on-policy data. On public reward modeling benchmarks, rubric-based prompting massively improves a 30B-A3B judge and brings it close to GPT-5.4. On public downstream benchmarks, rubric-based filtering raises the macro average to 82.69, whereas outcome-based filtering drops it to 75.82 from 81.14. When evaluating scalability on a comprehensive benchmark, rDPO achieves 61.01, markedly outperforming the style-constrained baseline (52.36) and surpassing the 59.48 base model. Together, these results show that visual preference optimization benefits from combining on-policy data construction with instance-specific criterion-level feedback. --- ### Conflated Inverse Modeling to Generate Diverse and Temperature-Change Inducing Urban Vegetation Patterns Authors: Baris Sarper Tezcan, Hrishikesh Viswanath, Rubab Saher, Daniel Aliaga Published: 2026-04-14 Categories: cs.CV arXiv: https://arxiv.org/abs/2604.13028v1 Urban areas are increasingly vulnerable to thermal extremes driven by rapid urbanization and climate change. Traditionally, thermal extremes have been monitored using Earth-observing satellites and numerical modeling frameworks. For example, land surface temperature derived from Landsat or Sentinel imagery is commonly used to characterize surface heating patterns. These approaches operate as forward models, translating radiative observations or modeled boundary conditions into estimates of surface thermal states. While forward models can predict land surface temperature from vegetation and urban form, the inverse problem of determining spatial vegetation configurations that achieve a desired regional temperature shift remains largely unexplored. This task is inherently underdetermined, as multiple spatial vegetation patterns can yield similar aggregated temperature responses. Conventional regression and deterministic neural networks fail to capture this ambiguity and often produce averaged solutions, particularly under data-scarce conditions. We propose a conflated inverse modeling framework that combines a predictive forward model with a diffusion-based generative inverse model to produce diverse, physically plausible image-based vegetation patterns conditioned on specific temperature goals. Our framework maintains control over thermal outcomes while enabling diverse spatial vegetation configurations, even when such combinations are absent from training data. Altogether, this work introduces a controllable inverse modeling approach for urban climate adaptation that accounts for the inherent diversity of the problem. Code is available at the GitHub repository. --- ### CLAD: Efficient Log Anomaly Detection Directly on Compressed Representations Authors: Benzhao Tang, Shiyu Yang Published: 2026-04-14 Categories: cs.LG, cs.DB arXiv: https://arxiv.org/abs/2604.13024v1 The explosive growth of system logs makes streaming compression essential, yet existing log anomaly detection (LAD) methods incur severe pre-processing overhead by requiring full decompression and parsing. We introduce CLAD, the first deep learning framework to perform LAD directly on compressed byte streams. CLAD bypasses these bottlenecks by exploiting a key insight: normal logs compress into regular byte patterns, while anomalies systematically disrupt them. To extract these multi-scale deviations from opaque bytes, we propose a purpose-built architecture integrating a dilated convolutional byte encoder, a hybrid Transformer--mLSTM, and four-way aggregation pooling. This is coupled with a two-stage training strategy of masked pre-training and focal-contrastive fine-tuning to effectively handle severe class imbalance. Evaluated across five datasets, CLAD achieves a state-of-the-art average F1-score of 0.9909 and outperforms the best baseline by 2.72 percentage points. It delivers superior accuracy while completely eliminating decompression and parsing overheads, offering a robust solution that generalizes to structured streaming compressors. --- ### Classical and Quantum Speedups for Non-Convex Optimization via Energy Conserving Descent Authors: Yihang Sun, Huaijin Wang, Patrick Hayden, Jose Blanchet Published: 2026-04-14 Categories: quant-ph, cs.LG, math.OC, stat.ML arXiv: https://arxiv.org/abs/2604.13022v1 The Energy Conserving Descent (ECD) algorithm was recently proposed (De Luca & Silverstein, 2022) as a global non-convex optimization method. Unlike gradient descent, appropriately configured ECD dynamics escape strict local minima and converge to a global minimum, making it appealing for machine learning optimization. We present the first analytical study of ECD, focusing on the one-dimensional setting for this first installment. We formalize a stochastic ECD dynamics (sECD) with energy-preserving noise, as well as a quantum analog of the ECD Hamiltonian (qECD), providing the foundation for a quantum algorithm through Hamiltonian simulation. For positive double-well objectives, we compute the expected hitting time from a local to the global minimum. We prove that both sECD and qECD yield exponential speedup over respective gradient descent baselines--stochastic gradient descent and its quantization. For objectives with tall barriers, qECD achieves a further speedup over sECD. --- ### Representation geometry shapes task performance in vision-language modeling for CT enterography Authors: Cristian Minoccheri, Emily Wittrup, Kayvan Najarian, Ryan Stidham Published: 2026-04-14 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2604.13021v1 Computed tomography (CT) enterography is a primary imaging modality for assessing inflammatory bowel disease (IBD), yet the representational choices that best support automated analysis of this modality are unknown. We present the first study of vision-language transfer learning on abdominal CT enterography and identify two main findings. First, mean pooling of slice embeddings gives better categorical disease assessment (59.2\% three-class accuracy), whereas attention pooling gives better cross-modal retrieval (0.235 text-to-image MRR). This pattern holds across all LoRA configurations tested and suggests that the two aggregators emphasize different properties of the learned representation. Second, per-slice tissue contrast matters more than broader spatial coverage: multi-window RGB encoding, which maps complementary Hounsfield Unit windows to RGB channels, outperforms all strategies that increase spatial coverage through multiplanar sampling, and in this setting adding coronal and sagittal views reduces classification performance. For report generation, fine-tuning without retrieval context yields within-1 severity accuracy at the prevalence-matched chance level (70.4\% vs.\ 71\% random), suggesting little learned ordering beyond the class distribution. Retrieval-augmented generation (RAG) improves this across all configurations, scoring 7--14 percentage points above the chance baseline and improving ordinal MAE from 0.98 to 0.80--0.89. A three-teacher pseudolabel framework enables all comparisons without expert annotations. Together, these findings provide the first baselines for this underexplored modality and offer practical guidance for building vision-language systems for volumetric medical imaging. --- ### PrecisionCUA: Iterative Visual Refinement for Pixel-Precise Cursor Grounding in Code Editors Authors: Himangi Mittal, Gaurav Mittal, Nelson Daniel Troncoso, Yu Hu Published: 2026-04-14 Categories: cs.CV arXiv: https://arxiv.org/abs/2604.13019v3 Computer Use Agents (CUAs) fundamentally rely on graphical user interface (GUI) grounding to translate language instructions into executable screen actions, but editing-level grounding in dense coding interfaces (such as VS Code and Cursor), where sub-pixel accuracy is required to interact with dense IDE elements, remains underexplored. Existing approaches typically rely on single-shot coordinate prediction, which lacks a mechanism for error correction and often fails in high-density interfaces. In this technical report, we conduct an empirical study of pixel-precise cursor localization in coding environments. Instead of a single-step execution, our agent engages in an iterative refinement process, utilizing visual feedback from previous attempts to reach the target element. This closed-loop grounding mechanism allows the agent to self-correct displacement errors and adapt to dynamic UI changes. We evaluate our approach across Claude, Qwen, and GPT on a suite of complex coding benchmarks, demonstrating that multi-turn refinement significantly outperforms state-of-the-art single-shot models in both click precision and overall task success rate. Our results suggest that iterative visual reasoning is a critical component for the next generation of reliable software engineering agents. Code: https://github.com/microsoft/precision-cua-bench/tree/main. --- ### Toward Autonomous Long-Horizon Engineering for ML Research Authors: Guoxin Chen, Jie Chen, Lei Chen, Jiale Zhao, Fanzhe Meng, Wayne Xin Zhao, Ruihua Song, Cheng Chen, Ji-Rong Wen, Kai Jia Published: 2026-04-14 Categories: cs.CL arXiv: https://arxiv.org/abs/2604.13018v2 Agentic systems increasingly automate pieces of AI research. Yet turning underspecified research objectives into runnable, experimentally validated ML systems remains a central bottleneck. We study this operational setting as \emph{long-horizon ML research engineering}: converting a research specification into a runnable ML system through repeated implementation, experimentation, and refinement. The central challenge is to sustain cumulative project progress across heterogeneous stages under delayed, confounded feedback. We introduce AiScientist, a multi-agent system built around thin control over thick state: a lightweight hierarchical research team coordinates through a File-as-Bus workspace that preserves decision-relevant artifacts across roles and invocations. On PaperBench, AiScientist improves over the strongest matched baselines by 9.92 and 11.15 points with Gemini-3-Flash and GLM-5, respectively. On MLE-Bench Lite, it reaches 81.82 Any Medal\% under both backbones, improving over the strongest matched baselines by 4.55 and 16.67 points, and exceeding a Codex/GPT-5.5 xhigh frontier harness reference by 13.64 Any Medal points. Ablations and process analyses show that durable project state is central to later-round refinement: removing File-as-Bus lowers PaperBench score by 6.41 points and MLE-Bench Lite Any Medal\% by 31.82 points. These results suggest that long-horizon AI research is not only a problem of stronger local reasoning, but a systems problem of maintaining cumulative, inspectable project progress. --- ### PAL: Personal Adaptive Learner Authors: Megha Chakraborty, Darssan L. Eswaramoorthi, Madhur Thareja, Het Riteshkumar Shah, Finlay Palmer, Aryaman Bahl, Michelle A Ihetu, Amit Sheth Published: 2026-04-14 Categories: cs.AI, cs.HC arXiv: https://arxiv.org/abs/2604.13017v1 AI-driven education platforms have made some progress in personalisation, yet most remain constrained to static adaptation--predefined quizzes, uniform pacing, or generic feedback--limiting their ability to respond to learners' evolving understanding. This shortfall highlights the need for systems that are both context-aware and adaptive in real time. We introduce PAL (Personal Adaptive Learner), an AI-powered platform that transforms lecture videos into interactive learning experiences. PAL continuously analyzes multimodal lecture content and dynamically engages learners through questions of varying difficulty, adjusting to their responses as the lesson unfolds. At the end of a session, PAL generates a personalized summary that reinforces key concepts while tailoring examples to the learner's interests. By uniting multimodal content analysis with adaptive decision-making, PAL contributes a novel framework for responsive digital learning. Our work demonstrates how AI can move beyond static personalization toward real-time, individualized support, addressing a core challenge in AI-enabled education. --- ### Rethinking On-Policy Distillation of Large Language Models: Phenomenology, Mechanism, and Recipe Authors: Yaxuan Li, Yuxin Zuo, Bingxiang He, Jinqian Zhang, Chaojun Xiao, Cheng Qian, Tianyu Yu, Huan-ang Gao, Wenkai Yang, Zhiyuan Liu, Ning Ding Published: 2026-04-14 Categories: cs.LG, cs.AI, cs.CL arXiv: https://arxiv.org/abs/2604.13016v2 On-policy distillation (OPD) has become a core technique in the post-training of large language models, yet its training dynamics remain poorly understood. This paper provides a systematic investigation of OPD dynamics and mechanisms. We first identify that two conditions govern whether OPD succeeds or fails: (i) the student and teacher should share compatible thinking patterns; and (ii) even with consistent thinking patterns and higher scores, the teacher must offer genuinely new capabilities beyond what the student has seen during training. We validate these findings through weak-to-strong reverse distillation, showing that same-family 1.5B and 7B teachers are distributionally indistinguishable from the student's perspective. Probing into the token-level mechanism, we show that successful OPD is characterized by progressive alignment on high-probability tokens at student-visited states, a small shared token set that concentrates most of the probability mass (97%-99%). We further propose two practical strategies to recover failing OPD: off-policy cold start and teacher-aligned prompt selection. Finally, we show that OPD's apparent free lunch of dense token-level reward comes at a cost, raising the question of whether OPD can scale to long-horizon distillation. --- ### Bilevel Late Acceptance Hill Climbing for the Electric Capacitated Vehicle Routing Problem Authors: Yinghao Qin, Mosab Bazargani, Edmund K. Burke, Carlos A. Coello Coello, Zhongmin Song, Jun Chen Published: 2026-04-14 Categories: cs.AI, math.OC arXiv: https://arxiv.org/abs/2604.13013v1 This paper tackles the Electric Capacitated Vehicle Routing Problem (E-CVRP) through a bilevel optimization framework that handles routing and charging decisions separately or jointly depending on the search stage. By analyzing their interaction, we introduce a surrogate objective at the upper level to guide the search and accelerate convergence. A bilevel Late Acceptance Hill Climbing algorithm (b-LAHC) is introduced that operates through three phases: greedy descent, neighborhood exploration, and final solution refinement. b-LAHC operates with fixed parameters, eliminating the need for complex adaptation while remaining lightweight and effective. Extensive experiments on the IEEE WCCI-2020 benchmark show that b-LAHC achieves superior or competitive performance against eight state-of-the-art algorithms. Under a fixed evaluation budget, it attains near-optimal solutions on small-scale instances and sets 9/10 new best-known results on large-scale benchmarks, improving existing records by an average of 1.07%. Moreover, the strong correlation (though not universal) observed between the surrogate objective and the complete cost justifies the use of the surrogate objective while still necessitating a joint solution of both levels, thereby validating the effectiveness of the proposed bilevel framework and highlighting its potential for efficiently solving large-scale routing problems with a hierarchical structure. --- ### Lightning OPD: Efficient Post-Training for Large Reasoning Models with Offline On-Policy Distillation Authors: Yecheng Wu, Song Han, Han Cai Published: 2026-04-14 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2604.13010v3 On-policy distillation (OPD) is an effective post-training paradigm for large language models but requires a live teacher server throughout training, resulting in substantial infrastructure overhead. We investigate whether OPD can be performed offline by precomputing teacher log-probabilities once over SFT rollouts and reusing them during training. We find that naively doing so fails to reliably match standard OPD, and trace the root cause to a previously overlooked condition we term teacher consistency, requiring that the same teacher be used for both supervised fine-tuning and OPD. Violating this condition introduces a gradient bias that degrades performance for both offline and online OPD. Building on this insight, we propose Lightning OPD, an offline on-policy distillation framework that enforces teacher consistency and eliminates the need for a live teacher server entirely. We prove that, under teacher consistency, Lightning OPD shares the same optimum as standard OPD, with bounded gradient discrepancy and an implicit regularization effect that helps prevent policy drift. Experiments on math reasoning and code generation show that Lightning OPD achieves comparable performance to standard OPD while delivering 4.0x higher training efficiency. Starting from an SFT-initialized Qwen3-8B-Base model, Lightning OPD reaches 69.9% on AIME 2024 in just 30 GPU hours. Lightning OPD further scales to MoE architectures, training Qwen3-30B-A3B to 71.0% on AIME 2024 on a single 8xH100 node, substantially lowering the barrier for academic research on LLM post-training. Our code is released at https://github.com/jet-ai-projects/Lightning-OPD. --- ### Lightning OPD: Efficient Post-Training for Large Reasoning Models with Offline On-Policy Distillation Authors: Yecheng Wu, Song Han, Hai Cai Published: 2026-04-14 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2604.13010v2 On-policy distillation (OPD) is an effective post-training paradigm for large language models but requires a live teacher server throughout training, resulting in substantial infrastructure overhead. We investigate whether OPD can be performed offline by precomputing teacher log-probabilities once over SFT rollouts and reusing them during training. We find that naively doing so fails to reliably match standard OPD, and trace the root cause to a previously overlooked condition we term teacher consistency, requiring that the same teacher be used for both supervised fine-tuning and OPD. Violating this condition introduces a gradient bias that degrades performance for both offline and online OPD. Building on this insight, we propose Lightning OPD, an offline on-policy distillation framework that enforces teacher consistency and eliminates the need for a live teacher server entirely. We prove that, under teacher consistency, Lightning OPD shares the same optimum as standard OPD, with bounded gradient discrepancy and an implicit regularization effect that helps prevent policy drift. Experiments on math reasoning and code generation show that Lightning OPD achieves comparable performance to standard OPD while delivering 4.0x higher training efficiency. Starting from an SFT-initialized Qwen3-8B-Base model, Lightning OPD reaches 69.9% on AIME 2024 in just 30 GPU hours. Lightning OPD further scales to MoE architectures, training Qwen3-30B-A3B to 71.0% on AIME 2024 on a single 8xH100 node, substantially lowering the barrier for academic research on LLM post-training. Our code is released at https://github.com/jet-ai-projects/Lightning-OPD. --- ### Nonparametric efficient inference for network quantile causal effects under partial interference Authors: Chao Cheng, Fan Li Published: 2026-04-14 Categories: stat.ME, stat.AP, stat.ML arXiv: https://arxiv.org/abs/2604.13008v1 Interference arises when the treatment assigned to one individual affects the outcomes of other individuals. Commonly, individuals are naturally grouped into clusters, and interference occurs only among individuals within the same cluster, a setting referred to as partial interference. We study network causal effects on outcome quantiles in the presence of partial interference. We develop a general nonparametric efficiency theory for estimating these network quantile causal effects, which leads to a nonparametrically efficient estimator. The proposed estimator is consistent and asymptotically normal with parametric convergence rates, while allowing for flexible, data-adaptive estimation of complex nuisance functions. We leverage a three-way cross-fitting procedure that avoids direct estimation of the conditional outcome distribution. Simulations demonstrate adequate finite-sample performance of the proposed estimators, and we apply the methods to a clustered observational study. --- ### One Token Away from Collapse: The Fragility of Instruction-Tuned Helpfulness Authors: Erfan Baghaei Potraghloo, Seyedarmin Azizi, Souvik Kundu, Massoud Pedram Published: 2026-04-14 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2604.13006v2 Instruction-tuned large language models produce helpful, structured responses, but how robust is this helpfulness under trivial constraints? We show that simple lexical constraints (banning a single punctuation character or common word) cause instruction-tuned LLMs to collapse their responses, losing 14--48\% of comprehensiveness across seven models spanning five families (7B--70B, open- and closed-weight). A blinded human evaluation with 10 STEM-trained evaluators confirms genuine content loss, with information criteria degrading $1.5$--$2.3\times$ more than surface criteria, a finding corroborated by over 4,100 automated pairwise comparisons (77--100\% baseline preference) across three LLM judges from two model families. Diagnostic analysis identifies this as a \emph{planning failure}: two-pass generation recovers 59--96\% of response length, and linear probes on prompt representations predict response length with $R^2 = 0.51$--$0.94$ before generation begins. The same probes yield negative $R^2$ on base models, confirming that instruction tuning introduces the representational structure underlying the collapse. Base models show no systematic degradation under identical constraints, demonstrating that instruction tuning couples task competence to narrow surface-form templates. The effect extends to realistic deployment constraints (preamble suppression, corporate tone guidelines, legal compliance hedging, accessibility requirements) causing comparable degradation ($-$22\% to $-$34\%), with suppressing the conversational opener alone (``Certainly!'') causing 40\% collapse on our most fragile model despite restricting only the opening tokens. We further show that standard independent LLM-as-judge evaluation detects only a 3.5\% quality drop where pairwise evaluation reveals 23\%, exposing a methodological blind spot in current evaluation practice. --- ### Agentic Discovery with Active Hypothesis Exploration for Visual Recognition Authors: Jaywon Koo, Jefferson Hernandez, Ruozhen He, Hanjie Chen, Chen Wei, Vicente Ordonez Published: 2026-04-14 Categories: cs.CV arXiv: https://arxiv.org/abs/2604.12999v1 We introduce HypoExplore, an agentic framework that formulates neural architecture discovery for visual recognition as a hypothesis-driven scientific inquiry. Given a human-specified high-level research direction, HypoExplore ideates, implements, evaluates, and improves neural architectures through evolutionary branching. New hypotheses are created using a large language model by selecting a parent hypothesis to build upon, guided by a dual strategy that balances exploiting validated principles with resolving uncertain ones. Our proposed framework maintains a Trajectory Tree that records the lineage of all proposed architectures, and a Hypothesis Memory Bank that actively tracks confidence scores acquired through experimental evidence. After each experiment, multiple feedback agents analyze the results from different perspectives and consolidate their findings into hypothesis confidence updates. Our framework is tested on discovering lightweight vision architectures on CIFAR-10, with the best achieving 94.11% accuracy evolved from a root node baseline that starts at 18.91%, and generalizes to CIFAR-100 and Tiny-ImageNet. We further demonstrate applicability to a specialized domain by conducting independent architecture discovery runs on MedMNIST, which yield a state-of-the-art performance. We show that hypothesis confidence scores grow increasingly predictive as evidence accumulates, and that the learned principles transfer across independent evolutionary lineages, suggesting that HypoExplore not only discovers stronger architectures, but can help build a genuine understanding of the design space. --- ### PolicyLLM: Towards Excellent Comprehension of Public Policy for Large Language Models Authors: Han Bao, Penghao Zhang, Yue Huang, Zhengqing Yuan, Yanchi Ru, Rui Su, Yujun Zhou, Xiangqi Wang, Kehan Guo, Nitesh V Chawla, Yanfang Ye, Xiangliang Zhang Published: 2026-04-14 Categories: cs.CL, cs.CY arXiv: https://arxiv.org/abs/2604.12995v1 Large Language Models (LLMs) are increasingly integrated into real-world decision-making, including in the domain of public policy. Yet, their ability to comprehend and reason about policy-related content remains underexplored. To fill this gap, we present \textbf{\textit{PolicyBench}}, the first large-scale cross-system benchmark (US-China) evaluating policy comprehension, comprising 21K cases across a broad spectrum of policy areas, capturing the diversity and complexity of real-world governance. Following Bloom's taxonomy, the benchmark assesses three core capabilities: (1) \textbf{Memorization}: factual recall of policy knowledge, (2) \textbf{Understanding}: conceptual and contextual reasoning, and (3) \textbf{Application}: problem-solving in real-life policy scenarios. Building on this benchmark, we further propose \textbf{\textit{PolicyMoE}}, a domain-specialized Mixture-of-Experts (MoE) model with expert modules aligned to each cognitive level. The proposed models demonstrate stronger performance on application-oriented policy tasks than on memorization or conceptual understanding, and yields the highest accuracy on structured reasoning tasks. Our results reveal key limitations of current LLMs in policy understanding and suggest paths toward more reliable, policy-focused models. --- ### LogicEval: A Systematic Framework for Evaluating Automated Repair Techniques for Logical Vulnerabilities in Real-World Software Authors: Syed Md Mukit Rashid, Abdullah Al Ishtiaq, Kai Tu, Yilu Dong, Tianwei Wu, Ali Ranjbar, Tianchang Yang, Najrin Sultana, Shagufta Mehnaz, Syed Rafiul Hussain Published: 2026-04-14 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2604.12994v2 Logical vulnerabilities in software stem from flaws in program logic rather than memory safety, which can lead to critical security failures. Although existing automated program repair techniques primarily focus on repairing memory corruption vulnerabilities, they struggle with logical vulnerabilities because of their limited semantic understanding of the vulnerable code and its expected behavior. On the other hand, recent successes of large language models (LLMs) in understanding and repairing code are promising. However, no framework currently exists to analyze the capabilities and limitations of such techniques for logical vulnerabilities. We aim to systematically evaluate both traditional and LLM based repair approaches for addressing real world logical vulnerabilities. To facilitate our assessment, we created the first ever dataset, LogicDS, comprising 122 logical vulnerabilities that reflect tangible security impact. We also developed a systematic framework, LogicEval, to evaluate patches for logical vulnerabilities. Evaluations suggest that compilation and testing failures are primarily driven by prompt sensitivity, loss of code context, and difficulty in patch localization. --- ### Causal Diffusion Models for Counterfactual Outcome Distributions in Longitudinal Data Authors: Farbod Alinezhad, Jianfei Cao, Gary J. Young, Brady Post Published: 2026-04-14 Categories: stat.ML, cs.LG, econ.EM arXiv: https://arxiv.org/abs/2604.12992v1 Predicting counterfactual outcomes in longitudinal data, where sequential treatment decisions heavily depend on evolving patient states, is critical yet notoriously challenging due to complex time-dependent confounding and inadequate uncertainty quantification in existing methods. We introduce the Causal Diffusion Model (CDM), the first denoising diffusion probabilistic approach explicitly designed to generate full probabilistic distributions of counterfactual outcomes under sequential interventions. CDM employs a novel residual denoising architecture with relational self-attention, capturing intricate temporal dependencies and multimodal outcome trajectories without requiring explicit adjustments (e.g., inverse-probability weighting or adversarial balancing) for confounding. In rigorous evaluation on a pharmacokinetic-pharmacodynamic tumor-growth simulator widely adopted in prior work, CDM consistently outperforms state-of-the-art longitudinal causal inference methods, achieving a 15-30% relative improvement in distributional accuracy (1-Wasserstein distance) while maintaining competitive or superior point-estimate accuracy (RMSE) under high-confounding regimes. By unifying uncertainty quantification and robust counterfactual prediction in complex, sequentially confounded settings, without tailored deconfounding, CDM offers a flexible, high-impact tool for decision support in medicine, policy evaluation, and other longitudinal domains. --- ### Accelerating Speculative Decoding with Block Diffusion Draft Trees Authors: Liran Ringel, Yaniv Romano Published: 2026-04-14 Categories: cs.CL arXiv: https://arxiv.org/abs/2604.12989v1 Speculative decoding accelerates autoregressive language models by using a lightweight drafter to propose multiple future tokens, which the target model then verifies in parallel. DFlash shows that a block diffusion drafter can generate an entire draft block in a single forward pass and achieve state-of-the-art speculative decoding performance, outperforming strong autoregressive drafters such as EAGLE-3. Vanilla DFlash, however, still verifies only a single drafted trajectory per round, potentially limiting its acceptance length. We introduce DDTree (Diffusion Draft Tree), a method that constructs a draft tree directly from the per-position distributions of a block diffusion drafter. Under a fixed node budget, DDTree uses a simple best-first heap algorithm to select the continuations that are most likely to match the target model according to a surrogate defined by the draft model's output. The resulting tree is verified efficiently in a single target model forward pass using an ancestor-only attention mask. Because DDTree builds on DFlash, a leading draft model for speculative decoding, these gains place DDTree among the leading approaches to speculative decoding. --- ### ROSE: An Intent-Centered Evaluation Metric for NL2SQL Authors: Wenqi Pei, Shizheng Hou, Boyan Li, Han Chen, Zhichao Shi, Yuyu Luo Published: 2026-04-14 Categories: cs.DB, cs.AI arXiv: https://arxiv.org/abs/2604.12988v1 Execution Accuracy (EX), the widely used metric for evaluating the effectiveness of Natural Language to SQL (NL2SQL) solutions, is becoming increasingly unreliable. It is sensitive to syntactic variation, ignores that questions may admit multiple interpretations, and is easily misled by erroneous ground-truth SQL. To address this, we introduce ROSE, an intent-centered metric that focuses on whether the predicted SQL answers the question, rather than consistency with the ground-truth SQL under the reference-dependent paradigm. ROSE employs an adversarial Prover-Refuter cascade: SQL Prover assesses the semantic correctness of a predicted SQL against the user's intent independently, while Adversarial Refuter uses the ground-truth SQL as evidence to challenge and refine this judgment. On our expert-aligned validation set ROSE-VEC, ROSE achieves the best agreement with human experts, outperforming the next-best metric by nearly 24% in Cohen's Kappa. We also conduct a largescale re-evaluation of 19 NL2SQL methods, revealing four valuable insights. We release ROSE and ROSE-VEC to facilitate more reliable NL2SQL research. --- ### Parallax: Why AI Agents That Think Must Never Act Authors: Joel Fokou Published: 2026-04-14 Categories: cs.CR, cs.AI arXiv: https://arxiv.org/abs/2604.12986v1 Autonomous AI agents are rapidly transitioning from experimental tools to operational infrastructure, with projections that 80% of enterprise applications will embed AI copilots by the end of 2026. As agents gain the ability to execute real-world actions (reading files, running commands, making network requests, modifying databases), a fundamental security gap has emerged. The dominant approach to agent safety relies on prompt-level guardrails: natural language instructions that operate at the same abstraction level as the threats they attempt to mitigate. This paper argues that prompt-based safety is architecturally insufficient for agents with execution capability and introduces Parallax, a paradigm for safe autonomous AI execution grounded in four principles: Cognitive-Executive Separation, which structurally prevents the reasoning system from executing actions; Adversarial Validation with Graduated Determinism, which interposes an independent, multi-tiered validator between reasoning and execution; Information Flow Control, which propagates data sensitivity labels through agent workflows to detect context-dependent threats; and Reversible Execution, which captures pre-destructive state to enable rollback when validation fails. We present OpenParallax, an open-source reference implementation in Go, and evaluate it using Assume-Compromise Evaluation, a methodology that bypasses the reasoning system entirely to test the architectural boundary under full agent compromise. Across 280 adversarial test cases in nine attack categories, Parallax blocks 98.9% of attacks with zero false positives under its default configuration, and 100% of attacks under its maximum-security configuration. When the reasoning system is compromised, prompt-level guardrails provide zero protection because they exist only within the compromised system; Parallax's architectural boundary holds regardless. --- ### GlotOCR Bench: OCR Models Still Struggle Beyond a Handful of Unicode Scripts Authors: Amir Hossein Kargaran, Nafiseh Nikeghbal, Jana Diesner, François Yvon, Hinrich Schütze Published: 2026-04-14 Categories: cs.CL, cs.CV arXiv: https://arxiv.org/abs/2604.12978v1 Optical character recognition (OCR) has advanced rapidly with the rise of vision-language models, yet evaluation has remained concentrated on a small cluster of high- and mid-resource scripts. We introduce GlotOCR Bench, a comprehensive benchmark evaluating OCR generalization across 100+ Unicode scripts. Our benchmark comprises clean and degraded image variants rendered from real multilingual texts. Images are rendered using fonts from the Google Fonts repository, shaped with HarfBuzz and rasterized with FreeType, supporting both LTR and RTL scripts. Samples of rendered images were manually reviewed to verify correct rendering across all scripts. We evaluate a broad suite of open-weight and proprietary vision-language models and find that most perform well on fewer than ten scripts, and even the strongest frontier models fail to generalize beyond thirty scripts. Performance broadly tracks script-level pretraining coverage, suggesting that current OCR systems rely on language model pretraining as much as on visual recognition. Models confronted with unfamiliar scripts either produce random noise or hallucinate characters from similar scripts they already know. We release the benchmark and pipeline for reproducibility. Pipeline Code: https://github.com/cisnlp/glotocr-bench, Benchmark: https://hf.co/datasets/cis-lmu/glotocr-bench. --- ### Probabilistic Feature Imputation and Uncertainty-Aware Multimodal Federated Aggregation Authors: Nafis Fuad Shahid, Maroof Ahmed, Md Akib Haider, Saidur Rahman Sagor, Aashnan Rahman, Md Azam Hossain Published: 2026-04-14 Categories: eess.IV, cs.CV arXiv: https://arxiv.org/abs/2604.12970v1 Multimodal federated learning enables privacy-preserving collaborative model training across healthcare institutions. However, a fundamental challenge arises from modality heterogeneity: many clinical sites possess only a subset of modalities due to resource constraints or workflow variations. Existing approaches address this through feature imputation networks that synthesize missing modality representations, yet these methods produce point estimates without reliability measures, forcing downstream classifiers to treat all imputed features as equally trustworthy. In safety-critical medical applications, this limitation poses significant risks. We propose the Probabilistic Feature Imputation Network (P-FIN), which outputs calibrated uncertainty estimates alongside imputed features. This uncertainty is leveraged at two levels: (1) locally, through sigmoid gating that attenuates unreliable feature dimensions before classification, and (2) globally, through Fed-UQ-Avg, an aggregation strategy that prioritizes updates from clients with reliable imputation. Experiments on federated chest X-ray classification using CheXpert, NIH Open-I, and PadChest demonstrate consistent improvements over deterministic baselines, with +5.36% AUC gain in the most challenging configuration. --- ### AbdomenGen: Sequential Volume-Conditioned Diffusion Framework for Abdominal Anatomy Generation Authors: Yubraj Bhandari, Lavsen Dahal, Paul Segars, Joseph Y. Lo Published: 2026-04-14 Categories: cs.CV arXiv: https://arxiv.org/abs/2604.12969v1 Computational phantoms are widely used in medical imaging research, yet current systems to generate controlled, clinically meaningful anatomical variations remain limited. We present AbdomenGen, a sequential volume-conditioned diffusion framework for controllable abdominal anatomy generation. We introduce the \textbf{Volume Control Scalar (VCS)}, a standardized residual that decouples organ size from body habitus, enabling interpretable volume modulation. Organ masks are synthesized sequentially, conditioning on the body mask and previously generated structures to preserve global anatomical coherence while supporting independent, multi-organ control. Across 11 abdominal organs, the proposed framework achieves strong geometric fidelity (e.g., liver dice $0.83 \pm 0.05$), stable single-organ calibration over $(-3,+3)$ VCS, and disentangled multi-organ modulation. To showcase clinical utility with a hepatomegaly cohort selected from MERLIN, Wasserstein-based VCS selection reduces distributional distance of training data by 73.6\% . These results demonstrate calibrated, distribution-aware anatomical generation suitable for controllable abdominal phantom construction and simulation studies. --- ### Evolution of Optimization Methods: Algorithms, Scenarios, and Evaluations Authors: Tong Zhang, Jiangning Zhang, Zhucun Xue, Juntao Jiang, Yicheng Xu, Chengming Xu, Teng Hu, Xingyu Xie, Xiaobin Hu, Yabiao Wang, Yong Liu, Shuicheng Yan Published: 2026-04-14 Categories: cs.LG, cs.CV arXiv: https://arxiv.org/abs/2604.12968v1 Balancing convergence speed, generalization capability, and computational efficiency remains a core challenge in deep learning optimization. First-order gradient descent methods, epitomized by stochastic gradient descent (SGD) and Adam, serve as the cornerstone of modern training pipelines. However, large-scale model training, stringent differential privacy requirements, and distributed learning paradigms expose critical limitations in these conventional approaches regarding privacy protection and memory efficiency. To mitigate these bottlenecks, researchers explore second-order optimization techniques to surpass first-order performance ceilings, while zeroth-order methods reemerge to alleviate memory constraints inherent to large-scale training. Despite this proliferation of methodologies, the field lacks a cohesive framework that unifies underlying principles and delineates application scenarios for these disparate approaches. In this work, we retrospectively analyze the evolutionary trajectory of deep learning optimization algorithms and present a comprehensive empirical evaluation of mainstream optimizers across diverse model architectures and training scenarios. We distill key emerging trends and fundamental design trade-offs, pinpointing promising directions for future research. By synthesizing theoretical insights with extensive empirical evidence, we provide actionable guidance for designing next-generation highly efficient, robust, and trustworthy optimization methods. The code is available at https://github.com/APRIL-AIGC/Awesome-Optimizer. --- ### Cycle-Consistent Search: Question Reconstructability as a Proxy Reward for Search Agent Training Authors: Sohyun An, Shuibenyang Yuan, Hayeon Lee, Cho-Jui Hsieh, Alexander Min Published: 2026-04-14 Categories: cs.AI arXiv: https://arxiv.org/abs/2604.12967v1 Reinforcement Learning (RL) has shown strong potential for optimizing search agents in complex information retrieval tasks. However, existing approaches predominantly rely on gold supervision, such as ground-truth answers, which is difficult to scale. To address this limitation, we propose Cycle-Consistent Search (CCS), a gold-supervision-free framework for training search agents, inspired by cycle-consistency techniques from unsupervised machine translation and image-to-image translation. Our key hypothesis is that an optimal search trajectory, unlike insufficient or irrelevant ones, serves as a lossless encoding of the question's intent. Consequently, a high-quality trajectory should preserve the information required to accurately reconstruct the original question, thereby inducing a reward signal for policy optimization. However, naive cycle-consistency objectives are vulnerable to information leakage, as reconstruction may rely on superficial lexical cues rather than the underlying search process. To reduce this effect, we apply information bottlenecks, including exclusion of the final response and named entity recognition (NER) masking of search queries. These constraints force reconstruction to rely on retrieved observations together with the structural scaffold, ensuring that the resulting reward signal reflects informational adequacy rather than linguistic redundancy. Experiments on question-answering benchmarks show that CCS achieves performance comparable to supervised baselines while outperforming prior methods that do not rely on gold supervision. These results suggest that CCS provides a scalable training paradigm for training search agents in settings where gold supervision is unavailable. --- ### Boosting Visual Instruction Tuning with Self-Supervised Guidance Authors: Sophia Sirko-Galouchenko, Monika Wysoczanska, Andrei Bursuc, Nicolas Thome, Spyros Gidaris Published: 2026-04-14 Categories: cs.CV arXiv: https://arxiv.org/abs/2604.12966v1 Multimodal large language models (MLLMs) perform well on many vision-language tasks but often struggle with vision-centric problems that require fine-grained visual reasoning. Recent evidence suggests that this limitation arises not from weak visual representations, but from under-utilization of visual information during instruction tuning, where many tasks can be partially solved using language priors alone. We propose a simple and lightweight approach that augments visual instruction tuning with a small number of visually grounded self-supervised tasks expressed as natural language instructions. By reformulating classical self-supervised pretext tasks, such as rotation prediction, color matching, and cross-view correspondence, as image-instruction-response triplets, we introduce supervision that cannot be solved without relying on visual evidence. Our approach requires no human annotations, no architectural modifications, and no additional training stages. Across multiple models, training regimes, and benchmarks, injecting only a small fraction (3-10%) of such visually grounded instructions consistently improves performance on vision-centric evaluations. Our findings highlight instruction tuning with visually grounded SSL tasks as a powerful lever for improving visual reasoning in MLLMs through simple adjustments to the training data distribution. Code available at: https://github.com/sirkosophia/V-GIFT --- ### Text2Model: Modeling Copilots for Text-to-Model Translation Authors: Serdar Kadioglu, Karthik Uppuluri, Akash Singirikonda Published: 2026-04-14 Categories: cs.AI arXiv: https://arxiv.org/abs/2604.12955v3 There is growing interest in leveraging large language models (LLMs) for text-to-model translation and optimization tasks. This paper aims to advance this line of research by introducing \textsc{Text2Model} and \textsc{Text2Zinc}. \textsc{Text2Model} is a suite of copilots based on several LLM strategies with varying complexity, along with an online leaderboard. \textsc{Text2Zinc} is a cross-domain dataset for capturing optimization and satisfaction problems specified in natural language, along with an interactive editor with built-in AI assistant. While there is an emerging literature on using LLMs for translating combinatorial problems into formal models, our work is the first attempt to integrate \textit{both} satisfaction and optimization problems within a \textit{unified architecture} and \textit{dataset}. Moreover, our approach is \textit{solver-agnostic} unlike existing work that focuses on translation to a solver-specific model. To achieve this, we leverage \textsc{MiniZinc}'s solver-and-paradigm-agnostic modeling capabilities to formulate combinatorial problems. We conduct comprehensive experiments to compare execution and solution accuracy across several single- and multi-call strategies, including; zero-shot prompting, chain-of-thought reasoning, intermediate representations via knowledge-graphs, grammar-based syntax encoding, and agentic approaches that decompose the model into sequential sub-tasks. Our copilot strategies are competitive, and in parts improve, recent research in this domain. Our findings indicate that while LLMs are promising they are not yet a push-button technology for combinatorial modeling. We contribute \textsc{Text2Model} copilots and leaderboard, and \textsc{Text2Zinc} and interactive editor to open-source to support closing this performance gap. --- ### An Optimal Sauer Lemma Over $k$-ary Alphabets Authors: Steve Hanneke, Qinglin Meng, Shay Moran, Amirreza Shaeiri Published: 2026-04-14 Categories: cs.LG, math.CO, stat.ML arXiv: https://arxiv.org/abs/2604.12952v1 The Sauer-Shelah-Perles Lemma is a cornerstone of combinatorics and learning theory, bounding the size of a binary hypothesis class in terms of its Vapnik-Chervonenkis (VC) dimension. For classes of functions over a $k$-ary alphabet, namely the multiclass setting, the Natarajan dimension has long served as an analogue of VC dimension, yet the corresponding Sauer-type bounds are suboptimal for alphabet sizes $k>2$. In this work, we establish a sharp Sauer inequality for multiclass and list prediction. Our bound is expressed in terms of the Daniely--Shalev-Shwartz (DS) dimension, and more generally with its extension, the list-DS dimension -- the combinatorial parameters that characterize multiclass and list PAC learnability. Our bound is tight for every alphabet size $k$, list size $\ell$, and dimension value, replacing the exponential dependence on $\ell$ in the Natarajan-based bound by the optimal polynomial dependence, and improving the dependence on $k$ as well. Our proof uses the polynomial method. In contrast to the classical VC case, where several direct combinatorial proofs are known, we are not aware of any purely combinatorial proof in the DS setting. This motivates several directions for future research, which are discussed in the paper. As consequences, we obtain improved sample complexity upper bounds for list PAC learning and for uniform convergence of list predictors, sharpening the recent results of Charikar et al.~(STOC~2023), Hanneke et al.~(COLT~2024), and Brukhim et al.~(NeurIPS~2024). --- ### The Verification Tax: Fundamental Limits of AI Auditing in the Rare-Error Regime Authors: Jason Z Wang Published: 2026-04-14 Categories: cs.LG arXiv: https://arxiv.org/abs/2604.12951v1 The most cited calibration result in deep learning -- post-temperature-scaling ECE of 0.012 on CIFAR-100 (Guo et al., 2017) -- is below the statistical noise floor. We prove this is not a failure of the experiment but a law: the minimax rate for estimating calibration error with model error rate epsilon is Theta((Lepsilon/m)^{1/3}), and no estimator can beat it. This "verification tax" implies that as AI models improve, verifying their calibration becomes fundamentally harder -- with the same exponent in opposite directions. We establish four results that contradict standard evaluation practice: (1) self-evaluation without labels provides exactly zero information about calibration, bounded by a constant independent of compute; (2) a sharp phase transition at mepsilon approx 1 below which miscalibration is undetectable; (3) active querying eliminates the Lipschitz constant, collapsing estimation to detection; (4) verification cost grows exponentially with pipeline depth at rate L^K. We validate across five benchmarks (MMLU, TruthfulQA, ARC-Challenge, HellaSwag, WinoGrande; ~27,000 items) with 6 LLMs from 5 families (8B-405B parameters, 27 benchmark-model pairs with logprob-based confidence), 95% bootstrap CIs, and permutation tests. Self-evaluation non-significance holds in 80% of pairs. Across frontier models, 23% of pairwise comparisons are indistinguishable from noise, implying that credible calibration claims must report verification floors and prioritize active querying once gains approach benchmark resolution. --- ### Drawing on Memory: Dual-Trace Encoding Improves Cross-Session Recall in LLM Agents Authors: Benjamin Stern, Peter Nadel Published: 2026-04-14 Categories: cs.AI arXiv: https://arxiv.org/abs/2604.12948v1 LLM agents with persistent memory store information as flat factual records, providing little context for temporal reasoning, change tracking, or cross-session aggregation. Inspired by the drawing effect (3), we introduce dual-trace memory encoding. In this method, each stored fact is paired with a concrete scene trace, a narrative reconstruction of the moment and context in which the information was learned. The agent is forced to commit to specific contextual details during encoding, creating richer, more distinctive memory traces. Using the LongMemEval-S benchmark (4,575 sessions, 100 recall questions), we compare dual-trace encoding against a fact-only control with matched coverage and format over 99 shared questions. Dual-trace achieves 73.7% overall accuracy versus 53.5%, a +20.2 percentage point (pp) gain (95% CI: (+12.1, +29.3), bootstrap p < 0.0001). Gains concentrate in temporal reasoning (+40pp), knowledge-update tracking (+25pp), and multi-session aggregation (+30pp), with no benefit for single-session retrieval, consistent with encoding specificity theory (8). Token analysis shows dual-trace encoding achieves this gain at no additional cost. We additionally sketch an architectural design for adapting dual-trace encoding to coding agents, with preliminary pilot validation. --- ### Parcae: Scaling Laws For Stable Looped Language Models Authors: Hayden Prairie, Zachary Novack, Taylor Berg-Kirkpatrick, Daniel Y. Fu Published: 2026-04-14 Categories: cs.LG arXiv: https://arxiv.org/abs/2604.12946v1 Traditional fixed-depth architectures scale quality by increasing training FLOPs, typically through increased parameterization, at the expense of a higher memory footprint, or data. A potential alternative is looped architectures, which instead increase FLOPs by sending activations through a block of layers in a loop. While promising, existing recipes for training looped architectures can be unstable, suffering from residual explosion and loss spikes. We address these challenges by recasting looping as a nonlinear time-variant dynamical system over the residual stream. Via a linear approximation to this system, we find that instability occurs in existing looped architectures as a result of large spectral norms in their injection parameters. To address these instability issues, we propose Parcae, a novel stable, looped architecture that constrains the spectral norm of the injection parameters via discretization of a negative diagonal parameterization. As a result, Parcae achieves up to 6.3% lower validation perplexity over prior large-scale looped models. Using our stable looped architecture, we investigate the scaling properties of looping as a medium to improve quality by increasing FLOPs in training and test-time. For training, we derive predictable power laws to scale FLOPs while keeping parameter count fixed. Our initial scaling laws suggest that looping and data should be increased in tandem, given a fixed FLOP budget. At test-time, we find that Parcae can use looping to scale compute, following a predictable, saturating exponential decay. When scaled up to 1.3B parameters, we find that Parcae improves CORE and Core-Extended quality by 2.99 and 1.18 points when compared to strong Transformer baselines under a fixed parameter and data budget, achieving a relative quality of up to 87.5% a Transformer twice the size. --- ### Adaptive Data Dropout: Towards Self-Regulated Learning in Deep Neural Networks Authors: Amar Gahir, Varshil Patel, Shreyank N Gowda Published: 2026-04-14 Categories: cs.LG, cs.CV arXiv: https://arxiv.org/abs/2604.12945v1 Deep neural networks are typically trained by uniformly sampling large datasets across epochs, despite evidence that not all samples contribute equally throughout learning. Recent work shows that progressively reducing the amount of training data can improve efficiency and generalization, but existing methods rely on fixed schedules that do not adapt during training. In this work, we propose Adaptive Data Dropout, a simple framework that dynamically adjusts the subset of training data based on performance feedback. Inspired by self-regulated learning, our approach treats data selection as an adaptive process, increasing or decreasing data exposure in response to changes in training accuracy. We introduce a lightweight stochastic update mechanism that modulates the dropout schedule online, allowing the model to balance exploration and consolidation over time. Experiments on standard image classification benchmarks show that our method reduces effective training steps while maintaining competitive accuracy compared to static data dropout strategies. These results highlight adaptive data selection as a promising direction for efficient and robust training. Code will be released. --- ### Distorted or Fabricated? A Survey on Hallucination in Video LLMs Authors: Yiyang Huang, Yitian Zhang, Yizhou Wang, Mingyuan Zhang, Liang Shi, Huimin Zeng, Yun Fu Published: 2026-04-14 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2604.12944v1 Despite significant progress in video-language modeling, hallucinations remain a persistent challenge in Video Large Language Models (Vid-LLMs), referring to outputs that appear plausible yet contradict the content of the input video. This survey presents a comprehensive analysis of hallucinations in Vid-LLMs and introduces a systematic taxonomy that categorizes them into two core types: dynamic distortion and content fabrication, each comprising two subtypes with representative cases. Building on this taxonomy, we review recent advances in the evaluation and mitigation of hallucinations, covering key benchmarks, metrics, and intervention strategies. We further analyze the root causes of dynamic distortion and content fabrication, which often result from limited capacity for temporal representation and insufficient visual grounding. These insights inform several promising directions for future work, including the development of motion-aware visual encoders and the integration of counterfactual learning techniques. This survey consolidates scattered progress to foster a systematic understanding of hallucinations in Vid-LLMs, laying the groundwork for building robust and reliable video-language systems. An up-to-date curated list of related works is maintained at https://github.com/hukcc/Awesome-Video-Hallucination . --- ### Direct Discrepancy Replay: Distribution-Discrepancy Condensation and Manifold-Consistent Replay for Continual Face Forgery Detection Authors: Tianshuo Zhang, Haoyuan Zhang, Siran Peng, Weisong Zhao, Xiangyu Zhu, Zhen Lei Published: 2026-04-14 Categories: cs.CV arXiv: https://arxiv.org/abs/2604.12941v1 Continual face forgery detection (CFFD) requires detectors to learn emerging forgery paradigms without forgetting previously seen manipulations. Existing CFFD methods commonly rely on replaying a small amount of past data to mitigate forgetting. Such replay is typically implemented either by storing a few historical samples or by synthesizing pseudo-forgeries from detector-dependent perturbations. Under strict memory budgets, the former cannot adequately cover diverse forgery cues and may expose facial identities, while the latter remains strongly tied to past decision boundaries. We argue that the core role of replay in CFFD is to reinstate the distributions of previous forgery tasks during subsequent training. To this end, we directly condense the discrepancy between real and fake distributions and leverage real faces from the current stage to perform distribution-level replay. Specifically, we introduce Distribution-Discrepancy Condensation (DDC), which models the real-to-fake discrepancy via a surrogate factorization in characteristic-function space and condenses it into a tiny bank of distribution discrepancy maps. We further propose Manifold-Consistent Replay (MCR), which synthesizes replay samples through variance-preserving composition of these maps with current-stage real faces, yielding samples that reflect previous-task forgery cues while remaining compatible with current real-face statistics. Operating under an extremely small memory budget and without directly storing raw historical face images, our framework consistently outperforms prior CFFD baselines and significantly mitigates catastrophic forgetting. Replay-level privacy analysis further suggests reduced identity leakage risk relative to selection-based replay. --- ### Task Alignment: A Simple Proxy for Practical Model Merging Across Diverse Vision Tasks Authors: Pau de Jorge, César Roberto de Souza, Björn Michele, Mert Bülent Sarıyıldız, Philippe Weinzaepfel, Florent Perronnin, Diane Larlus, Yannis Kalantidis Published: 2026-04-14 Categories: cs.CV arXiv: https://arxiv.org/abs/2604.12935v3 Efficiently merging several models fine-tuned for different tasks, but stemming from the same pretrained base model, is of great practical interest. Despite extensive prior work, most evaluations of model merging in computer vision are restricted to image classification using CLIP, where different classification datasets define different tasks. In this work, our goal is to make model merging more practical and show its relevance on challenging scenarios beyond this specific setting. In most vision scenarios, different tasks rely on trainable and usually heterogeneous decoders. Differently from previous studies with frozen decoders, where merged models can be evaluated right away, the non-trivial cost of decoder training renders hyperparameter selection based on downstream performance impractical. To address this, we introduce the task alignment proxy, and show how it can be used to speed up hyperparameter selection by orders of magnitude while retaining performance. Equipped with the task alignment proxy, we extend the applicability of model merging to multi-task vision models beyond CLIP-based classification. Project page: https://europe.naverlabs.com/task-alignment --- ### Task Alignment: A Simple Proxy for Practical Model Merging Across Diverse Vision Tasks Authors: Pau de Jorge, César Roberto de Souza, Björn Michele, Mert Bülent Sarıyıldız, Philippe Weinzaepfel, Florent Perronnin, Diane Larlus, Yannis Kalantidis Published: 2026-04-14 Categories: cs.CV arXiv: https://arxiv.org/abs/2604.12935v2 Efficiently merging several models fine-tuned for different tasks, but stemming from the same pretrained base model, is of great practical interest. Despite extensive prior work, most evaluations of model merging in computer vision are restricted to image classification using CLIP, where different classification datasets define different tasks. In this work, our goal is to make model merging more practical and show its relevance on challenging scenarios beyond this specific setting. In most vision scenarios, different tasks rely on trainable and usually heterogeneous decoders. Differently from previous studies with frozen decoders, where merged models can be evaluated right away, the non-trivial cost of decoder training renders hyperparameter selection based on downstream performance impractical. To address this, we introduce the task alignment proxy, and show how it can be used to speed up hyperparameter selection by orders of magnitude while retaining performance. Equipped with the task alignment proxy, we extend the applicability of model merging to multi-task vision models beyond CLIP-based classification. Project page: https://europe.naverlabs.com/task-alignment --- ### DINO-Explorer: Active Underwater Discovery via Ego-Motion Compensated Semantic Predictive Coding Authors: Yuhan Jin, Nayari Marie Lessa, Mariela De Lucas Alvarez, Melvin Laux, Lucas Amparo Barbosa, Frank Kirchner, Rebecca Adam Published: 2026-04-14 Categories: cs.RO, cs.CV arXiv: https://arxiv.org/abs/2604.12933v1 Marine ecosystem degradation necessitates continuous, scientifically selective underwater monitoring. However, most autonomous underwater vehicles (AUVs) operate as passive data loggers, capturing exhaustive video for offline review and frequently missing transient events of high scientific value. Transitioning to active perception requires a causal, online signal that highlights significant phenomena while suppressing maneuver-induced visual changes. We propose DINO-Explorer, a novelty-aware perception framework driven by a continuous semantic surprise signal. Operating within the latent space of a frozen DINOv3 foundation model, it leverages a lightweight, action-conditioned recurrent predictor to anticipate short-horizon semantic evolution. An efference-copy-inspired module utilizes globally pooled optical flow to discount self-induced visual changes without suppressing genuine environmental novelty. We evaluate this signal on the downstream task of asynchronous event triage under variant telemetry constraints. Results demonstrate that DINO-Explorer provides a robust, bandwidth-efficient attention mechanism. At a fixed operating point, the system retains 78.8% of post-discovery human-reviewer consensus events with a 56.8% trigger confirmation rate, effectively surfacing mission-relevant phenomena. Crucially, ego-motion conditioning suppresses 45.5% of false positives relative to an uncompensated surprise signal baseline. In a replay-side Pareto ablation study, DINO-Explorer robustly dominates the validated peak F1 versus telemetry bandwidth frontier, reducing telemetry bandwidth by 48.2% at the selected operating point while maintaining a 62.2% peak F1 score, successfully concentrating data transmission around human-verified novelty events. ---