# Research Papers — 2026-03 · bundle page 65 of 243 > Full-text markdown of papers #3201–3250 of 12123 published in 2026-03, newest first. AI-readable mirrors of the full 12123-paper 2026-03 archive (50 papers per file). Canonical page: https://aipod.dev/papers/archive/2026-03/page/65 [← Previous bundle](https://aipod.dev/papers/archive/2026-03/page/64.md) [Next bundle →](https://aipod.dev/papers/archive/2026-03/page/66.md) ### Noise Titration: Exact Distributional Benchmarking for Probabilistic Time Series Forecasting Authors: Qilin Wang Published: 2026-03-23 Categories: cs.LG, stat.ML arXiv: https://arxiv.org/abs/2603.22219v1 Modern time series forecasting is evaluated almost entirely through passive observation of single historical trajectories, rendering claims about a model's robustness to non-stationarity fundamentally unfalsifiable. We propose a paradigm shift toward interventionist, exact-statistical benchmarking. By systematically titrating calibrated Gaussian observation noise into known chaotic and stochastic dynamical systems, we transform forecasting from a black-box sequence matching game into an exact distributional inference task. Because the underlying data-generating process and noise variance are mathematically explicit, evaluation can rely on exact negative log-likelihoods and calibrated distributional tests rather than heuristic approximations. To fully leverage this framework, we extend the Fern architecture into a probabilistic generative model that natively parameterizes the Symmetric Positive Definite (SPD) cone, outputting calibrated joint covariance structures without the computational bottleneck of generic Jacobian modeling. Under this rigorous evaluation, we find that state-of-the-art zero-shot foundation models behave consistently with the context-parroting mechanism, failing systematically under non-stationary regime shifts and elevated noise. In contrast, Fern explicitly captures the invariant measure and multivariate geometry of the underlying dynamics, maintaining structural fidelity and statistically sharp calibration precisely where massive sequence-matching models collapse. --- ### Gumbel Distillation for Parallel Text Generation Authors: Chi Zhang, Xixi Hu, Bo Liu, Qiang Liu Published: 2026-03-23 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2603.22216v2 The slow, sequential nature of autoregressive (AR) language models has driven the adoption of parallel decoding methods. However, these non-AR models often sacrifice generation quality as they struggle to model the complex joint distribution of token sequences. To narrow this performance gap, we introduce Gumbel Distillation, a novel distillation technique that enables parallel decoders to learn this distribution effectively. Our method leverages the Gumbel-Max trick to create a deterministic mapping from a latent Gumbel noise space to the output tokens of a high-performing AR teacher. As a model-agnostic technique, Gumbel Distillation seamlessly integrates with diverse parallel decoding architectures, including MDLM and BD3-LM. Experiments on LM1B and OpenWebText show that Gumbel Distillation substantially improves the generation quality of parallel language models, achieving a 30.0% improvement in MAUVE score and 10.5% in generative perplexity over MDLM trained on OpenWebText dataset. Code available at https://github.com/hxixixh/gumbel-distill. --- ### Evaluating the Reliability and Fidelity of Automated Judgment Systems of Large Language Models Authors: Tom Biskupski, Stephan Kleber Published: 2026-03-23 Categories: cs.CR, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2603.22214v1 A Large Language Model (LLM) as judge evaluates the quality of victim Machine Learning (ML) models, specifically LLMs, by analyzing their outputs. An LLM as judge is the combination of one model and one specifically engineered judge prompt that contains the criteria for the analysis. The resulting automation of the analysis scales up the complex evaluation of the victim models' free-form text outputs by faster and more consistent judgments compared to human reviewers. Thus, quality and security assessments of LLMs can cover a wide range of the victim models' use cases. Being a comparably new technique, LLMs as judges lack a thorough investigation for their reliability and agreement to human judgment. Our work evaluates the applicability of LLMs as automated quality assessors of victim LLMs. We test the efficacy of 37 differently sized conversational LLMs in combination with 5 different judge prompts, the concept of a second-level judge, and 5 models fine-tuned for the task as assessors. As assessment objective, we curate datasets for eight different categories of judgment tasks and the corresponding ground-truth labels based on human assessments. Our empirical results show a high correlation of LLMs as judges with human assessments, when combined with a suitable prompt, in particular for GPT-4o, several open-source models with $\geqslant$ 32B parameters, and a few smaller models like Qwen2.5 14B. --- ### SPA: A Simple but Tough-to-Beat Baseline for Knowledge Injection Authors: Kexian Tang, Jiani Wang, Shaowen Wang, Kaifeng Lyu Published: 2026-03-23 Categories: cs.LG, cs.AI, cs.CL arXiv: https://arxiv.org/abs/2603.22213v2 While large language models (LLMs) are pretrained on massive amounts of data, their knowledge coverage remains incomplete in specialized, data-scarce domains, motivating extensive efforts to study synthetic data generation for knowledge injection. We propose SPA (Scaling Prompt-engineered Augmentation), a simple but tough-to-beat baseline that uses a small set of carefully designed prompts to generate large-scale synthetic data for knowledge injection. Through systematic comparisons, we find that SPA outperforms several strong baselines. Furthermore, we identify two key limitations of prior approaches: (1) while RL-based methods may improve the token efficiency of LLM-based data augmentation at small scale, they suffer from diversity collapse as data scales, leading to diminishing returns; and (2) while multi-stage prompting may outperform simple augmentation methods, their advantages can disappear after careful prompt tuning. Our results suggest that, for knowledge injection, careful prompt design combined with straightforward large-scale augmentation can be surprisingly effective, and we hope SPA can serve as a strong baseline for future studies in this area. Our code is available at https://github.com/Tangkexian/SPA. --- ### SPA: A Simple but Tough-to-Beat Baseline for Knowledge Injection Authors: Kexian Tang, Jiani Wang, Shaowen Wang, Kaifeng Lyu Published: 2026-03-23 Categories: cs.LG, cs.AI, cs.CL arXiv: https://arxiv.org/abs/2603.22213v1 While large language models (LLMs) are pretrained on massive amounts of data, their knowledge coverage remains incomplete in specialized, data-scarce domains, motivating extensive efforts to study synthetic data generation for knowledge injection. We propose SPA (Scaling Prompt-engineered Augmentation), a simple but tough-to-beat baseline that uses a small set of carefully designed prompts to generate large-scale synthetic data for knowledge injection. Through systematic comparisons, we find that SPA outperforms several strong baselines. Furthermore, we identify two key limitations of prior approaches: (1) while RL-based methods may improve the token efficiency of LLM-based data augmentation at small scale, they suffer from diversity collapse as data scales, leading to diminishing returns; and (2) while multi-stage prompting may outperform simple augmentation methods, their advantages can disappear after careful prompt tuning. Our results suggest that, for knowledge injection, careful prompt design combined with straightforward large-scale augmentation can be surprisingly effective, and we hope SPA can serve as a strong baseline for future studies in this area. Our code is available at https://github.com/Tangkexian/SPA. --- ### Omni-WorldBench: Towards a Comprehensive Interaction-Centric Evaluation for World Models Authors: Meiqi Wu, Zhixin Cai, Fufangchen Zhao, Xiaokun Feng, Rujing Dang, Bingze Song, Ruitian Tian, Jiashu Zhu, Jiachen Lei, Hao Dou, Jing Tang, Lei Sun et al. Published: 2026-03-23 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.22212v1 Video--based world models have emerged along two dominant paradigms: video generation and 3D reconstruction. However, existing evaluation benchmarks either focus narrowly on visual fidelity and text--video alignment for generative models, or rely on static 3D reconstruction metrics that fundamentally neglect temporal dynamics. We argue that the future of world modeling lies in 4D generation, which jointly models spatial structure and temporal evolution. In this paradigm, the core capability is interactive response: the ability to faithfully reflect how interaction actions drive state transitions across space and time. Yet no existing benchmark systematically evaluates this critical dimension. To address this gap, we propose Omni--WorldBench, a comprehensive benchmark specifically designed to evaluate the interactive response capabilities of world models in 4D settings. Omni--WorldBench comprises two key components: Omni--WorldSuite, a systematic prompt suite spanning diverse interaction levels and scene types; and Omni--Metrics, an agent-based evaluation framework that quantifies world modeling capabilities by measuring the causal impact of interaction actions on both final outcomes and intermediate state evolution trajectories. We conduct extensive evaluations of 18 representative world models across multiple paradigms. Our analysis reveals critical limitations of current world models in interactive response, providing actionable insights for future research. Omni-WorldBench will be publicly released to foster progress in interactive 4D world modeling. --- ### Identification of physiological shock in intensive care units via Bayesian regime switching models Authors: Emmett B. Kendall, Jonathan P. Williams, Curtis B. Storlie, Misty A. Radosevich, Erica D. Wittwer, Matthew A. Warner Published: 2026-03-23 Categories: stat.AP, stat.ME, stat.ML, stat.OT arXiv: https://arxiv.org/abs/2603.22208v2 Detection of occult hemorrhage (i.e., internal bleeding) in patients in intensive care units (ICUs) can pose significant challenges for critical care workers. Because blood loss may not always be clinically apparent, clinicians rely on monitoring vital signs for specific trends indicative of a hemorrhage event. The inherent difficulties of diagnosing such an event can lead to late intervention by clinicians which has catastrophic consequences. Therefore, a methodology for early detection of hemorrhage has wide utility. We develop a Bayesian regime switching model (RSM) that analyzes trends in patients' vitals and labs to provide a probabilistic assessment of the underlying physiological state that a patient is in at any given time. This article is motivated by a comprehensive dataset we curated from Mayo Clinic of 33,924 real ICU patient encounters. Longitudinal response measurements are modeled as a vector autoregressive process conditional on all latent states up to the current time point, and the latent states follow a Markov process. We present a novel Bayesian sampling routine to learn the posterior probability distribution of the latent physiological states, as well as develop an approach to account for pre-ICU-admission physiological changes. A simulation and real case study illustrate the effectiveness of our approach. --- ### Chimera: Latency- and Performance-Aware Multi-agent Serving for Heterogeneous LLMs Authors: Kangqi Ni, Wenyue Hua, Xiaoxiang Shi, Jiang Guo, Shiyu Chang, Tianlong Chen Published: 2026-03-23 Categories: cs.LG arXiv: https://arxiv.org/abs/2603.22206v1 Multi-agent applications often execute complex tasks as multi-stage workflows, where each stage is an LLM call whose output becomes part of context for subsequent steps. Existing LLM serving systems largely assume homogeneous clusters with identical model replicas. This design overlooks the potential of heterogeneous deployments, where models of different sizes and capabilities enable finer trade-offs between latency and performance. However, heterogeneity introduces new challenges in scheduling across models with diverse throughput and performance. We present Chimera, a predictive scheduling system for multi-agent workflow serving on heterogeneous LLM clusters that jointly improves end-to-end latency and task performance. Chimera applies semantic routing to estimate per-model confidence scores for each request, predicts the total remaining output length of the workflow, and estimates per-model congestion using in-flight predicted token volumes for load balancing. We evaluate Chimera on representative agentic workflows for code generation and math reasoning using multiple heterogeneous LLM configurations. Across comparable settings, Chimera traces the best latency-performance frontier, reducing end-to-end latency by 1.2--2.4$\times$ and improving task performance by 8.0-9.5 percentage points on average over competitive baselines including vLLM. --- ### Mixture of Mini Experts: Overcoming the Linear Layer Bottleneck in Multiple Instance Learning Authors: Daniel Shao, Joel Runevic, Richard J. Chen, Drew F. K. Williamson, Ahrong Kim, Andrew H. Song, Faisal Mahmood Published: 2026-03-23 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.22198v1 Multiple Instance Learning (MIL) is the predominant framework for classifying gigapixel whole-slide images in computational pathology. MIL follows a sequence of 1) extracting patch features, 2) applying a linear layer to obtain task-specific patch features, and 3) aggregating the patches into a slide feature for classification. While substantial efforts have been devoted to optimizing patch feature extraction and aggregation, none have yet addressed the second point, the critical layer which transforms general-purpose features into task-specific features. We hypothesize that this layer constitutes an overlooked performance bottleneck and that stronger representations can be achieved with a low-rank transformation tailored to each patch's phenotype, yielding synergistic effects with any of the existing MIL approaches. To this end, we introduce MAMMOTH, a parameter-efficient, multi-head mixture of experts module designed to improve the performance of any MIL model with minimal alterations to the total number of parameters. Across eight MIL methods and 19 different classification tasks, we find that such task-specific transformation has a larger effect on performance than the choice of aggregation method. For instance, when equipped with MAMMOTH, even simple methods such as max or mean pooling attain higher average performance than any method with the standard linear layer. Overall, MAMMOTH improves performance in 130 of the 152 examined configurations, with an average $+3.8\%$ change in performance. Code is available at https://github.com/mahmoodlab/mammoth. --- ### CayleyPy-4: AI-Holography. Towards analogs of holographic string dualities for AI tasks Authors: A. Chervov, F. Levkovich-Maslyuk, A. Smolensky, F. Khafizov, I. Kiselev, D. Melnikov, I. Koltsov, S. Kudashev, D. Shiltsov, M. Obozov, S. Krymskii, V. Kirova et al. Published: 2026-03-23 Categories: hep-th, cs.AI, cs.LG, math.CO, math.GR arXiv: https://arxiv.org/abs/2603.22195v1 This is the fourth paper in the CayleyPy project, which applies AI methods to the exploration of large graphs. In this work, we suggest the existence of a new discrete version of holographic string dualities for this setup, and discuss their relevance to AI systems and mathematics. Many modern AI tasks -- such as those addressed by GPT-style language models or RL systems -- can be viewed as direct analogues of predicting particle trajectories on graphs. We investigate this problem for a large family of Cayley graphs, for which we show that surprisingly it admits a dual description in terms of discrete strings. We hypothesize that such dualities may extend to a range of AI systems where they can lead to more efficient computational approaches. In particular, string holographic images of states are proposed as natural candidates for data embeddings, motivated by the "complexity = volume" principle in AdS/CFT. For Cayley graphs of the symmetric group S_n, our results indicate that the corresponding dual objects are flat, planar polygons. The diameter of the graph is equal to the number of integer points inside the polygon scaled by n. Vertices of the graph can be mapped holographically to paths inside the polygon, and the usual graph distances correspond to the area under the paths, thus directly realising the "complexity = volume" paradigm. We also find evidence for continuous CFTs and dual strings in the large n limit. We confirm this picture and other aspects of the duality in a large initial set of examples. We also present new datasets (obtained by a combination of ML and conventional tools) which should be instrumental in establishing the duality for more general cases. --- ### PAM: A Pose-Appearance-Motion Engine for Sim-to-Real HOI Video Generation Authors: Mingju Gao, Kaisen Yang, Huan-ang Gao, Bohan Li, Ao Ding, Wenyi Li, Yangcheng Yu, Jinkun Liu, Shaocong Xu, Yike Niu, Haohan Chi, Hao Chen et al. Published: 2026-03-23 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.22193v3 Hand-object interaction (HOI) reconstruction and synthesis are becoming central to embodied AI and AR/VR. Yet, despite rapid progress, existing HOI generation research remains fragmented across three disjoint tracks: (1) pose-only synthesis that predicts MANO trajectories without producing pixels; (2) single-image HOI generation that hallucinates appearance from masks or 2D cues but lacks dynamics; and (3) video generation methods that require both the entire pose sequence and the ground-truth first frame as inputs, preventing true sim-to-real deployment. Inspired by the philosophy of Joo et al. (2018), we think that HOI generation requires a unified engine that brings together pose, appearance, and motion within one coherent framework. Thus we introduce PAM: a Pose-Appearance-Motion Engine for controllable HOI video generation. The performance of our engine is validated by: (1) On DexYCB, we obtain an FVD of 29.13 (vs. 38.83 for InterDyn), and MPJPE of 19.37 mm (vs. 30.05 mm for CosHand), while generating higher-resolution 480x720 videos compared to 256x256 and 256x384 baselines. (2) On OAKINK2, our full multi-condition model improves FVD from 68.76 to 46.31. (3) An ablation over input conditions on DexYCB shows that combining depth, segmentation, and keypoints consistently yields the best results. (4) For a downstream hand pose estimation task using SimpleHand, augmenting training with 3,400 synthetic videos (207k frames) allows a model trained on only 50% of the real data plus our synthetic data to match the 100% real baseline. --- ### A Backbone Benchmarking Study on Self-supervised Learning as a Auxiliary Task with Texture-based Local Descriptors for Face Analysis Authors: Shukesh Reddy, Abhijit Das Published: 2026-03-23 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.22190v1 In this work, we benchmark with different backbones and study their impact for self-supervised learning (SSL) as an auxiliary task to blend texture-based local descriptors into feature modelling for efficient face analysis. It is established in previous work that combining a primary task and a self-supervised auxiliary task enables more robust and discriminative representation learning. We employed different shallow to deep backbones for the SSL task of Masked Auto-Encoder (MAE) as an auxiliary objective to reconstruct texture features such as local patterns alongside the primary task in local pattern SSAT (L-SSAT), ensuring robust and unbiased face analysis. To expand the benchmark, we conducted a comprehensive comparative analysis across multiple model configurations within the proposed framework. To this end, we address the three research questions: "What is the role of the backbone in performance L-SSAT?", "What type of backbone is effective for different face analysis tasks?", and "Is there any generalized backbone for effective face analysis with L-SSAT?". Towards answering these questions, we provide a detailed study and experiments. The performance evaluation demonstrates that the backbone for the proposed method is highly dependent on the downstream task, achieving average accuracies of 0.94 on FaceForensics++, 0.87 on CelebA, and 0.88 on AffectNet. For consistency of feature representation quality and generalisation capability across various face analysis paradigms, including face attribute prediction, emotion classification, and deepfake detection, there is no unified backbone. --- ### Seeing is Improving: Visual Feedback for Iterative Text Layout Refinement Authors: Junrong Guo, Shancheng Fang, Yadong Qu, Hongtao Xie Published: 2026-03-23 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2603.22187v1 Recent advances in Multimodal Large Language Models (MLLMs) have enabled automated generation of structured layouts from natural language descriptions. Existing methods typically follow a code-only paradigm that generates code to represent layouts, which are then rendered by graphic engines to produce final images. However, they are blind to the rendered visual outcome, making it difficult to guarantee readability and aesthetics. In this paper, we identify visual feedback as a critical factor in layout generation and propose Visual Feedback Layout Model (VFLM), a self-improving framework that leverages visual feedback iterative refinement. VFLM is capable of performing adaptive reflective generation, which leverages visual information to reflect on previous issues and iteratively generates outputs until satisfactory quality is achieved. It is achieved through reinforcement learning with a visually grounded reward model that incorporates OCR accuracy. By rewarding only the final generated outcome, we can effectively stimulate the model's iterative and reflective generative capabilities. Experiments across multiple benchmarks show that VFLM consistently outperforms advanced MLLMs, existing layout models, and code-only baselines, establishing visual feedback as critical for design-oriented MLLMs. Our code and data are available at https://github.com/FolSpark/VFLM. --- ### Enhancing Document-Level Machine Translation via Filtered Synthetic Corpora and Two-Stage LLM Adaptation Authors: Ireh Kim, Tesia Sker, Chanwoo Kim Published: 2026-03-23 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2603.22186v1 In Machine Translation, Large Language Models (LLMs) have generally underperformed compared to conventional encoder-decoder systems and thus see limited adoption. However, LLMs excel at modeling contextual information, making them a natural fit for document-level translation tasks where coherence across sentences is crucial. Despite this potential, document-level MT with LLMs faces two key challenges: (1) the scarcity of large-scale, high-quality document-level parallel data; and (2) the propensity of LLMs to introduce hallucinations and omissions during generation. To address these challenges, we propose a two-stage fine-tuning strategy leveraging LLM-augmented document-level data. First, we augment data by converting summarization data into document-level parallel data using a LLM, and then filter it using multiple metrics, leveraging sacreBLEU, COMET, and LaBSE-based cosine similarity-to improve data quality. Finally, we employ a two-stage fine-tuning strategy: first fine-tuning on the abundant sentence-level MT resources, and then on the filtered document-level corpus. --- ### Revisiting Quantum Code Generation: Where Should Domain Knowledge Live? Authors: Oscar Novo, Oscar Bastidas-Jossa, Alberto Calvo, Antonio Peris, Carlos Kuchkovsky Published: 2026-03-23 Categories: cs.LG, quant-ph arXiv: https://arxiv.org/abs/2603.22184v1 Recent advances in large language models (LLMs) have enabled the automation of an increasing number of programming tasks, including code generation for scientific and engineering domains. In rapidly evolving software ecosystems such as quantum software development, where frameworks expose complex abstractions, a central question is how best to incorporate domain knowledge into LLM-based assistants while preserving maintainability as libraries evolve. In this work, we study specialization strategies for Qiskit code generation using the Qiskit-HumanEval benchmark. We compare a parameter-specialized fine-tuned baseline introduced in prior work against a range of recent general-purpose LLMs enhanced with retrieval-augmented generation (RAG) and agent-based inference with execution feedback. Our results show that modern general-purpose LLMs consistently outperform the parameter-specialized baseline. While the fine-tuned model achieves approximately 47% pass@1 on Qiskit-HumanEval, recent general-purpose models reach 60-65% under zero-shot and retrieval-augmented settings, and up to 85% for the strongest evaluated model when combined with iterative execution-feedback agents -representing an improvement of more than 20% over zero-shot general-purpose performance and more than 35% over the parameter-specialized baseline. Agentic execution feedback yields the most consistent improvements, albeit at increased runtime cost, while RAG provides modest and model-dependent gains. These findings indicate that performance gains can be achieved without domain-specific fine-tuning, instead relying on inference-time augmentation, thereby enabling a more flexible and maintainable approach to LLM-assisted quantum software development. --- ### MARCUS: An agentic, multimodal vision-language model for cardiac diagnosis and management Authors: Jack W O'Sullivan, Mohammad Asadi, Lennart Elbe, Akshay Chaudhari, Tahoura Nedaee, Francois Haddad, Ivan Lopez, Fang Cao, Michael Salerno, Li Fe-Fei, Ehsan Adeli, Rima Arnaout et al. Published: 2026-03-23 Categories: cs.AI arXiv: https://arxiv.org/abs/2603.22179v2 Cardiovascular disease remains the leading cause of global mortality, with progress hindered by human interpretation of complex cardiac tests. Current AI vision-language models are limited to single-modality inputs and are non-interactive. We present MARCUS (Multimodal Autonomous Reasoning and Chat for Ultrasound and Signals), an agentic vision-language system for end-to-end interpretation of electrocardiograms (ECGs), echocardiograms, and cardiac magnetic resonance imaging (CMR) independently and as multimodal input. MARCUS employs a hierarchical agentic architecture comprising modality-specific vision-language expert models, each integrating domain-trained visual encoders with multi-stage language model optimization, coordinated by a multimodal orchestrator. Trained on 13.5 million images (0.25M ECGs, 1.3M echocardiogram images, 12M CMR images) and our novel expert-curated dataset spanning 1.6 million questions, MARCUS achieves state-of-the-art performance surpassing frontier models (GPT-5 Thinking, Gemini 2.5 Pro Deep Think). Across internal (Stanford) and external (UCSF) test cohorts, MARCUS achieves accuracies of 87-91% for ECG, 67-86% for echocardiography, and 85-88% for CMR, outperforming frontier models by 34-45% (P<0.001). On multimodal cases, MARCUS achieved 70% accuracy, nearly triple that of frontier models (22-28%), with 1.7-3.0x higher free-text quality scores. Our agentic architecture also confers resistance to mirage reasoning, whereby vision-language models derive reasoning from unintended textual signals or hallucinated visual content. MARCUS demonstrates that domain-specific visual encoders with an agentic orchestrator enable multimodal cardiac interpretation. We release our models, code, and benchmark open-source. --- ### MARCUS: An agentic, multimodal vision-language model for cardiac diagnosis and management Authors: Jack W O'Sullivan, Mohammad Asadi, Lennart Elbe, Akshay Chaudhari, Tahoura Nedaee, Francois Haddad, Michael Salerno, Li Fe-Fei, Ehsan Adeli, Rima Arnaout, Euan A Ashley Published: 2026-03-23 Categories: cs.AI arXiv: https://arxiv.org/abs/2603.22179v1 Cardiovascular disease remains the leading cause of global mortality, with progress hindered by human interpretation of complex cardiac tests. Current AI vision-language models are limited to single-modality inputs and are non-interactive. We present MARCUS (Multimodal Autonomous Reasoning and Chat for Ultrasound and Signals), an agentic vision-language system for end-to-end interpretation of electrocardiograms (ECGs), echocardiograms, and cardiac magnetic resonance imaging (CMR) independently and as multimodal input. MARCUS employs a hierarchical agentic architecture comprising modality-specific vision-language expert models, each integrating domain-trained visual encoders with multi-stage language model optimization, coordinated by a multimodal orchestrator. Trained on 13.5 million images (0.25M ECGs, 1.3M echocardiogram images, 12M CMR images) and our novel expert-curated dataset spanning 1.6 million questions, MARCUS achieves state-of-the-art performance surpassing frontier models (GPT-5 Thinking, Gemini 2.5 Pro Deep Think). Across internal (Stanford) and external (UCSF) test cohorts, MARCUS achieves accuracies of 87-91% for ECG, 67-86% for echocardiography, and 85-88% for CMR, outperforming frontier models by 34-45% (P<0.001). On multimodal cases, MARCUS achieved 70% accuracy, nearly triple that of frontier models (22-28%), with 1.7-3.0x higher free-text quality scores. Our agentic architecture also confers resistance to mirage reasoning, whereby vision-language models derive reasoning from unintended textual signals or hallucinated visual content. MARCUS demonstrates that domain-specific visual encoders with an agentic orchestrator enable multimodal cardiac interpretation. We release our models, code, and benchmark open-source. --- ### Calibeating Made Simple Authors: Yurong Chen, Zhiyi Huang, Michael I. Jordan, Haipeng Luo Published: 2026-03-23 Categories: cs.LG, cs.AI, cs.GT, econ.TH arXiv: https://arxiv.org/abs/2603.22167v1 We study calibeating, the problem of post-processing external forecasts online to minimize cumulative losses and match an informativeness-based benchmark. Unlike prior work, which analyzed calibeating for specific losses with specific arguments, we reduce calibeating to existing online learning techniques and obtain results for general proper losses. More concretely, we first show that calibeating is minimax-equivalent to regret minimization. This recovers the $O(\log T)$ calibeating rate of Foster and Hart (FH23) for the Brier and log losses and its optimality, and yields new optimal calibeating rates for mixable losses and general bounded losses. Second, we prove that multi-calibeating is minimax-equivalent to the combination of calibeating and the classical expert problem. This yields new optimal multi-calibeating rates for mixable losses, including Brier and log losses, and general bounded losses. Finally, we obtain new bounds for achieving calibeating and calibration simultaneously for the Brier loss. For binary predictions, our result gives the first calibrated algorithm that at the same time also achieves the optimal $O(\log T)$ calibeating rate. --- ### ACPO: Counteracting Likelihood Displacement in Vision-Language Alignment with Asymmetric Constraints Authors: Kaili Huang, Hongming Zhang, Rui Shen, Linjun Dai, Jiahao Wang, Hanming Deng, Lewei Lu Published: 2026-03-23 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.22165v1 While Direct Preference Optimization (DPO) has become the de facto approach for aligning Large Vision-Language Models (LVLMs), it suffers from Likelihood Displacement, where the probability of both chosen and rejected responses collapses. This optimization flaw is especially detrimental in multimodal settings: the erosion of chosen likelihoods -- a failure we term Visual Anchor Collapse -- causes models to abandon visual evidence for strong language priors, precipitating significant hallucinations. To address this, we propose Asymmetric Constrained Preference Optimization (ACPO), a modality-agnostic alignment mechanism that applies dynamic, target-oriented scaling to preference optimization. ACPO derives a complexity-aware scaling coefficient applied exclusively to the rejected reward, asymmetrically suppressing the gradient flow on the rejected term while preserving the chosen distribution as a gradient-stable reference. While fundamentally a general-purpose objective, breaking this gradient symmetry is crucial for multimodal tasks, as it mitigates the suppression of visual tokens by language priors. Experiments on InternVL models demonstrate that ACPO effectively reverses the chosen-reward degradation of standard DPO. By halting Visual Anchor Collapse, ACPO generally outperforms baselines on hallucination benchmarks (HallusionBench, MM-IFEval) and general leaderboards (MMBench, MMStar, OCRBenchV2) while driving concurrent improvements in general capabilities. --- ### Causal Evidence that Language Models use Confidence to Drive Behavior Authors: Dharshan Kumaran, Nathaniel Daw, Simon Osindero, Petar Veličković, Viorica Patraucean Published: 2026-03-23 Categories: cs.LG arXiv: https://arxiv.org/abs/2603.22161v3 Metacognition -- assessing the quality of one's own cognitive performance -- guides adaptive behavior across species. Substantial research demonstrates that confidence signals can be extracted from language model outputs, yet a fundamental question remains: do models actually use these signals to control behavior, such as deciding whether to answer or abstain? To investigate, we developed a four-phase paradigm. Phase~1 elicited baseline confidence estimates without an abstention option. Phase~2 revealed that LLMs apply an implicit threshold to internal confidence when deciding to abstain, with confidence effect sizes approximately an order of magnitude larger than alternative mechanisms. Phase~3 provided direct causal evidence through activation steering: boosting or suppressing confidence signals correspondingly decreased or increased abstention rates. Phase~4 extended this by systematically varying instructed thresholds, demonstrating that LLMs actively deploy confidence signals to implement abstention policies. Critically, beyond calibrated log-probability based confidence derived from the output distribution, verbal confidence independently predicted abstention across all models, despite being objectively less discriminatory of answer correctness. Activation decoding at the last pre-answer token further showed that both observable measures are lossy readouts of a richer internal representation. Together, these results suggest that abstention is not fully captured by the strength of evidence in the output distribution alone, but is better explained by the joint operation of a multidimensional internal confidence representation and threshold-based policies -- consistent with structured metacognitive control in LLMs, a capacity of growing importance as models transition to autonomous agents that must recognize their own uncertainty. --- ### Causal Evidence that Language Models use Confidence to Drive Behavior Authors: Dharshan Kumaran, Nathaniel Daw, Simon Osindero, Petar Veličković, Viorica Patraucean Published: 2026-03-23 Categories: cs.LG arXiv: https://arxiv.org/abs/2603.22161v2 Metacognition -- assessing the quality of one's own cognitive performance -- guides adaptive behavior across species. Substantial research demonstrates that confidence signals can be extracted from language model outputs, yet a fundamental question remains: do models actually use these signals to control behavior, such as deciding whether to answer or abstain? To investigate, we developed a four-phase paradigm. Phase~1 elicited baseline confidence estimates without an abstention option. Phase~2 revealed that LLMs apply an implicit threshold to internal confidence when deciding to abstain, with confidence effect sizes approximately an order of magnitude larger than alternative mechanisms. Phase~3 provided direct causal evidence through activation steering: boosting or suppressing confidence signals correspondingly decreased or increased abstention rates. Phase~4 extended this by systematically varying instructed thresholds, demonstrating that LLMs actively deploy confidence signals to implement abstention policies. Critically, beyond calibrated log-probability based confidence derived from the output distribution, verbal confidence independently predicted abstention across all models, despite being objectively less discriminatory of answer correctness. Activation decoding at the last pre-answer token further showed that both observable measures are lossy readouts of a richer internal representation. Together, these results suggest that abstention is not fully captured by the strength of evidence in the output distribution alone, but is better explained by the joint operation of a multidimensional internal confidence representation and threshold-based policies -- consistent with structured metacognitive control in LLMs, a capacity of growing importance as models transition to autonomous agents that must recognize their own uncertainty. --- ### Data Curation for Machine Learning Interatomic Potentials by Determinantal Point Processes Authors: Joanna Zou, Youssef Marzouk Published: 2026-03-23 Categories: stat.AP, cs.LG arXiv: https://arxiv.org/abs/2603.22160v1 The development of machine learning interatomic potentials faces a critical computational bottleneck with the generation and labeling of useful training datasets. We present a novel application of determinantal point processes (DPPs) to the task of selecting informative subsets of atomic configurations to label with reference energies and forces from costly quantum mechanical methods. Through experiments with hafnium oxide data, we show that DPPs are competitive with existing approaches to constructing compact but diverse training sets by utilizing kernels of molecular descriptors, leading to improved accuracy and robustness in machine learning representations of molecular systems. Our work identifies promising directions to employ DPPs for unsupervised training data curation with heterogeneous or multimodal data, or in online active learning schemes for iterative data augmentation during molecular dynamics simulation. --- ### Multimodal Survival Analysis with Locally Deployable Large Language Models Authors: Moritz Gögl, Christopher Yau Published: 2026-03-23 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2603.22158v1 We study multimodal survival analysis integrating clinical text, tabular covariates, and genomic profiles using locally deployable large language models (LLMs). As many institutions face tight computational and privacy constraints, this setting motivates the use of lightweight, on-premises models. Our approach jointly estimates calibrated survival probabilities and generates concise, evidence-grounded prognosis text via teacher-student distillation and principled multimodal fusion. On a TCGA cohort, it outperforms standard baselines, avoids reliance on cloud services and associated privacy concerns, and reduces the risk of hallucinated or miscalibrated estimates that can be observed in base LLMs. --- ### RAMPAGE: RAndomized Mid-Point for debiAsed Gradient Extrapolation Authors: Zhankun Luo, M. Berk Sahin, Antesh Upadhyay, Behzad Sharif, Abolfazl Hashemi Published: 2026-03-23 Categories: cs.LG, math.OC arXiv: https://arxiv.org/abs/2603.22155v2 A celebrated method for Variational Inequalities (VIs) is Extragradient (EG), which can be viewed as a standard discrete-time integration scheme. With this view in mind, in this paper we show that EG may suffer from discretization bias when applied to non-linear vector fields, conservative or otherwise. To resolve this discretization shortcoming, we introduce RAndomized Mid-Point for debiAsed Gradient Extrapolation (RAMPAGE) and its variance-reduced counterpart, RAMPAGE+, which leverages antithetic sampling. In contrast with EG, both methods are unbiased. Furthermore, leveraging negative correlation, RAMPAGE+ acts as an unbiased, geometric path-integrator that completely removes internal first-order terms from the variance, provably improving upon RAMPAGE. We further demonstrate that both methods enjoy provable $\mathcal{O}(1/k)$ convergence guarantees for a range of problems including root finding under co-coercive, co-hypomonotone, and generalized Lipschitzness regimes. Furthermore, we introduce symmetrically scaled variants to extend our results to constrained VIs. Finally, we provide convergence guarantees of both methods for stochastic and deterministic smooth convex-concave games. Somewhat interestingly, despite being a randomized method, RAMPAGE+ attains purely deterministic bounds for a number of the studied settings. --- ### dynActivation: A Trainable Activation Family for Adaptive Nonlinearity Authors: Alois Bachmann Published: 2026-03-23 Categories: cs.LG, cs.CV arXiv: https://arxiv.org/abs/2603.22154v1 This paper proposes $\mathrm{dynActivation}$, a per-layer trainable activation defined as $f_i(x) = \mathrm{BaseAct}(x)(α_i - β_i) + β_i x$, where $α_i$ and $β_i$ are lightweight learned scalars that interpolate between the base nonlinearity and a linear path and $\mathrm{BaseAct}(x)$ resembles any ReLU-like function. The static and dynamic ReLU-like variants are then compared across multiple vision tasks, language modeling tasks, and ablation studies. The results suggest that dynActivation variants tend to linearize deep layers while maintaining high performance, which can improve training efficiency by up to $+54\%$ over ReLU. On CIFAR-10, dynActivation(Mish) improves over static Mish by up to $+14.02\%$ on AttentionCNN with an average improvment by $+6.00\%$, with a $24\%$ convergence-AUC reduction relative to Mish (2120 vs. 2785). In a 1-to-75-layer MNIST depth-scaling study, dynActivation never drops below $95\%$ test accuracy ($95.3$--$99.3\%$), while ReLU collapses below $80\%$ at 25 layers. Under FGSM at $\varepsilon{=}0.08$, dynActivation(Mish) incurs a $55.39\%$ accuracy drop versus $62.79\%$ for ReLU ($7.40\%$ advantage). Transferred to language modeling, a new proposed dynActGLU-variant achieves a $10.3\%$ relative perplexity reduction over SwiGLU at 5620 steps (4.047 vs. 4.514), though the gap vanishes at 34300 steps. --- ### Beyond Matching to Tiles: Bridging Unaligned Aerial and Satellite Views for Vision-Only UAV Navigation Authors: Kejia Liu, Haoyang Zhou, Ruoyu Xu, Peicheng Wang, Mingli Song, Haofei Zhang Published: 2026-03-23 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2603.22153v3 Recent advances in cross-view geo-localization (CVGL) methods have shown strong potential for supporting unmanned aerial vehicle (UAV) navigation in GNSS-denied environments. However, existing work predominantly focuses on matching UAV views to onboard map tiles, which introduces an inherent trade-off between accuracy and storage overhead, and overlooks the importance of the UAV's heading during navigation. Moreover, the substantial discrepancies and varying overlaps in cross-view scenarios have been insufficiently considered, limiting their generalization to real-world scenarios. In this paper, we present Bearing-UAV, a purely vision-driven cross-view navigation method that jointly predicts UAV absolute location and heading from neighboring features, enabling accurate, lightweight, and robust navigation in the wild. Our method leverages global and local structural features and explicitly encodes relative spatial relationships, making it robust to cross-view variations, misalignment, and feature-sparse conditions. We also present Bearing-UAV-90k, a multi-city benchmark for evaluating cross-view localization and navigation. Extensive experiments show encouraging results that Bearing-UAV yields lower localization error than previous matching/retrieval paradigm across diverse terrains. Our code and dataset will be made publicly available. --- ### More Isn't Always Better: Balancing Decision Accuracy and Conformity Pressures in Multi-AI Advice Authors: Yuta Tsuchiya, Yukino Baba Published: 2026-03-23 Categories: cs.HC, cs.AI arXiv: https://arxiv.org/abs/2603.22152v1 Just as people improve decision-making by consulting diverse human advisors, they can now also consult with multiple AI systems. Prior work on group decision-making shows that advice aggregation creates pressure to conform, leading to overreliance. However, the conditions under which multi-AI consultation improves or undermines human decision-making remain unclear. We conducted experiments with three tasks in which participants received advice from panels of AIs. We varied panel size, within-panel consensus, and the human-likeness of presentation. Accuracy improved for small panels relative to a single AI; larger panels yielded no gains. The level of within-panel consensus affected participants' reliance on AI advice: High consensus fostered overreliance; a single dissent reduced pressure to conform; wide disagreement created confusion and undermined appropriate reliance. Human-like presentations increased perceived usefulness and agency in certain tasks, without raising conformity pressure. These findings yield design implications for presenting multi-AI advice that preserve accuracy while mitigating conformity. --- ### OpenEarth-Agent: From Tool Calling to Tool Creation for Open-Environment Earth Observation Authors: Sijie Zhao, Feng Liu, Xueliang Zhang, Hao Chen, Xinyu Gu, Zhe Jiang, Fenghua Ling, Ben Fei, Wenlong Zhang, Junjue Wang, Weihao Xuan, Pengfeng Xiao et al. Published: 2026-03-23 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.22148v1 Earth Observation (EO) is essential for perceiving dynamic land surface changes, yet deploying autonomous EO in open environments is hindered by the immense diversity of multi-source data and heterogeneous tasks. While remote sensing agents have emerged to streamline EO workflows, existing tool-calling agents are confined to closed environments. They rely on pre-defined tools and are restricted to narrow scope, limiting their generalization to the diverse data and tasks. To overcome these limitations, we introduce OpenEarth-Agent, the first tool-creation agent framework tailored for open-environment EO. Rather than calling predefined tools, OpenEarth-Agent employs adaptive workflow planning and tool creation to generalize to unseen data and tasks. This adaptability is bolstered by an open-ended integration of multi-stage tools and cross-domain knowledge bases, enabling robust execution in the entire EO pipeline across multiple application domains. To comprehensively evaluate EO agents in open environments, we propose OpenEarth-Bench, a novel benchmark comprising 596 real-world, full-pipeline cases across seven application domains, explicitly designed to assess agents' adaptive planning and tool creation capabilities. Only essential pre-trained model tools are provided in this benchmark, devoid of any other predefined task-specific tools. Extensive experiments demonstrate that OpenEarth-Agent successfully masters full-pipeline EO across multiple domains in the open environment. Notably, on the cross-benchmark Earth-Bench, our tool-creating agent equipped with 6 essential pre-trained models achieves performance comparable to tool-calling agents relying on 104 specialized tools, and significantly outperforms them when provided with the complete toolset. In several cases, the created tools exhibit superior robustness to data anomalies compared to human-engineered counterparts. --- ### The Semantic Ladder: A Framework for Progressive Formalization of Natural Language Content for Knowledge Graphs and AI Systems Authors: Lars Vogt Published: 2026-03-23 Categories: cs.CL, cs.DB arXiv: https://arxiv.org/abs/2603.22136v1 Semantic data and knowledge infrastructures must reconcile two fundamentally different forms of representation: natural language, in which most knowledge is created and communicated, and formal semantic models, which enable machine-actionable integration, interoperability, and reasoning. Bridging this gap remains a central challenge, particularly when full semantic formalization is required at the point of data entry. Here, we introduce the Semantic Ladder, an architectural framework that enables the progressive formalization of data and knowledge. Building on the concept of modular semantic units as identifiable carriers of meaning, the framework organizes representations across levels of increasing semantic explicitness, ranging from natural language text snippets to ontology-based and higher-order logical models. Transformations between levels support semantic enrichment, statement structuring, and logical modelling while preserving semantic continuity and traceability. This approach enables the incremental construction of semantic knowledge spaces, reduces the semantic parsing burden, and supports the integration of heterogeneous representations, including natural language, structured semantic models, and vector-based embeddings. The Semantic Ladder thereby provides a foundation for scalable, interoperable, and AI-ready data and knowledge infrastructures. --- ### Computationally lightweight classifiers with frequentist bounds on predictions Authors: Shreeram Murali, Cristian R. Rojas, Dominik Baumann Published: 2026-03-23 Categories: cs.LG, stat.ML arXiv: https://arxiv.org/abs/2603.22128v2 While both classical and neural network classifiers can achieve high accuracy, they fall short on offering uncertainty bounds on their predictions, making them unfit for safety-critical applications. Existing kernel-based classifiers that provide such bounds scale with $\mathcal O (n^{\sim3})$ in time, making them computationally intractable for large datasets. To address this, we propose a novel, computationally efficient classification algorithm based on the Nadaraya-Watson estimator, for whose estimates we derive frequentist uncertainty intervals. We evaluate our classifier on synthetically generated data and on electrocardiographic heartbeat signals from the MIT-BIH Arrhythmia database. We show that the method achieves competitive accuracy $>$\SI{96}{\percent} at $\mathcal O(n)$ and $\mathcal O(\log n)$ operations, while providing actionable uncertainty bounds. These bounds can, e.g., aid in flagging low-confidence predictions, making them suitable for real-time settings with resource constraints, such as diagnostic monitoring or implantable devices. --- ### DA-VAE: Plug-in Latent Compression for Diffusion via Detail Alignment Authors: Xin Cai, Zhiyuan You, Zhoutong Zhang, Tianfan Xue Published: 2026-03-23 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.22125v1 Reducing token count is crucial for efficient training and inference of latent diffusion models, especially at high resolution. A common strategy is to build high-compression image tokenizers with more channels per token. However, when trained only for reconstruction, high-dimensional latent spaces often lose meaningful structure, making diffusion training harder. Existing methods address this with extra objectives such as semantic alignment or selective dropout, but usually require costly diffusion retraining. Pretrained diffusion models, however, already exhibit a structured, lower-dimensional latent space; thus, a simpler idea is to expand the latent dimensionality while preserving this structure. We therefore propose \textbf{D}etail-\textbf{A}ligned VAE, which increases the compression ratio of a pretrained VAE with only lightweight adaptation of the pretrained diffusion backbone. DA-VAE uses an explicit latent layout: the first $C$ channels come directly from the pretrained VAE at a base resolution, while an additional $D$ channels encode higher-resolution details. A simple detail-alignment mechanism encourages the expanded latent space to retain the structure of the original one. With a warm-start fine-tuning strategy, our method enables $1024 \times 1024$ image generation with Stable Diffusion 3.5 using only $32 \times 32$ tokens, $4\times$ fewer than the original model, within 5 H100-days. It further unlocks $2048 \times 2048$ generation with SD3.5, achieving a $6\times$ speedup while preserving image quality. We also validate the method and its design choices quantitatively on ImageNet. --- ### Biophysics-Enhanced Neural Representations for Patient-Specific Respiratory Motion Modeling Authors: Jan Boysen, Hristina Uzunova, Heinz Handels, Jan Ehrhardt Published: 2026-03-23 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.22123v1 A precise spatial delivery of the radiation dose is crucial for the treatment success in radiotherapy. In the lung and upper abdominal region, respiratory motion introduces significant treatment uncertainties, requiring special motion management techniques. To address this, respiratory motion models are commonly used to infer the patient-specific respiratory motion and target the dose more efficiently. In this work, we investigate the possibility of using implicit neural representations (INR) for surrogate-based motion modeling. Therefore, we propose physics-regularized implicit surrogate-based modeling for respiratory motion (PRISM-RM). Our new integrated respiratory motion model is free of a fixed reference breathing state. Unlike conventional pairwise registration techniques, our approach provides a trajectory-aware spatio-temporally continuous and diffeomorphic motion representation, improving generalization to extrapolation scenarios. We introduce biophysical constraints, ensuring physiologically plausible motion estimation across time beyond the training data. Our results show that our trajectory-aware approach performs on par in interpolation and improves the extrapolation ability compared to our initially proposed INR-based approach. Compared to sequential registration-based approaches both our approaches perform equally well in interpolation, but underperform in extrapolation scenarios. However, the methodical features of INRs make them particularly effective for respiratory motion modeling, and with their performance steadily improving, they demonstrate strong potential for advancing this field. --- ### GenSpan: Generation-Calibrated Motion Span Priors for Multi-Verb Video Corpus Moment Retrieval Authors: Yunzhuo Sun, Xinyue Liu, Yanyang Li, Nanding Wu, Linlin Zong, Xianchao Zhang, Wenxin Liang Published: 2026-03-23 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2603.22121v2 Video Corpus Moment Retrieval (VCMR) aims to retrieve both the correct video and its temporal segment corresponding to a natural-language query, a task that is especially challenging for multi-verb queries where temporal action ordering is critical. Existing approaches often rely solely on text or static images and struggle to capture implicit motion dynamics, leading to retrieval errors and temporal misalignment. We propose GenSpan, a generation-calibrated VCMR framework that constructs short auxiliary videos from LLM-selected subtitle cues and decomposed sub-events, using these as temporal priors rather than direct retrieval targets. A token selector filters candidate-video features aligned with generated motion, and a bidirectional state-space model efficiently predicts video-moment tuples. Experiments on TVR and ActivityNet-Captions demonstrate that GenSpan improves corpus-level retrieval and moment localization, particularly for complex multi-action queries, while reducing computational cost compared to state-of-the-art multimodal baselines. --- ### StreamingClaw Technical Report Authors: Jiawei Chen, Zhe Chen, Chaoqun Du, Maokui He, Wei He, Hengtao Li, Qizhen Li, Zide Liu, Hao Ma, Xuhao Pan, Chang Ren, Xudong Rao et al. Published: 2026-03-23 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.22120v2 Emerging applications such as embodied intelligence, AI hardware, autonomous driving, and intelligent cockpits rely on a real-time perception-decision-action closed loop, posing stringent challenges for streaming video understanding. However, current agents mostly suffer from fragmented capabilities, such as supporting only offline video understanding, lacking long-term multimodal memory mechanisms, or struggling to achieve real-time reasoning and proactive interaction under streaming input. These shortcomings have become a key bottleneck for preventing agents from sustaining perception, making real-time decisions, and executing closed-loop actions in complex real-world environments, constraining their deployment and potential in dynamic, open physical worlds. To alleviate these issues, we propose StreamingClaw, a unified agent framework for streaming video understanding and embodied intelligence. Beyond maintaining full compatibility with the OpenClaw framework, it natively supports real-time, multimodal streaming interactions. StreamingClaw integrates five core capabilities: (1) It supports real-time streaming reasoning. (2) It supports reasoning about future events and proactive interaction under the online evolution of interaction objectives. (3) It supports multimodal long-term memory storage, hierarchical memory evolution, efficient memory retrieval, and memory sharing across multiple agents. (4) It supports a closed loop of perception-decision-action. In addition to conventional tools and skills, it also provides streaming tools and action-centric skills tailored for real-world physical environments. (5) It is compatible with the OpenClaw framework, allowing it to leverage the resources and support of the open-source community. --- ### On the Direction of RLVR Updates for LLM Reasoning: Identification and Exploitation Authors: Kexin Huang, Haoming Meng, Junkang Wu, Jinda Lu, Chiyu Ma, Ziqian Chen, Xue Wang, Bolin Ding, Jiancan Wu, Xiang Wang, Xiangnan He, Guoyin Wang et al. Published: 2026-03-23 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2603.22117v1 Reinforcement learning with verifiable rewards (RLVR) has substantially improved the reasoning capabilities of large language models. While existing analyses identify that RLVR-induced changes are sparse, they primarily focus on the \textbf{magnitude} of these updates, largely overlooking their \textbf{direction}. In this work, we argue that the direction of updates is a more critical lens for understanding RLVR's effects, which can be captured by the signed, token-level log probability difference $Δ\log p$ between the base and final RLVR models. Through statistical analysis and token-replacement interventions, we demonstrate that $Δ\log p$ more effectively identifies sparse, yet reasoning-critical updates than magnitude-based metrics (\eg divergence or entropy). Building on this insight, we propose two practical applications: (1) a \textit{test-time extrapolation} method that amplifies the policy along the learned $Δ\log p$ direction to improve reasoning accuracy without further training; (2) a \textit{training-time reweighting} method that focuses learning on low-probability (corresponding to higher $Δ\log p$) tokens, which improves reasoning performance across models and benchmarks. Our work establishes the direction of change as a key principle for analyzing and improving RLVR. --- ### Multiperspectivity as a Resource for Narrative Similarity Prediction Authors: Max Upravitelev, Veronika Solopova, Jing Yang, Charlott Jakob, Premtim Sahitaj, Ariana Sahitaj, Vera Schmitt Published: 2026-03-23 Categories: cs.CL arXiv: https://arxiv.org/abs/2603.22103v1 Predicting narrative similarity can be understood as an inherently interpretive task: different, equally valid readings of the same text can produce divergent interpretations and thus different similarity judgments, posing a fundamental challenge for semantic evaluation benchmarks that encode a single ground truth. Rather than treating this multiperspectivity as a challenge to overcome, we propose to incorporate it in the decision making process of predictive systems. To explore this strategy, we created an ensemble of 31 LLM personas. These range from practitioners following interpretive frameworks to more intuitive, lay-style characters. Our experiments were conducted on the SemEval-2026 Task 4 dataset, where the system achieved an accuracy score of 0.705. Accuracy improves with ensemble size, consistent with Condorcet Jury Theorem-like dynamics under weakened independence. Practitioner personas perform worse individually but produce less correlated errors, yielding larger ensemble gains under majority voting. Our error analysis reveals a consistent negative association between gender-focused interpretive vocabulary and accuracy across all persona categories, suggesting either attention to dimensions not relevant for the benchmark or valid interpretations absent from the ground truth. This finding underscores the need for evaluation frameworks that account for interpretive plurality. --- ### FreeArtGS: Articulated Gaussian Splatting Under Free-moving Scenario Authors: Hang Dai, Hongwei Fan, Han Zhang, Duojin Wu, Jiyao Zhang, Hao Dong Published: 2026-03-23 Categories: cs.CV, cs.GR, cs.RO arXiv: https://arxiv.org/abs/2603.22102v1 The increasing demand for augmented reality and robotics is driving the need for articulated object reconstruction with high scalability. However, existing settings for reconstructing from discrete articulation states or casual monocular videos require non-trivial axis alignment or suffer from insufficient coverage, limiting their applicability. In this paper, we introduce FreeArtGS, a novel method for reconstructing articulated objects under free-moving scenario, a new setting with a simple setup and high scalability. FreeArtGS combines free-moving part segmentation with joint estimation and end-to-end optimization, taking only a monocular RGB-D video as input. By optimizing with the priors from off-the-shelf point-tracking and feature models, the free-moving part segmentation module identifies rigid parts from relative motion under unconstrained capture. The joint estimation module calibrates the unified object-to-camera poses and recovers joint type and axis robustly from part segmentation. Finally, 3DGS-based end-to-end optimization is implemented to jointly reconstruct visual textures, geometry, and joint angles of the articulated object. We conduct experiments on two benchmarks and real-world free-moving articulated objects. Experimental results demonstrate that FreeArtGS consistently excels in reconstructing free-moving articulated objects and remains highly competitive in previous reconstruction settings, proving itself a practical and effective solution for realistic asset generation. The project page is available at: https://freeartgs.github.io/ --- ### SpecTM: Spectral Targeted Masking for Trustworthy Foundation Models Authors: Syed Usama Imtiaz, Mitra Nasr Azadani, Nasrin Alamdari Published: 2026-03-23 Categories: cs.AI, cs.LG arXiv: https://arxiv.org/abs/2603.22097v2 Foundation models are now increasingly being developed for Earth observation (EO), yet they often rely on stochastic masking that do not explicitly enforce physics constraints; a critical trustworthiness limitation, in particular for predictive models that guide public health decisions. In this work, we propose SpecTM (Spectral Targeted Masking), a physics-informed masking design that encourages the reconstruction of targeted bands from cross-spectral context during pretraining. To achieve this, we developed an adaptable multi-task (band reconstruction, bio-optical index inference, and 8-day-ahead temporal prediction) self-supervised learning (SSL) framework that encodes spectrally intrinsic representations via joint optimization, and evaluated it on a downstream microcystin concentration regression model using NASA PACE hyperspectral imagery over Lake Erie. SpecTM achieves R^2 = 0.695 (current week) and R^2 = 0.620 (8-day-ahead) predictions surpassing all baseline models by (+34% (0.51 Ridge) and +99% (SVR 0.31)) respectively. Our ablation experiments show targeted masking improves predictions by +0.037 R^2 over random masking. Furthermore, it outperforms strong baselines with 2.2x superior label efficiency under extreme scarcity. SpecTM enables physics-informed representation learning across EO domains and improves the interpretability of foundation models. --- ### GSEM: Graph-based Self-Evolving Memory for Experience Augmented Clinical Reasoning Authors: Xiao Han, Yuzheng Fan, Sendong Zhao, Haochun Wang, Bing Qin Published: 2026-03-23 Categories: cs.AI arXiv: https://arxiv.org/abs/2603.22096v1 Clinical decision-making agents can benefit from reusing prior decision experience. However, many memory-augmented methods store experiences as independent records without explicit relational structure, which may introduce noisy retrieval, unreliable reuse, and in some cases even hurt performance compared to direct LLM inference. We propose GSEM (Graph-based Self-Evolving Memory), a clinical memory framework that organizes clinical experiences into a dual-layer memory graph, capturing both the decision structure within each experience and the relational dependencies across experiences, and supporting applicability-aware retrieval and online feedback-driven calibration of node quality and edge weights. Across MedR-Bench and MedAgentsBench with two LLM backbones, GSEM achieves the highest average accuracy among all baselines, reaching 70.90\% and 69.24\% with DeepSeek-V3.2 and Qwen3.5-35B, respectively. Code is available at https://github.com/xhan1022/gsem. --- ### Principled Steering via Null-space Projection for Jailbreak Defense in Vision-Language Models Authors: Xingyu Zhu, Beier Zhu, Shuo Wang, Junfeng Fang, Kesen Zhao, Hanwang Zhang, Xiangnan He Published: 2026-03-23 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.22094v2 As vision-language models (VLMs) are increasingly deployed in open-world scenarios, they can be easily induced by visual jailbreak attacks to generate harmful content, posing serious risks to model safety and trustworthy usage. Recent activation steering methods inject directional vectors into model activations during inference to induce refusal behaviors and have demonstrated effectiveness. However, a steering vector may both enhance refusal ability and cause over-refusal, thereby degrading model performance on benign inputs. Moreover, due to the lack of theoretical interpretability, these methods still suffer from limited robustness and effectiveness. To better balance safety and utility, we propose NullSteer, a null-space projected activation defense framework. Our method constructs refusal directions within model activations through a linear transformation: it maintains zero perturbation within the benign subspace while dynamically inducing refusal along potentially harmful directions, thereby theoretically achieving safety enhancement without impairing the model's general capabilities. Extensive experiments show that NullSteer significantly reduces harmful outputs under various jailbreak attacks (average ASR reduction over 15 percent on MiniGPT-4) while maintaining comparable performance to the original model on general benchmarks. --- ### P-Flow: Prompting Visual Effects Generation Authors: Rui Zhao, Mike Zheng Shou Published: 2026-03-23 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.22091v1 Recent advancements in video generation models have significantly improved their ability to follow text prompts. However, the customization of dynamic visual effects, defined as temporally evolving and appearance-driven visual phenomena like object crushing or explosion, remains underexplored. Prior works on motion customization or control mainly focus on low-level motions of the subject or camera, which can be guided using explicit control signals such as motion trajectories. In contrast, dynamic visual effects involve higher-level semantics that are more naturally suited for control via text prompts. However, it is hard and time-consuming for humans to craft a single prompt that accurately specifies these effects, as they require complex temporal reasoning and iterative refinement over time. To address this challenge, we propose P-Flow, a novel training-free framework for customizing dynamic visual effects in video generation without modifying the underlying model. By leveraging the semantic and temporal reasoning capabilities of vision-language models, P-Flow performs test-time prompt optimization, refining prompts based on the discrepancy between the visual effects of the reference video and the generated output. Through iterative refinement, the prompts evolve to better induce the desired dynamic effect in novel scenes. Experiments demonstrate that P-Flow achieves high-fidelity and diverse visual effect customization and outperforms other models on both text-to-video and image-to-video generation tasks. Code is available at https://github.com/showlab/P-Flow. --- ### A Context Engineering Framework for Improving Enterprise AI Agents based on Digital-Twin MDP Authors: Xi Yang, Aurelie Lozano, Naoki Abe, Bhavya, Saurabh Jha, Noah Zheutlin, Rohan R. Arora, Yu Deng, Daby M. Sow Published: 2026-03-23 Categories: cs.AI arXiv: https://arxiv.org/abs/2603.22083v1 Despite rapid progress in AI agents for enterprise automation and decision-making, their real-world deployment and further performance gains remain constrained by limited data quality and quantity, complex real-world reasoning demands, difficulties with self-play, and the lack of reliable feedback signals. To address these challenges, we propose a lightweight, model-agnostic framework for improving LLM-based enterprise agents via offline reinforcement learning (RL). The proposed Context Engineering via DT-MDP (DT-MDP-CE) framework comprises three key components: (1) A Digital-Twin Markov Decision Process (DT-MDP), which abstracts the agent's reasoning behavior as a finite MDP; (2) A robust contrastive inverse RL, which, armed with the DT-MDP, to efficiently estimate a well-founded reward function and induces policies from mixed-quality offline trajectories; and (3) RL-guided context engineering, which uses the policy obtained from the integrated process of (1) and (2), to improve the agent's decision-making behavior. As a case study, we apply the framework to a representative task in the enterprise-oriented domain of IT automation. Extensive experimental results demonstrate consistent and significant improvements over baseline agents across a wide range of evaluation settings, suggesting that the framework can generalize to other agents sharing similar characteristics in enterprise environments. --- ### Autoregressive vs. Masked Diffusion Language Models: A Controlled Comparison Authors: Caio Vicentino Published: 2026-03-23 Categories: cs.CL arXiv: https://arxiv.org/abs/2603.22075v1 We present a controlled empirical comparison between autoregressive (AR) and masked diffusion (MDLM) language models. Both models are trained on identical data (50M tokens from TinyStories), identical compute budget (20,000 steps, batch size 32, sequence length 512), and identical hardware (NVIDIA H100 80GB), isolating the generation paradigm as the sole variable. We report three findings. First, both paradigms achieve comparable training throughput (~50K tokens/second), with MDLM requiring only 4.7% more wall-clock time. Second, AR converges faster and begins overfitting by step 14,000, while MDLM converges more slowly and is still improving at step 20,000, suggesting different compute-optimal training regimes. Third, quantitative diversity analysis over 1,000 generated samples reveals a structural diversity-fluency trade-off: AR produces fluent but repetitive outputs (99.8% begin with the same word), while MDLM generates more diverse narratives (93.4% unique 5-word openings, higher Distinct-n, lower Self-BLEU), at the cost of occasional grammatical inconsistencies. All code, trained checkpoints, and data pipelines are released for reproducibility. --- ### MIHT: A Hoeffding Tree for Time Series Classification using Multiple Instance Learning Authors: Aurora Esteban, Amelia Zafra, Sebastián Ventura Published: 2026-03-23 Categories: cs.LG arXiv: https://arxiv.org/abs/2603.22074v1 Due to the prevalence of temporal data and its inherent dependencies in many real-world problems, time series classification is of paramount importance in various domains. However, existing models often struggle with series of variable length or high dimensionality. This paper introduces the MIHT (Multi-instance Hoeffding Tree) algorithm, an efficient model that uses multi-instance learning to classify multivariate and variable-length time series while providing interpretable results. The algorithm uses a novel representation of time series as "bags of subseries," together with an optimization process based on incremental decision trees that distinguish relevant parts of the series from noise. This methodology extracts the underlying concept of series with multiple variables and variable lengths. The generated decision tree is a compact, white-box representation of the series' concept, providing interpretability insights into the most relevant variables and segments of the series. Experimental results demonstrate MIHT's superiority, as it outperforms 11 state-of-the-art time series classification models on 28 public datasets, including high-dimensional ones. MIHT offers enhanced accuracy and interpretability, making it a promising solution for handling complex, dynamic time series data. --- ### PreferRec: Learning and Transferring Pareto Preferences for Multi-objective Re-ranking Authors: Wei Zhou, Wuyang Li, Junkai Ji, Xueliang Li, Wenjing Hong, Zexuan Zhu, Xing Tang, Xiuqiang He Published: 2026-03-23 Categories: cs.IR, cs.NE arXiv: https://arxiv.org/abs/2603.22073v1 Multi-objective re-ranking has become a critical component of modern multi-stage recommender systems, as it tasked to balance multiple conflicting objectives such as accuracy, diversity, and fairness. Existing multi-objective re-ranking methods typically optimize aggregate objectives at the item level using static or handcrafted preference weights. This design overlooks that users inherently exhibit Pareto-optimal preferences at the intent level, reflecting personalized trade-offs among objectives rather than fixed weight combinations. Moreover, most approaches treat re-ranking task for each user as an isolated problem, and repeatedly learn the preferences from scratch. Such a paradigm not only incurs high computational cost, but also ignores the fact that users often share similar preference trade-off structures across objectives. Inspired by the existence of homogeneous multi-objective optimization spaces where Pareto-optimal patterns are transferable, we propose PreferRec, a novel framework that explicitly models and transfers Pareto preferences across users. Specifically, PreferRec is built upon three tightly coupled components: Preference-Aware Pareto Learning aims to capture user intrinsic trade-offs among multiple conflicting objectives at the intent level. By learning Pareto preference representations from re-ranking populations, this component explicitly models how users prioritize different objectives under diverse contexts. Knowledge-Guided Transfer facilitates efficient cross-user knowledge transfer by distilling shared optimization patterns across homogeneous optimization spaces. The transferred knowledge is then used to guide solution selection and personalized re-ranking, biasing the optimization process toward high-quality regions of the Pareto front while preserving user-specific preference characteristics. --- ### Adapting Point Cloud Analysis via Multimodal Bayesian Distribution Learning Authors: Xingyu Zhu, Liang Yi, Shuo Wang, Wenbo Zhu, Yonglinag Wu, Beier Zhu, Hanwang Zhang Published: 2026-03-23 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.22070v2 Multimodal 3D vision-language models show strong generalization across diverse 3D tasks, but their performance still degrades notably under domain shifts. This has motivated recent studies on test-time adaptation (TTA), which enables models to adapt online using test-time data. Among existing TTA methods, cache-based mechanisms are widely adopted for leveraging previously observed samples in online prediction refinement. However, they store only limited historical information, leading to progressive information loss as the test stream evolves. In addition, their prediction logits are fused heuristically, making adaptation unstable. To address these limitations, we propose BayesMM, a Multimodal Bayesian Distribution Learning framework for test-time point cloud analysis. BayesMM models textual priors and streaming visual features of each class as Gaussian distributions: textual parameters are derived from semantic prompts, while visual parameters are updated online with arriving samples. The two modalities are fused via Bayesian model averaging, which automatically adjusts their contributions based on posterior evidence, yielding a unified prediction that adapts continually to evolving test-time data without training. Extensive experiments on multiple point cloud benchmarks demonstrate that BayesMM maintains robustness under distributional shifts, yielding over 4% average improvement. --- ### On the Failure of Topic-Matched Contrast Baselines in Multi-Directional Refusal Abliteration Authors: Valentin Petrov Published: 2026-03-23 Categories: cs.LG, cs.AI arXiv: https://arxiv.org/abs/2603.22061v1 Inasmuch as the removal of refusal behavior from instruction-tuned language models by directional abliteration requires the extraction of refusal-mediating directions from the residual stream activation space, and inasmuch as the construction of the contrast baseline against which harmful prompt activations are compared has been treated in the existing literature as an implementation detail rather than a methodological concern, the present work investigates whether a topically matched contrast baseline yields superior refusal directions. The investigation is carried out on the Qwen~3.5 2B model using per-category matched prompt pairs, per-class Self-Organizing Map extraction, and Singular Value Decomposition orthogonalization. It was found that topic-matched contrast produces no functional refusal directions at any tested weight level on any tested layer, while unmatched contrast on the same model, same extraction code, and same evaluation protocol achieves complete refusal elimination on six layers. The geometric analysis of the failure establishes that topic-matched subtraction cancels the dominant activation component shared between harmful and harmless prompts of the same subject, reducing the extracted direction magnitude below the threshold at which weight-matrix projection perturbs the residual stream. The implications for the design of contrast baselines in abliteration research are discussed. --- ### SpatialBoost: Enhancing Visual Representation through Language-Guided Reasoning Authors: Byungwoo Jeon, Dongyoung Kim, Huiwon Jang, Insoo Kim, Jinwoo Shin Published: 2026-03-23 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.22057v1 Despite the remarkable success of large-scale pre-trained image representation models (i.e., vision encoders) across various vision tasks, they are predominantly trained on 2D image data and therefore often fail to capture 3D spatial relationships between objects and backgrounds in the real world, constraining their effectiveness in many downstream applications. To address this, we propose SpatialBoost, a scalable framework that enhances the spatial awareness of existing pre-trained vision encoders by injecting 3D spatial knowledge expressed in linguistic descriptions. The core idea involves converting dense 3D spatial information from 2D images into linguistic expressions, which is then used to inject such spatial knowledge into vision encoders through a Large Language Model (LLM). To this end, we adopt a multi-turn Chain-of-Thought (CoT) reasoning process that progressively incorporates dense spatial knowledge and builds hierarchical spatial understanding. To validate effectiveness, we adapt SpatialBoost to state-of-the-art vision encoders such as DINOv3, and evaluate its performance gains on a wide range of benchmarks requiring both 3D perception and general vision abilities. For instance, SpatialBoost improves DINOv3 performance from 55.9 to 59.7 mIoU on ADE20K, achieving state-of-the-art performance with 3.8% gain over the pre-trained DINOv3. --- ### Dual-Space Knowledge Distillation with Key-Query Matching for Large Language Models with Vocabulary Mismatch Authors: Stella Eva Tsiapali, Cong-Thanh Do, Kate Knill Published: 2026-03-23 Categories: cs.CL arXiv: https://arxiv.org/abs/2603.22056v2 Large language models (LLMs) achieve state-of-the-art (SOTA) performance across language tasks, but are costly to deploy due to their size and resource demands. Knowledge Distillation (KD) addresses this by training smaller Student models to mimic larger Teacher models, improving efficiency without significant performance loss. Dual-Space Knowledge Distillation with Cross-Model Attention (DSKD-CMA) has emerged as a SOTA method for KD between LLMs with distinct tokenizers, yet its internal workings remain largely opaque. In this work, we systematically analyse the attention mechanism of DSKD-CMA through manual token alignment probing and heatmap visualisations, revealing both strengths and limitations. Building on this, we introduce a novel method, DSKD-CMA-GA, based on Generative Adversarial (GA) learning, to address the mismatched distributions between the keys and queries computed from distinct models. Experiments show modest but consistent ROUGE-L gains in text generation quality, particularly on out-of-distribution data (+0.37 on average), narrowing the gap between cross- and same-tokenizer KD. --- ### FontCrafter: High-Fidelity Element-Driven Artistic Font Creation with Visual In-Context Generation Authors: Wuyang Luo, Chengkai Tan, Chang Ge, Binye Hong, Su Yang, Yongjiu Ma Published: 2026-03-23 Categories: cs.CV arXiv: https://arxiv.org/abs/2603.22054v2 Artistic font generation aims to synthesize stylized glyphs based on a reference style. However, existing approaches suffer from limited style diversity and coarse control. In this work, we explore the potential of element-driven artistic font generation. Elements are the fundamental visual units of a font, serving as reference images for the desired style. Conceptually, we categorize elements into object elements (e.g., flowers or stones) with distinct structures and amorphous elements (e.g., flames or clouds) with unstructured textures. We introduce FontCrafter, an element-driven framework for font creation, and construct a large-scale dataset, ElementFont, which contains diverse element types and high-quality glyph images. However, achieving high-fidelity reconstruction of both texture and structure of reference elements remains challenging. To address this, we propose an in-context generation strategy that treats element images as visual context and uses an inpainting model to transfer element styles into glyph regions at the pixel level. To further control glyph shapes, we design a lightweight Context-aware Mask Adapter (CMA) that injects shape information. Moreover, a training-free attention redirection mechanism enables region-aware style control and suppresses stroke hallucination. In addition, edge repainting is applied to make boundaries more natural. Extensive experiments demonstrate that FontCrafter achieves strong zero-shot generation performance, particularly in preserving structural and textural fidelity, while also supporting flexible controls such as style mixture. ---