# Research Papers — 2026-03 · bundle page 93 of 243

> Full-text markdown of papers #4601–4650 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/93
[← Previous bundle](https://aipod.dev/papers/archive/2026-03/page/92.md)
[Next bundle →](https://aipod.dev/papers/archive/2026-03/page/94.md)

### BeamAgent: LLM-Aided MIMO Beamforming with Decoupled Intent Parsing and Alternating Optimization for Joint Site Selection and Precoding

Authors: Xiucheng Wang, Yue Zhang, Nan Cheng
Published: 2026-03-19
Categories: cs.IT, cs.LG, eess.SP, eess.SY
arXiv: https://arxiv.org/abs/2603.18855v1

Integrating large language models (LLMs) into wireless communication optimization is a promising yet challenging direction. Existing approaches either use LLMs as black-box solvers or code generators, tightly coupling them with numerical computation. However, LLMs lack the precision required for physical-layer optimization, and the scarcity of wireless training data makes domain-specific fine-tuning impractical. We propose BeamAgent, an LLM-aided MIMO beamforming framework that explicitly decouples semantic intent parsing from numerical optimization. The LLM serves solely as a semantic translator that converts natural language descriptions into structured spatial constraints. A dedicated gradient-based optimizer then jointly solves the discrete base station site selection and continuous precoding design through an alternating optimization algorithm. A scene-aware prompt enables grounded spatial reasoning without fine-tuning, and a multi-round interaction mechanism with dual-layer intent classification ensures robust constraint verification. A penalty-based loss function enforces dark-zone power constraints while releasing optimization degrees of freedom for bright-zone gain maximization. Experiments on a ray-tracing-based urban MIMO scenario show that BeamAgent achieves a bright-zone power of 84.0\,dB, outperforming exhaustive zero-forcing by 7.1 dB under the same dark-zone constraint. The end-to-end system reaches within 3.3 dB of the expert upper bound, with the full optimization completing in under 2 s on a laptop.

---

### Learn for Variation: Efficient AAV Trajectory Learning through a Differentiable Wireless World Model

Authors: Xiucheng Wang, Zhenye Chen, Nan Cheng, Zhisheng Yin, Xuemin Shen
Published: 2026-03-19
Categories: eess.SY, cs.LG
arXiv: https://arxiv.org/abs/2603.18853v3

Autonomous aerial vehicles (AAVs) enable data collection for sixth-generation Internet-of-Things networks, but their trajectories couple nonlinear wireless rates with long-horizon service progress. This paper views the evolution of AAV kinematics, channel state, and user backlog as a structured differentiable world model and develops Learn for Variation (L4V) to exploit that model efficiently. L4V replaces a discontinuous completion-time objective with a cumulative-backlog surrogate, unrolls the mission dynamics, and propagates pathwise sensitivities to a neural policy through the discrete adjoint recursion. The resulting derivative is exact conditional on a fixed exogenous-noise realization; stochastic expected-objective optimization still requires sampling. We show that the structured adjoint grows at most polynomially with the horizon and establish a stationary-point rate for fixed-step full-gradient descent under standard smoothness assumptions. The framework also learns shared OFDMA allocation under reparameterized shadowing and Rician fading, while distributional pretraining amortizes model-based optimization into forward-only deployment on unseen layouts. Paired stress tests cover channel-generator mismatch, noisy partial observations, a fixed-resource two-AAV extension, and a circular no-fly region. Code and configurations are available at https://github.com/UNIC-Lab/L4V-AAV. Against genetic-algorithm, DQN, A2C, DDPG, and differentiable model-predictive-control implementations, L4V reduces mission time by up to $65\%$, executes a default mission in $53$ ms, and completes all $60$ frozen-policy tests after pretraining on $1{,}600$ layouts.

---

### Learn for Variation: Variationally Guided AAV Trajectory Learning in Differentiable Environments

Authors: Xiucheng Wang, Zhenye Chen, Nan Cheng
Published: 2026-03-19
Categories: eess.SY, cs.LG
arXiv: https://arxiv.org/abs/2603.18853v2

Autonomous aerial vehicles (AAVs) empower sixth-generation (6G) Internet-of-Things (IoT) networks through mobility-driven data collection. However, conventional reward-driven reinforcement learning for AAV trajectory planning suffers from severe credit assignment issues and training instability, because sparse scalar rewards fail to capture the long-term and nonlinear effects of sequential movements. To address these challenges, this paper proposes Learn for Variation (L4V), a gradient-informed trajectory learning framework that replaces high-variance scalar reward signals with dense and analytically grounded policy gradients. Particularly, the coupled evolution of AAV kinematics, distance-dependent channel gains, and per-user data-collection progress is first unrolled into an end-to-end differentiable computational graph. Backpropagation through time then serves as a discrete adjoint solver, which propagates exact sensitivities from the cumulative mission objective to every control action and policy parameter. These structured gradients are used to train a deterministic neural policy with temporal smoothness regularization and gradient clipping. Extensive simulations demonstrate that L4V consistently outperforms representative baselines, including a genetic algorithm, DQN, A2C, and DDPG, in mission completion time, average transmission rate, and training cost

---

### HORNet: Task-Guided Frame Selection for Video Question Answering with Vision-Language Models

Authors: Xiangyu Bai, Bishoy Galoaa, Sarah Ostadabbas
Published: 2026-03-19
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.18850v1

Video question answering (VQA) with vision-language models (VLMs) depends critically on which frames are selected from the input video, yet most systems rely on uniform or heuristic sampling that cannot be optimized for downstream answering quality. We introduce \textbf{HORNet}, a lightweight frame selection policy trained with Group Relative Policy Optimization (GRPO) to learn which frames a frozen VLM needs to answer questions correctly. With fewer than 1M trainable parameters, HORNet reduces input frames by up to 99\% and VLM processing time by up to 93\%, while improving answer quality on short-form benchmarks (+1.7\% F1 on MSVD-QA) and achieving strong performance on temporal reasoning tasks (+7.3 points over uniform sampling on NExT-QA). We formalize this as Select Any Frames (SAF), a task that decouples visual input curation from VLM reasoning, and show that GRPO-trained selection generalizes better out-of-distribution than supervised and PPO alternatives. HORNet's policy further transfers across VLM answerers without retraining, yielding an additional 8.5\% relative gain when paired with a stronger model. Evaluated across six benchmarks spanning 341,877 QA pairs and 114.2 hours of video, our results demonstrate that optimizing \emph{what} a VLM sees is a practical and complementary alternative to optimizing what it generates while improving efficiency. Code is available at https://github.com/ostadabbas/HORNet.

---

### Towards Interpretable Foundation Models for Retinal Fundus Images

Authors: Samuel Ofosu Mensah, Camila Roa, Kerol Djoumessi, Philipp Berens
Published: 2026-03-19
Categories: cs.CV, cs.LG, stat.CO
arXiv: https://arxiv.org/abs/2603.18846v4

Foundation models are used to extract transferable representations from large amounts of unlabeled data, typically via self-supervised learning (SSL). However, many of these models rely on architectures that offer limited interpretability, a critical issue in high-stakes domains such as medical imaging. We propose DualIFM, a foundation model that is interpretable-by-design via a BagNet backbone whose small receptive fields generate class evidence maps that are faithful to the model's decision-making process. Additionally, DualIFM incorporates a $2D$ projection layer during pretraining that enables direct visualization of the representation space, providing a dataset-level view of the learned structure including meaningful clinical clusters as well as potential spurious correlations. We trained DualIFM on over 800,000 color fundus photographs from various sources to learn generalizable representations for different downstream tasks. Our model achieves performance comparable to RETFound, which has $16\times$ more parameters, while providing interpretable predictions on out-of-distribution data. These results suggest that large-scale SSL pretraining paired with inherent interpretability can lead to robust representations for retinal imaging. Code and pretrained models are available at github.com/berenslab/interpretable_FM.

---

### A Model Ensemble-Based Post-Processing Framework for Fairness-Aware Prediction

Authors: Zhouting Zhao, Tin Lok James Ng
Published: 2026-03-19
Categories: cs.LG, stat.ML
arXiv: https://arxiv.org/abs/2603.18838v1

Striking an optimal balance between predictive performance and fairness continues to be a fundamental challenge in machine learning. In this work, we propose a post-processing framework that facilitates fairness-aware prediction by leveraging model ensembling. Designed to operate independently of any specific model internals, our approach is widely applicable across various learning tasks, model architectures, and fairness definitions. Through extensive experiments spanning classification, regression, and survival analysis, we demonstrate that the framework effectively enhances fairness while maintaining, or only minimally affecting, predictive accuracy.

---

### Model Order Reduction of Cerebrovascular Hemodynamics Using POD_Galerkin and Reservoir Computing_based Approach

Authors: Rahul Halder, Arash Hajisharifi, Kabir Bakhshaei, Gianluigi Rozza
Published: 2026-03-19
Categories: math.NA, cs.LG
arXiv: https://arxiv.org/abs/2603.18837v1

We investigate model order reduction (MOR) strategies for simulating unsteady hemodynamics within cerebrovascular systems, contrasting a physics-based intrusive approach with a data-driven non-intrusive framework. High-fidelity 3D Computational Fluid Dynamics (CFD) snapshots of an idealised basilar artery bifurcation are first compressed into a low-dimensional latent space using Proper Orthogonal Decomposition (POD). We evaluate the performance of a POD-Galerkin (POD-G) model, which projects the Navier-Stokes equations onto the reduced basis, against a POD-Reservoir Computing (POD-RC) model that learns the temporal evolution of coefficients through a recurrent architecture. A multi-harmonic and multi-amplitude training signal is introduced to improve training efficiency. Both methodologies achieve computational speed-ups on the order of 10^2 to 10^3 compared to full-order simulations, demonstrating their potential as efficient and accurate surrogates for predicting flow quantities such as wall shear stress.

---

### Statistical Characteristic-Guided Denoising for Rapid High-Resolution Transmission Electron Microscopy Imaging

Authors: Hesong Li, Ziqi Wu, Ruiwen Shao, Ying Fu
Published: 2026-03-19
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.18834v1

High-Resolution Transmission Electron Microscopy (HRTEM) enables atomic-scale observation of nucleation dynamics, which boosts the studies of advanced solid materials. Nonetheless, due to the millisecond-scale rapid change of nucleation, it requires short-exposure rapid imaging, leading to severe noise that obscures atomic positions. In this work, we propose a statistical characteristic-guided denoising network, which utilizes statistical characteristics to guide the denoising process in both spatial and frequency domains. In the spatial domain, we present spatial deviation-guided weighting to select appropriate convolution operations for each spatial position based on deviation characteristic. In the frequency domain, we present frequency band-guided weighting to enhance signals and suppress noise based on band characteristics. We also develop an HRTEM-specific noise calibration method and generate a dataset with disordered structures and realistic HRTEM image noises. It can ensure the denoising performance of models on real images for nucleation observation. Experiments on synthetic and real data show our method outperforms the state-of-the-art methods in HRTEM image denoising, with effectiveness in the localization downstream task. Code will be available at https://github.com/HeasonLee/SCGN.

---

### Agent Control Protocol: Admission Control for Agent Actions

Authors: Marcelo Fernandez
Published: 2026-03-19
Categories: cs.CR, cs.AI
arXiv: https://arxiv.org/abs/2603.18829v10

Autonomous agents can produce harmful behavioral patterns from individually valid requests -- a threat class per-request policy evaluation cannot address, because stateless engines evaluate each request in isolation. We present ACP, a temporal admission control protocol enforcing behavioral properties over execution traces via static risk scoring combined with stateful signals (anomaly accumulation, cooldown) through a LedgerQuerier abstraction. ACP blocks execution based on deterministic, history-aware risk scoring -- not anomaly detection. Under a 500-request workload where every request is individually valid (RS=35), a stateless engine approves all 500; ACP limits autonomous execution to 2 out of 500 (0.4%), escalating after 3 actions and denying after 11. We identify a state-mixing vulnerability in ACP-RISK-2.0 (cross-context false denials) and introduce ACP-RISK-3.0, scoping anomaly signals to PatternKey(agentID, capability, resource). Decision evaluation: 739-832 ns (p50); throughput 1,720,000 req/s. Safety and liveness model-checked via TLA+ (11 invariants + 4 temporal properties, 0 violations) across 4,294,930,695 distinct states. We formalize deviation collapse -- enforcement active but never exercised due to upstream constraints -- and introduce Boundary Activation Rate (BAR) as its detection mechanism. An adversary suppressing BAR to 0.00 is detected via DeltaBAR before collapse (BAR_C=1.00). N coordinated agents accumulate risk independently; coordination window CW_appr=2N with zero deviation: activity scales linearly, preventing superlinear amplification. ACP is Paper 1 of a 6-paper Agent Governance Series: P0 -- atomic decision boundaries; P2 -- behavioral drift detection (IML); P3/4 -- governance structure, fair allocation, and irreducibility; P5 -- runtime execution validity (RAM, arXiv:2604.22898); P6 -- operationalization of RAM.

---

### Student views in AI Ethics and Social Impact

Authors: Tudor-Dan Mihoc, Manuela-Andreea Petrescu, Emilia-Loredana Pop
Published: 2026-03-19
Categories: cs.CY, cs.AI
arXiv: https://arxiv.org/abs/2603.18827v1

An investigation, from a gender perspective, of how students view the ethical implications and societal effects of artificial intelligence is conducted, examining concepts that could have a big influence on how artificial intelligence may be taught in the future. For this, we conducted a survey on a cohort of 230 second year computer science students to reveal their opinions. The results revealed that AI, from the students' perspective, will significantly impact daily life, particularly in areas such as medicine, education, or media. Men are more aware of potential changes in Computer Science, autonomous driving, image and video processing, and chatbot usage, while women mention more the impact on social media. Both men and women perceive potential threats in the same manner, with men more aware of war, AI controlled drones, terrain recognition, and information war. Women seem to have a stronger tendency towards ethical considerations and helping others.

---

### Detecting Basic Values in A Noisy Russian Social Media Text Data: A Multi-Stage Classification Framework

Authors: Maria Milkova, Maksim Rudnev
Published: 2026-03-19
Categories: cs.CL
arXiv: https://arxiv.org/abs/2603.18822v1

This study presents a multi-stage classification framework for detecting human values in noisy Russian language social media, validated on a random sample of 7.5 million public text posts. Drawing on Schwartz's theory of basic human values, we design a multi-stage pipeline that includes spam and nonpersonal content filtering, targeted selection of value relevant and politically relevant posts, LLM based annotation, and multi-label classification. Particular attention is given to verifying the quality of LLM annotations and model predictions against human experts. We treat human expert annotations not as ground truth but as an interpretative benchmark with its own uncertainty. To account for annotation subjectivity, we aggregate multiple LLM generated judgments into soft labels that reflect varying levels of agreement. These labels are then used to train transformer based models capable of predicting the probability of each of the ten basic values. The best performing model, XLM RoBERTa large, achieves an F1 macro of 0.83 and an F1 of 0.71 on held out test data. By treating value detection as a multi perspective interpretive task, where expert labels, GPT annotations, and model predictions represent coherent but not identical readings of the same texts, we show that the model generally aligns with human judgments but systematically overestimates the Openness to Change value domain. Empirically, the study reveals distinct patterns of value expression and their co-occurrence in Russian social networks, contributing to a broader research agenda on cultural variation, communicative framing, and value based interpretation in digital environments. All models are released publicly.

---

### Seasoning Generative Models for a Generalization Aftertaste

Authors: Hisham Husain, Valentin De Bortoli, Richard Nock
Published: 2026-03-19
Categories: cs.LG
arXiv: https://arxiv.org/abs/2603.18817v1

The use of discriminators to train or fine-tune generative models has proven to be a rather successful framework. A notable example is Generative Adversarial Networks (GANs) that minimize a loss incurred by training discriminators along with other paradigms that boost generative models via discriminators that satisfy weak learner constraints. More recently, even diffusion models have shown advantages with some kind of discriminator guidance. In this work, we extend a strong-duality result related to $f$-divergences which gives rise to a discriminator-guided recipe that allows us to \textit{refine} any generative model. We then show that the refined generative models provably improve generalization, compared to its non-refined counterpart. In particular, our analysis reveals that the gap in generalization is improved based on the Rademacher complexity of the discriminator set used for refinement. Our recipe subsumes a recently introduced score-based diffusion approach (Kim et al., 2022) that has shown great empirical success, however allows us to shed light on the generalization guarantees of this method by virtue of our analysis. Thus, our work provides a theoretical validation for existing work, suggests avenues for new algorithms, and contributes to our understanding of generalization in generative models at large.

---

### ProRL Agent: Rollout-as-a-Service for RL Training of Multi-Turn LLM Agents

Authors: Hao Zhang, Mingjie Liu, Shaokun Zhang, Songyang Han, Jian Hu, Zhenghui Jin, Yuchi Zhang, Shizhe Diao, Ximing Lu, Binfeng Xu, Zhiding Yu, Jan Kautz et al.
Published: 2026-03-19
Categories: cs.AI
arXiv: https://arxiv.org/abs/2603.18815v1

Multi-turn LLM agents are increasingly important for solving complex, interactive tasks, and reinforcement learning (RL) is a key ingredient for improving their long-horizon behavior. However, RL training requires generating large numbers of sandboxed rollout trajectories, and existing infrastructures often couple rollout orchestration with the training loop, making systems hard to migrate and maintain. Under the rollout-as-a-service philosophy, we present ProRL Agent , a scalable infrastructure that serves the full agentic rollout lifecycle through an API service. ProRL Agent also provides standardized and extensible sandbox environments that support diverse agentic tasks in rootless HPC settings. We validate ProRL Agent through RL training on software engineering, math, STEM, and coding tasks. ProRL Agent is open-sourced and integrated as part of NVIDIA NeMo Gym.

---

### Can LLM generate interesting mathematical research problems?

Authors: Xiaoyang Chen, Xiang Jiang
Published: 2026-03-19
Categories: cs.AI
arXiv: https://arxiv.org/abs/2603.18813v1

This paper is the second one in a series of work on the mathematical creativity of LLM. In the first paper, the authors proposed three criteria for evaluating the mathematical creativity of LLM and constructed a benchmark dataset to measure it. This paper further explores the mathematical creativity of LLM, with a focus on investigating whether LLM can generate valuable and cutting-edge mathematical research problems. We develop an agent to generate unknown problems and produced 665 research problems in differential geometry. Through human verification, we find that many of these mathematical problems are unknown to experts and possess unique research value.

---

### dTRPO: Trajectory Reduction in Policy Optimization of Diffusion Large Language Models

Authors: Wenxuan Zhang, Lemeng Wu, Changsheng Zhao, Ernie Chang, Mingchen Zhuge, Zechun Liu, Andy Su, Hanxian Huang, Jun Chen, Chong Zhou, Raghuraman Krishnamoorthi, Vikas Chandra et al.
Published: 2026-03-19
Categories: cs.AI
arXiv: https://arxiv.org/abs/2603.18806v2

Diffusion Large Language Models (dLLMs) introduce a new paradigm for language generation, which in turn presents new challenges for aligning them with human preferences. In this work, we aim to improve the policy optimization for dLLMs by reducing the cost of the trajectory probability calculation, thereby enabling scaled-up offline policy training. We prove that: (i) under reference policy regularization, the probability ratio of the newly unmasked tokens is an unbiased estimate of that of intermediate diffusion states, and (ii) the probability of the full trajectory can be effectively estimated with a single forward pass of a re-masked final state. By integrating these two trajectory reduction strategies into a policy optimization objective, we propose Trajectory Reduction Policy Optimization (dTRPO). We evaluate dTRPO on 7B dLLMs across instruction-following and reasoning benchmarks. Results show that it substantially improves the core performance of state-of-the-art dLLMs, achieving gains of up to 9.6% on STEM tasks, up to 4.3% on coding tasks, and up to 3.0% on instruction-following tasks. Moreover, dTRPO exhibits strong training efficiency due to its offline, single-forward nature, and achieves improved generation efficiency through high-quality outputs.

---

### Signals of Success and Struggle: Early Prediction and Physiological Signatures of Human Performance across Task Complexity

Authors: Yufei Cao, Penny Sweetser, Ziyu Chen, Xuanying Zhu
Published: 2026-03-19
Categories: cs.LG, cs.HC
arXiv: https://arxiv.org/abs/2603.18798v1

User performance is crucial in interactive systems, capturing how effectively users engage with task execution. Prospectively predicting performance enables the timely identification of users struggling with task demands. While ocular and cardiac signals are widely used to characterise performance-relevant visual behaviour and physiological activation, their potential for early prediction and for revealing the physiological mechanisms underlying performance differences remains underexplored. We conducted a within-subject experiment in a game environment with naturally unfolding complexity, using early ocular and cardiac signals to predict later performance and to examine physiological and self-reported group differences. Results show that the ocular-cardiac fusion model achieves a balanced accuracy of 0.86, and the ocular-only model shows comparable predictive power. High performers exhibited targeted gaze and adjusted visual sampling, and sustained more stable cardiac activation as demands intensified, with a more positive affective experience. These findings demonstrate the feasibility of cross-session prediction from early physiology, providing interpretable insights into performance variation and facilitating future proactive intervention.

---

### VesselTok: Tokenizing Vessel-like 3D Biomedical Graph Representations for Reconstruction and Generation

Authors: Chinmay Prabhakar, Bastian Wittmann, Tamaz Amiranashvili, Paul Büschl, Ezequiel de la Rosa, Julian McGinnis, Benedikt Wiestler, Bjoern Menze, Suprosanna Shit
Published: 2026-03-19
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.18797v2

Spatial graphs provide a lightweight and elegant representation of curvilinear anatomical structures such as blood vessels, lung airways, and neuronal networks. Accurately modeling these graphs is crucial in clinical and (bio-)medical research. However, the high spatial resolution of large networks drastically increases their complexity, resulting in significant computational challenges. In this work, we aim to tackle these challenges by proposing VesselTok, a framework that approaches spatially dense graphs from a parametric shape perspective to learn latent representations (tokens). VesselTok leverages centerline points with a pseudo radius to effectively encode tubular geometry. Specifically, we learn a novel latent representation conditioned on centerline points to encode neural implicit representations of vessel-like, tubular structures. We demonstrate VesselTok's performance across diverse anatomies, including lung airways, lung vessels, and brain vessels, highlighting its ability to robustly encode complex topologies. To prove the effectiveness of VesselTok's learnt latent representations, we show that they (i) generalize to unseen anatomies, (ii) support generative modeling of plausible anatomical graphs, and (iii) transfer effectively to downstream inverse problems, such as link prediction.

---

### Perceptio: Perception Enhanced Vision Language Models via Spatial Token Generation

Authors: Yuchen Li, Amanmeet Garg, Shalini Chaudhuri, Rui Zhao, Garin Kessler
Published: 2026-03-19
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2603.18795v1

Large Vision Language Models (LVLMs) excel at semantic understanding but struggle with fine grained spatial grounding, as the model must implicitly infer complex geometry without ever producing a spatial interpretation. We present Perceptio, a perception enhanced LVLM with 2D and 3D spatial reasoning abilities, enabled via explicit semantic segmentation tokens and depth tokens generated directly within the autoregressive sequence. Concretely, we (i) distill a VQVAE depth codebook from a strong monocular teacher to tokenize dense depth into compact sequences, and (ii) integrate SAM2 based semantic segmentation tokens and VQ-VAE depth tokens inside the LLM so the model first emits spatial tokens and then answers. To stabilize depth token generation, we introduce novel composite depth-token objectives (marker, token, and count losses) and a soft-merging technique for differentiable reconstruction. We adopt a multi-task co-training strategy across diverse datasets, letting the model learn perception tokens to tackle multiple downstream tasks. Building on InternVL, Perceptio achieves state-of-the-art performance across benchmarks: improving referring expression segmentation by +0.8/+1.4/+1.1 cIoU on RefCOCO/+/g HardBLINK spatial understanding accuracy by 10.3%, and MMBench accuracy by 1.0%, demonstrating that explicit spatial chain-of-thought materially strengthens spatial grounding in LVLMs.

---

### Functional Subspace Watermarking for Large Language Models

Authors: Zikang Ding, Junhao Li, Suling Wu, Junchi Yao, Hongbo Liu, Lijie Hu
Published: 2026-03-19
Categories: cs.CR, cs.AI
arXiv: https://arxiv.org/abs/2603.18793v1

Model watermarking utilizes internal representations to protect the ownership of large language models (LLMs). However, these features inevitably undergo complex distortions during realistic model modifications such as fine-tuning, quantization, or knowledge distillation, making reliable extraction extremely challenging. Despite extensive research on model-side watermarking, existing methods still lack sufficient robustness against parameter-level perturbations. To address this gap, we propose \texttt{\textbf{Functional Subspace Watermarking (FSW)}}, a framework that anchors ownership signals into a low-dimensional functional backbone. Specifically, we first solve a generalized eigenvalue problem to extract a stable functional subspace for watermark injection, while introducing an adaptive spectral truncation strategy to achieve an optimal balance between robustness and model utility. Furthermore, a vector consistency constraint is incorporated to ensure that watermark injection does not compromise the original semantic performance. Extensive experiments across various LLM architectures and datasets demonstrate that our method achieves superior detection accuracy and statistical verifiability under multiple model attacks, maintaining robustness that outperforms existing state-of-the-art (SOTA) methods.

---

### Rethinking Uncertainty Quantification and Entanglement in Image Segmentation

Authors: Jakob Lønborg Christensen, Vedrana Andersen Dahl, Morten Rieger Hannemose, Anders Bjorholm Dahl, Christian F. Baumgartner
Published: 2026-03-19
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.18792v2

Uncertainty quantification (UQ) is crucial in safety-critical applications such as medical image segmentation. Total uncertainty is typically decomposed into data-related aleatoric uncertainty (AU) and model-related epistemic uncertainty (EU). Many methods exist for modeling AU (such as Probabilistic UNet, Diffusion) and EU (such as ensembles, MC Dropout), but it is unclear how they interact when combined. Additionally, recent work has revealed substantial entanglement between AU and EU, undermining the interpretability and practical usefulness of the decomposition. We present a comprehensive empirical study covering a broad range of AU-EU model combinations, propose a metric to quantify uncertainty entanglement, and evaluate both across downstream UQ tasks. Ensembles consistently exhibit lower entanglement and superior performance. Softmax models usually beat other AU methods, except in calibration where the results are dataset-dependent. A softmax ensemble performs remarkably well on all tasks. Finally, we analyze potential sources of uncertainty entanglement and outline directions for mitigating this effect.

---

### Mi:dm K 2.5 Pro

Authors: KT Tech innovation Group
Published: 2026-03-19
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2603.18788v2

The evolving LLM landscape requires capabilities beyond simple text generation, prioritizing multi-step reasoning, long-context understanding, and agentic workflows. This shift challenges existing models in enterprise environments, especially in Korean-language and domain-specific scenarios where scaling is insufficient. We introduce Mi:dm K 2.5 Pro, a 32B parameter flagship LLM designed to address enterprise-grade complexity through reasoning-focused optimization. Our methodology builds a robust data foundation via a quality-centric curation pipeline utilizing abstract syntax tree (AST) analysis for code, gap-filling synthesis for mathematics, and an LLM-based quality evaluator. Pre-training scales the model via layer-predictor-based Depth Upscaling (DuS) and a progressive strategy supporting a 128K token context window. Post-training introduces a specialized multi-stage pipeline, including Reasoning SFT, model merging, and asynchronous reinforcement learning (RL), to develop complex problem-solving skills. "Fusion Training" then rebalances these capabilities with conversational fluency, consistent response styling, and reliable tool-use. The evaluations show that Mi:dm K 2.5 Pro achieves competitive performance against leading global and domestic models. In addition, it sets state-of-the-art results on Korean-specific benchmarks, showcasing deep linguistic and cultural understanding. Finally, Responsible AI evaluations validate safety against attacks, ensuring a secure profile for deployment with a balance of harmlessness and responsiveness.

---

### Proceedings of the 2nd Workshop on Advancing Artificial Intelligence through Theory of Mind

Authors: Nitay Alon, Joseph M. Barnby, Reuth Mirsky, Stefan Sarkadi
Published: 2026-03-19
Categories: cs.AI
arXiv: https://arxiv.org/abs/2603.18786v1

This volume includes a selection of papers presented at the 2nd Workshop on Advancing Artificial Intelligence through Theory of Mind held at AAAI 2026 in Singapore on 26th January 2026. The purpose of this volume is to provide an open access and curated anthology for the ToM and AI research community.

---

### Points-to-3D: Structure-Aware 3D Generation with Point Cloud Priors

Authors: Jiatong Xia, Zicheng Duan, Anton van den Hengel, Lingqiao Liu
Published: 2026-03-19
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2603.18782v3

Recent progress in 3D generation has been driven largely by models conditioned on images or text, while readily available 3D priors are still underused. In many real-world scenarios, the visible-region point cloud are easy to obtain from active sensors such as LiDAR or from feed-forward predictors like VGGT, offering explicit geometric constraints that current methods fail to exploit. In this work, we introduce Points-to-3D, a diffusion-based framework that leverages point cloud priors for geometry-controllable 3D asset and scene generation. Built on a latent 3D diffusion model TRELLIS, Points-to-3D first replaces pure-noise sparse structure latent initialization with a point cloud priors tailored input formulation.A structure inpainting network, trained within the TRELLIS framework on task-specific data designed to learn global structural inpainting, is then used for inference with a staged sampling strategy (structural inpainting followed by boundary refinement), completing the global geometry while preserving the visible regions of the input priors. In practice, Points-to-3D can take either accurate point-cloud priors or VGGT-estimated point clouds from single images as input. Experiments on both objects and scene scenarios consistently demonstrate superior performance over state-of-the-art baselines in terms of rendering quality and geometric fidelity, highlighting the effectiveness of explicitly embedding point-cloud priors for achieving more accurate and structurally controllable 3D generation. Project page: https://jiatongxia.github.io/points2-3D/

---

### SRRM: Improving Recursive Transport Surrogates in the Small-Discrepancy Regime

Authors: Yufei Zhang, Tao Wang, Jingyi Zhang
Published: 2026-03-19
Categories: stat.ML, cs.LG, stat.AP
arXiv: https://arxiv.org/abs/2603.18781v1

Recursive partitioning methods provide computationally efficient surrogates for the Wasserstein distance, yet their statistical behavior and their resolution in the small-discrepancy regime remain insufficiently understood. We study Recursive Rank Matching (RRM) as a representative instance of this class under a population-anchored reference. In this setting, we establish consistency and an explicit convergence rate for the anchored empirical RRM under the quadratic cost. We then identify a dominant mismatch mechanism responsible for the loss of resolution in the small-discrepancy regime. Based on this analysis, we introduce Selective Recursive Rank Matching (SRRM), which suppresses the resulting dominant mismatches and yields a higher-fidelity practical surrogate for the Wasserstein distance at moderate additional computational cost.

---

### SEAR: Simple and Efficient Adaptation of Visual Geometric Transformers for Unpaired RGB+Thermal 3D Reconstruction

Authors: Vsevolod Skorokhodov, Chenghao Xu, Shuo Sun, Olga Fink, Malcolm Mielle
Published: 2026-03-19
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.18774v2

Foundational feed-forward visual geometry models enable accurate and efficient camera pose estimation and scene reconstruction by learning strong scene priors from massive RGB datasets. However, their effectiveness drops when applied to mixed sensing modalities, such as RGB-thermal (RGB-T) images. We observe that while a visual geometry grounded transformer pretrained on RGB data generalizes well to thermal-only reconstruction, it struggles to align RGB and thermal modalities when processed jointly. To address this, we propose SEAR, a simple yet efficient fine-tuning strategy that adapts a pretrained geometry transformer to multimodal RGB-T inputs. Despite being trained on a relatively small RGB-T dataset, our approach significantly outperforms state-of-the-art methods for 3D reconstruction and camera pose estimation, achieving significant improvements over all metrics and delivering higher detail and consistency between modalities with negligible overhead in inference time compared to the original pretrained model. Notably, SEAR enables reliable multimodal pose estimation and reconstruction even under challenging conditions, such as low lighting and dense smoke. We validate our architecture through extensive ablation studies and demonstrate how the model aligns both modalities. Additionally, we introduce a new dataset featuring RGB and thermal sequences captured at different times, viewpoints, and illumination conditions, providing a robust benchmark for future work in multimodal 3D scene reconstruction. Code and models are publicly available at https://doi.org/10.5281/ZENODO.21077295.

---

### Automatic Configuration of LLM Post-Training Pipelines

Authors: Channe Chwa, Xinle Wu, Yao Lu
Published: 2026-03-19
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2603.18773v1

LLM post-training pipelines that combine supervised fine-tuning and reinforcement learning are difficult to configure under realistic compute budgets: the configuration space is high-dimensional and heterogeneous, stages are strongly coupled, and each end-to-end evaluation is expensive. We propose AutoPipe, a budget-aware two-stage framework for configuration selection in LLM post-training. Offline, AutoPipe learns a dataset-conditioned learning-to-rank surrogate from historical runs, capturing within-dataset preferences and providing transferable guidance toward promising regions of the configuration space. Online, for a new dataset, AutoPipe uses the offline guidance to steer Bayesian optimization and models dataset-specific deviations with a Gaussian-process residual surrogate. To reduce evaluation cost, each trial is early-stopped and scored by a learned predictor that maps early training signals to a low-cost proxy for final post-training performance. Experiments on biomedical reasoning tasks show that AutoPipe consistently outperforms offline-only baselines and achieves comparable performance with the strongest online HPO baselines while using less than 10\% of their computational cost.

---

### A Concept is More Than a Word: Diversified Unlearning in Text-to-Image Diffusion Models

Authors: Duc Hao Pham, Van Duy Truong, Duy Khanh Dinh, Tien Cuong Nguyen, Dien Hy Ngo, Tuan Anh Bui
Published: 2026-03-19
Categories: cs.AI
arXiv: https://arxiv.org/abs/2603.18767v1

Concept unlearning has emerged as a promising direction for reducing the risks of harmful content generation in text-to-image diffusion models by selectively erasing undesirable concepts from a model's parameters. Existing approaches typically rely on keywords to identify the target concept to be unlearned. However, we show that this keyword-based formulation is inherently limited: a visual concept is multi-dimensional, can be expressed in diverse textual forms, and often overlap with related concepts in the latent space, making keyword-only unlearning, which imprecisely indicate the target concept is brittle and prone to over-forgetting. This occurs because a single keyword represents only a narrow point estimate of the concept, failing to cover its full semantic distribution and entangled variations in the latent space. To address this limitation, we propose Diversified Unlearning, a distributional framework that represents a concept through a set of contextually diverse prompts rather than a single keyword. This richer representation enables more precise and robust unlearning. Through extensive experiments across multiple benchmarks and state-of-the-art baselines, we demonstrate that integrating Diversified Unlearning as an add-on component into existing unlearning pipelines consistently achieves stronger erasure, better retention of unrelated concepts, and improved robustness against adversarial recovery attacks.

---

### Enhancing the Parameterization of Reservoir Properties for Data Assimilation Using Deep VAE-GAN

Authors: M. A. Sampaio, P. H. Ranazzi, M. J. Blunt
Published: 2026-03-19
Categories: cs.LG
arXiv: https://arxiv.org/abs/2603.18766v2

Currently, the methods called Iterative Ensemble Smoothers, especially the method called Ensemble Smoother with Multiple Data Assimilation (ESMDA) can be considered state-of-the-art for history matching in petroleum reservoir simulation. However, this approach has two important limitations: the use of an ensemble with finite size to represent the distributions and the Gaussian assumption in parameter and data uncertainties. This latter is particularly important because many reservoir properties have non-Gaussian distributions. Parameterization involves mapping non-Gaussian parameters to a Gaussian field before the update and then mapping them back to the original domain to forward the ensemble through the reservoir simulator. A promising approach to perform parameterization is through deep learning models. Recent studies have shown that Generative Adversarial Networks (GAN) performed poorly concerning data assimilation, but generated more geologically plausible realizations of the reservoir, while the Variational Autoencoder (VAE) performed better than the GAN in data assimilation, but generated less geologically realistic models. This work is innovative in combining the strengths of both to implement a deep learning model called Variational Autoencoder Generative Adversarial Network (VAE-GAN) integrated with ESMDA. The methodology was applied in two case studies, one case being categorical and the other with continuous values of permeability. Our findings demonstrate that by applying the VAE-GAN model we can obtain high quality reservoir descriptions (just like GANs) and a good history matching on the production curves (just like VAEs) simultaneously.

---

### Implicit Grading Bias in Large Language Models: How Writing Style Affects Automated Assessment Across Math, Programming, and Essay Tasks

Authors: Rudra Jadhav, Janhavi Danve, Sonalika Shaw
Published: 2026-03-19
Categories: cs.CL
arXiv: https://arxiv.org/abs/2603.18765v1

As large language models (LLMs) are increasingly deployed as automated graders in educational settings, concerns about fairness and bias in their evaluations have become critical. This study investigates whether LLMs exhibit implicit grading bias based on writing style when the underlying content correctness remains constant. We constructed a controlled dataset of 180 student responses across three subjects (Mathematics, Programming, and Essay/Writing), each with three surface-level perturbation types: grammar errors, informal language, and non-native phrasing. Two state-of-the-art open-source LLMs -- LLaMA 3.3 70B (Meta) and Qwen 2.5 72B (Alibaba) -- were prompted to grade responses on a 1-10 scale with explicit instructions to evaluate content correctness only and to disregard writing style. Our results reveal statistically significant grading bias in Essay/Writing tasks across both models and all perturbation types (p < 0.05), with effect sizes ranging from medium (Cohen's d = 0.64) to very large (d = 4.25). Informal language received the heaviest penalty, with LLaMA deducting an average of 1.90 points and Qwen deducting 1.20 points on a 10-point scale -- penalties comparable to the difference between a B+ and C+ letter grade. Non-native phrasing was penalized 1.35 and 0.90 points respectively. In sharp contrast, Mathematics and Programming tasks showed minimal bias, with most conditions failing to reach statistical significance. These findings demonstrate that LLM grading bias is subject-dependent, style-sensitive, and persists despite explicit counter-bias instructions in the grading prompt. We discuss implications for equitable deployment of LLM-based grading systems and recommend bias auditing protocols before institutional adoption.

---

### ProCal: Probability Calibration for Neighborhood-Guided Source-Free Domain Adaptation

Authors: Ying Zheng, Yiyi Zhang, Yi Wang, Lap-Pui Chau
Published: 2026-03-19
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.18764v1

Source-Free Domain Adaptation (SFDA) adapts pre-trained models to unlabeled target domains without requiring access to source data. Although state-of-the-art methods leveraging local neighborhood structures show promise for SFDA, they tend to over-rely on prediction similarity among neighbors. This over-reliance accelerates the forgetting of source knowledge and increases susceptibility to local noise overfitting. To address these issues, we introduce ProCal, a probability calibration method that dynamically calibrates neighborhood-based predictions through a dual-model collaborative prediction mechanism. ProCal integrates the source model's initial predictions with the current model's online outputs to effectively calibrate neighbor probabilities. This strategy not only mitigates the interference of local noise but also preserves the discriminative information from the source model, thereby achieving a balance between knowledge retention and domain adaptation. Furthermore, we design a joint optimization objective that combines a soft supervision loss with a diversity loss to guide the target model. Our theoretical analysis shows that ProCal converges to an equilibrium where source knowledge and target information are effectively fused, reducing both knowledge forgetting and overfitting. We validate the effectiveness of our approach through extensive experiments on 31 cross-domain tasks across four public datasets. Our code is available at: https://github.com/zhengyinghit/ProCal.

---

### ClawTrap: A MITM-Based Red-Teaming Framework for Real-World OpenClaw Security Evaluation

Authors: Haochen Zhao, Shaoyang Cui
Published: 2026-03-19
Categories: cs.CR, cs.AI
arXiv: https://arxiv.org/abs/2603.18762v1

Autonomous web agents such as \textbf{OpenClaw} are rapidly moving into high-impact real-world workflows, but their security robustness under live network threats remains insufficiently evaluated. Existing benchmarks mainly focus on static sandbox settings and content-level prompt attacks, which leaves a practical gap for network-layer security testing. In this paper, we present \textbf{ClawTrap}, a \textbf{MITM-based red-teaming framework for real-world OpenClaw security evaluation}. ClawTrap supports diverse and customizable attack forms, including \textit{Static HTML Replacement}, \textit{Iframe Popup Injection}, and \textit{Dynamic Content Modification}, and provides a reproducible pipeline for rule-driven interception, transformation, and auditing. This design lays the foundation for future research to construct richer, customizable MITM attacks and to perform systematic security testing across agent frameworks and model backbones. Our empirical study shows clear model stratification: weaker models are more likely to trust tampered observations and produce unsafe outputs, while stronger models demonstrate better anomaly attribution and safer fallback strategies. These findings indicate that reliable OpenClaw security evaluation should explicitly incorporate dynamic real-world MITM conditions rather than relying only on static sandbox protocols.

---

### NeuroGame Transformer: Gibbs-Inspired Attention Driven by Game Theory and Statistical Physics

Authors: Djamel Bouchaffra, Faycal Ykhlef, Hanene Azzag, Mustapha Lebbah, Bilal Faye
Published: 2026-03-19
Categories: cs.AI
arXiv: https://arxiv.org/abs/2603.18761v1

Standard attention mechanisms in transformers are limited by their pairwise formulation, which hinders the modeling of higher-order dependencies among tokens. We introduce the NeuroGame Transformer (NGT) to overcome this by reconceptualizing attention through a dual perspective: tokens are treated simultaneously as players in a cooperative game and as interacting spins in a statistical physics system. Token importance is quantified using two complementary game-theoretic concepts -- Shapley values for global, permutation-based attribution and Banzhaf indices for local, coalition-level influence. These are combined via a learnable gating parameter to form an external magnetic field, while pairwise interaction potentials capture synergistic relationships. The system's energy follows an Ising Hamiltonian, with attention weights emerging as marginal probabilities under the Gibbs distribution, efficiently computed via mean-field equations. To ensure scalability despite the exponential coalition space, we develop importance-weighted Monte Carlo estimators with Gibbs-distributed weights. This approach avoids explicit exponential factors, ensuring numerical stability for long sequences. We provide theoretical convergence guarantees and characterize the fairness-sensitivity trade-off governed by the interpolation parameter. Experimental results demonstrate that the NeuroGame Transformer achieves strong performance across SNLI, and MNLI-matched, outperforming some major efficient transformer baselines. On SNLI, it attains a test accuracy of 86.4\% (with a peak validation accuracy of 86.6\%), surpassing ALBERT-Base and remaining highly competitive with RoBERTa-Base. Code is available at https://github.com/dbouchaffra/NeuroGame-Transformer.

---

### Dual-Model Prediction of Affective Engagement and Vocal Attractiveness from Speaker Expressiveness in Video Learning

Authors: Hung-Yue Suen, Kuo-En Hung, Fan-Hsun Tseng
Published: 2026-03-19
Categories: cs.HC, cs.CV, cs.SD
arXiv: https://arxiv.org/abs/2603.18758v1

This paper outlines a machine learning-enabled speaker-centric Emotion AI approach capable of predicting audience-affective engagement and vocal attractiveness in asynchronous video-based learning, relying solely on speaker-side affective expressions. Inspired by the demand for scalable, privacy-preserving affective computing applications, this speaker-centric Emotion AI approach incorporates two distinct regression models that leverage a massive corpus developed within Massive Open Online Courses (MOOCs) to enable affectively engaging experiences. The regression model predicting affective engagement is developed by assimilating emotional expressions emanating from facial dynamics, oculomotor features, prosody, and cognitive semantics, while incorporating a second regression model to predict vocal attractiveness based exclusively on speaker-side acoustic features. Notably, on speaker-independent test sets, both regression models yielded impressive predictive performance (R2 = 0.85 for affective engagement and R2 = 0.88 for vocal attractiveness), confirming that speaker-side affect can functionally represent aggregated audience feedback. This paper provides a speaker-centric Emotion AI approach substantiated by an empirical study discovering that speaker-side multimodal features, including acoustics, can prospectively forecast audience feedback without necessarily employing audience-side input information.

---

### DA-Mamba: Learning Domain-Aware State Space Model for Global-Local Alignment in Domain Adaptive Object Detection

Authors: Haochen Li, Rui Zhang, Hantao Yao, Xin Zhang, Yifan Hao, Shaohui Peng, Yongwei Zhao, Ling Li
Published: 2026-03-19
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.18757v1

Domain Adaptive Object Detection (DAOD) aims to transfer detectors from a labeled source domain to an unlabeled target domain. Existing DAOD methods employ multi-granularity feature alignment to learn domain-invariant representations. However, the local connectivity of their CNN-based backbone and detection head restricts alignment to local regions, failing to extract global domain-invariant features. Although transformer-based DAOD methods capture global dependencies via attention mechanisms, their quadratic computational cost hinders practical deployment. To solve this, we propose DA-Mamba, a hybrid CNN-State Space Models (SSMs) architecture that combines the efficiency of CNNs with the linear-time long-range modeling capability of State Space Models (SSMs) to capture both global and local domain-invariant features. Specifically, we introduce two novel modules: Image-Aware SSM (IA-SSM) and Object-Aware SSM (OA-SSM). IA-SSM is integrated into the backbone to enhance global domain awareness, enabling image-level global and local alignment. OA-SSM is inserted into the detection head to model spatial and semantic dependencies among objects, enhancing instance-level alignment. Comprehensive experiments demonstrate that the proposed method can efficiently improve the cross-domain performance of the object detector.

---

### Are complicated loss functions necessary for teaching LLMs to reason?

Authors: Gabriele Carrino, Andrea Sassella, Nicolo Brunello, Federico Toschi, Mark James Carman
Published: 2026-03-19
Categories: cs.LG, cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2603.18756v1

Recent advances in large language models (LLMs) highlight the importance of post training techniques for improving reasoning and mathematical ability. Group Relative Policy Optimization (GRPO) has shown promise in this domain by combining group relative advantage estimation, PPO style clipping, and KL regularization. However, its complexity raises the question of whether all components are necessary for fostering reasoning behaviors. We conduct a systematic analysis of GRPO and identify two key findings: (1) incorporating negative feedback is essential training solely on actions above a baseline limits learning; and (2) PPO style constraints, such as policy ratio clipping, are not required to improve mathematical reasoning or performance. Building on these insights, we propose REINFORCE with Group Relative Advantage (RGRA), a simplified variant that retains group relative advantage estimation but removes PPO style clipping and policy ratio terms. Experiments across standard mathematical benchmarks indicate that RGRA has the potential to achieve stronger performance than GRPO. Our results suggest that simpler REINFORCE based approaches can effectively enhance reasoning in LLMs, offering a more transparent and efficient alternative to GRPO.

---

### WeNLEX: Weakly Supervised Natural Language Explanations for Multilabel Chest X-ray Classification

Authors: Isabel Rio-Torto, Jaime S. Cardoso, Luís F. Teixeira
Published: 2026-03-19
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2603.18752v1

Natural language explanations provide an inherently human-understandable way to explain black-box models, closely reflecting how radiologists convey their diagnoses in textual reports. Most works explicitly supervise the explanation generation process using datasets annotated with explanations. Thus, though plausible, the generated explanations are not faithful to the model's reasoning. In this work, we propose WeNLEX, a weakly supervised model for the generation of natural language explanations for multilabel chest X-ray classification. Faithfulness is ensured by matching images generated from their corresponding natural language explanations with original images, in the black-box model's feature space. Plausibility is maintained via distribution alignment with a small database of clinician-annotated explanations. We empirically demonstrate, through extensive validation on multiple metrics to assess faithfulness, simulatability, diversity, and plausibility, that WeNLEX is able to produce faithful and plausible explanations, using as little as 5 ground-truth explanations per diagnosis. Furthermore, WeNLEX can operate in both post-hoc and in-model settings. In the latter, i.e., when the multilabel classifier is trained together with the rest of the network, WeNLEX improves the classification AUC of the standalone classifier by 2.21%, thus showing that adding interpretability to the training process can actually increase the downstream task performance. Additionally, simply by changing the database, WeNLEX explanations are adaptable to any target audience, and we showcase this flexibility by training a layman version of WeNLEX, where explanations are simplified for non-medical users.

---

### Automatic detection of Gen-AI texts: A comparative framework of neural models

Authors: Cristian Buttaro, Irene Amerini
Published: 2026-03-19
Categories: cs.CL
arXiv: https://arxiv.org/abs/2603.18750v1

The rapid proliferation of Large Language Models has significantly increased the difficulty of distinguishing between human-written and AI generated texts, raising critical issues across academic, editorial, and social domains. This paper investigates the problem of AI generated text detection through the design, implementation, and comparative evaluation of multiple machine learning based detectors. Four neural architectures are developed and analyzed: a Multilayer Perceptron, a one-dimensional Convolutional Neural Network, a MobileNet-based CNN, and a Transformer model. The proposed models are benchmarked against widely used online detectors, including ZeroGPT, GPTZero, QuillBot, Originality.AI, Sapling, IsGen, Rephrase, and Writer. Experiments are conducted on the COLING Multilingual Dataset, considering both English and Italian configurations, as well as on an original thematic dataset focused on Art and Mental Health. Results show that supervised detectors achieve more stable and robust performance than commercial tools across different languages and domains, highlighting key strengths and limitations of current detection strategies.

---

### Memento-Skills: Let Agents Design Agents

Authors: Huichi Zhou, Siyuan Guo, Anjie Liu, Zhongwei Yu, Ziqin Gong, Bowen Zhao, Zhixun Chen, Menglong Zhang, Yihang Chen, Jinsong Li, Runyu Yang, Qiangbin Liu et al.
Published: 2026-03-19
Categories: cs.AI, cs.CL, cs.LG
arXiv: https://arxiv.org/abs/2603.18743v1

We introduce \emph{Memento-Skills}, a generalist, continually-learnable LLM agent system that functions as an \emph{agent-designing agent}: it autonomously constructs, adapts, and improves task-specific agents through experience. The system is built on a memory-based reinforcement learning framework with \emph{stateful prompts}, where reusable skills (stored as structured markdown files) serve as persistent, evolving memory. These skills encode both behaviour and context, enabling the agent to carry forward knowledge across interactions. Starting from simple elementary skills (like Web search and terminal operations), the agent continually improves via the \emph{Read--Write Reflective Learning} mechanism introduced in \emph{Memento~2}~\cite{wang2025memento2}. In the \emph{read} phase, a behaviour-trainable skill router selects the most relevant skill conditioned on the current stateful prompt; in the \emph{write} phase, the agent updates and expands its skill library based on new experience. This closed-loop design enables \emph{continual learning without updating LLM parameters}, as all adaptation is realised through the evolution of externalised skills and prompts. Unlike prior approaches that rely on human-designed agents, Memento-Skills enables a generalist agent to \emph{design agents end-to-end} for new tasks. Through iterative skill generation and refinement, the system progressively improves its own capabilities. Experiments on the \emph{General AI Assistants} benchmark and \emph{Humanity's Last Exam} demonstrate sustained gains, achieving 26.2\% and 116.2\% relative improvements in overall accuracy, respectively. Code is available at https://github.com/Memento-Teams/Memento-Skills.

---

### 6Bit-Diffusion: Inference-Time Mixed-Precision Quantization for Video Diffusion Models

Authors: Rundong Su, Jintao Zhang, Zhihang Yuan, Haojie Duanmu, Jianfei Chen, Jun Zhu
Published: 2026-03-19
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.18742v1

Diffusion transformers have demonstrated remarkable capabilities in generating videos. However, their practical deployment is severely constrained by high memory usage and computational cost. Post-Training Quantization provides a practical way to reduce memory usage and boost computation speed. Existing quantization methods typically apply a static bit-width allocation, overlooking the quantization difficulty of activations across diffusion timesteps, leading to a suboptimal trade-off between efficiency and quality. In this paper, we propose a inference time NVFP4/INT8 Mixed-Precision Quantization framework. We find a strong linear correlation between a block's input-output difference and the quantization sensitivity of its internal linear layers. Based on this insight, we design a lightweight predictor that dynamically allocates NVFP4 to temporally stable layers to maximize memory compression, while selectively preserving INT8 for volatile layers to ensure robustness. This adaptive precision strategy enables aggressive quantization without compromising generation quality. Beside this, we observe that the residual between the input and output of a Transformer block exhibits high temporal consistency across timesteps. Leveraging this temporal redundancy, we introduce Temporal Delta Cache (TDC) to skip computations for these invariant blocks, further reducing the computational cost. Extensive experiments demonstrate that our method achieves 1.92$\times$ end-to-end acceleration and 3.32$\times$ memory reduction, setting a new baseline for efficient inference in Video DiTs.

---

### Measuring and Exploiting Contextual Bias in LLM-Assisted Security Code Review

Authors: Georgios Alexopoulos, Nikolaos Alexopoulos, Diomidis Spinellis, Dimitris Mitropoulos
Published: 2026-03-19
Categories: cs.SE, cs.AI, cs.CR
arXiv: https://arxiv.org/abs/2603.18740v4

Automated Code Review (ACR) systems integrating Large Language Models (LLMs) are increasingly adopted in software development workflows, ranging from interactive assistants to autonomous agents in CI/CD pipelines. In this paper, we study how LLM-based vulnerability detection in ACR is affected by the framing effect: the tendency to let the presentation of information override its semantic content in forming judgments. We examine whether adversaries can exploit this through contextual-bias injection (crafting PR metadata to bias ACR security judgments) as a supply-chain attack vector against real-world ACR pipelines. To this end, we first conduct a large-scale exploratory study across 6 LLMs under five framing conditions, establishing the framing effect as a systematic and widespread phenomenon in LLM-based vulnerability detection. We then design a realistic and controlled experimental environment, evaluating 33 CVEs across 20 real-world projects and two popular ACR pipelines (Claude Code and CodeRabbit), to assess the susceptibility of real-world ACR pipelines to vulnerability re-introduction attacks. We employ two attack strategies: a template-based attack inspired by prior related work, and a novel LLM-assisted refinement attack. We find that template-based attacks are ineffective and may even backfire, as direct biasing attempts raise suspicions. Our refinement attack, on the other hand, is successful in 32/33 (97%) cases, exploiting a fundamental asymmetry: attackers can iteratively refine attacks against a local clone of the review pipeline, while defenders have only one chance to detect them. Overall, our findings highlight the dangers of over-relying on ACR and stress the importance of human oversight and contributor trust in the development process.

---

### Measuring and Exploiting Contextual Bias in LLM-Assisted Security Code Review

Authors: Dimitris Mitropoulos, Nikolaos Alexopoulos, Georgios Alexopoulos, Diomidis Spinellis
Published: 2026-03-19
Categories: cs.SE, cs.AI, cs.CR
arXiv: https://arxiv.org/abs/2603.18740v2

Automated Code Review (ACR) systems integrating Large Language Models (LLMs) are increasingly adopted in software development workflows, ranging from interactive assistants to autonomous agents in CI/CD pipelines. In this paper, we study how LLM-based vulnerability detection in ACR is affected by the framing effect: the tendency to let the presentation of information override its semantic content in forming judgments. We examine whether adversaries can exploit this through contextual-bias injection: crafting PR metadata to bias ACR security judgments as a supply-chain attack vector against real-world ACR pipelines. To this end, we first conduct a large-scale exploratory study across 6 LLMs under five framing conditions, establishing the framing effect as a systematic and widespread phenomenon in LLM-based vulnerability detection, with bug-free framing producing the strongest effect. We then design a realistic and controlled experimental environment, evaluating 17 CVEs across 10 real-world projects, to assess the susceptibility of real-world ACR pipelines to vulnerability reintroduction attacks. We employ two attack strategies: a template-based attack inspired by prior related work, and a novel LLM-assisted iterative refinement attack. We find that template-based attacks are ineffective and may even backfire, as direct biasing attempts raise suspicions. Our iterative refinement attack, on the other hand, achieves 100% success, exploiting a fundamental asymmetry: attackers can iteratively refine attacks against a local clone of the review pipeline, while defenders have only one chance to detect them. Debiasing via metadata redaction and explicit instructions restores detection in all affected cases. Overall, our findings highlight the dangers of over-relying on ACR and stress the importance of human oversight and contributor trust in the development process.

---

### EdgeCrafter: Compact ViTs for Edge Dense Prediction via Task-Specialized Distillation

Authors: Longfei Liu, Yongjie Hou, Yang Li, Qirui Wang, Youyang Sha, Yongjun Yu, Yinzhi Wang, Peizhe Ru, Xuanlong Yu, Xi Shen
Published: 2026-03-19
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.18739v4

Deploying high-performance dense prediction models on resource-constrained edge devices remains challenging due to strict computation and memory budgets. In practice, lightweight systems for object detection, instance segmentation, and pose estimation are still dominated by CNN-based architectures such as YOLO, while compact Vision Transformers (ViTs) often struggle to achieve comparable accuracy-efficiency trade-offs, even with large-scale pretraining. We argue that this gap arises primarily from insufficient task-specific representation learning in small-scale ViTs, rather than from an inherent mismatch between ViTs and edge dense prediction. To address this issue, we introduce EdgeCrafter, a unified compact ViT framework for edge dense prediction centered on ECDet, a detection model built on a distilled compact backbone and an edge-friendly encoder-decoder design. The resulting detection-distilled representation transfers directly to instance segmentation and human pose estimation through lightweight task-specific prediction modules. Without using task annotations beyond COCO, ECDet-S achieves 51.7 box AP with fewer than 10M parameters, while ECInsSeg-X and ECPose-X reach 48.4 mask AP and 74.8 keypoint AP, respectively. As a complementary but more compute-intensive setting, Objects365 detection pretraining consistently improves performance across all scales, with the X variants reaching 59.9 box AP, 49.8 mask AP, and 75.9 keypoint AP. These results show that compact ViTs, when combined with task-specialized distillation and edge-aware design, can be a practical and competitive solution for edge dense prediction. Code is available at: https://intellindust-ai-lab.github.io/projects/EdgeCrafter

---

### EdgeCrafter: Compact ViTs for Edge Dense Prediction via Task-Specialized Distillation

Authors: Longfei Liu, Yongjie Hou, Yang Li, Qirui Wang, Youyang Sha, Yongjun Yu, Yinzhi Wang, Peizhe Ru, Xuanlong Yu, Xi Shen
Published: 2026-03-19
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.18739v3

Deploying high-performance dense prediction models on resource-constrained edge devices remains challenging due to strict limits on computation and memory. In practice, lightweight systems for object detection, instance segmentation, and pose estimation are still dominated by CNN-based architectures such as YOLO, while compact Vision Transformers (ViTs) often struggle to achieve similarly strong accuracy efficiency tradeoff, even with large scale pretraining. We argue that this gap is largely due to insufficient task specific representation learning in small scale ViTs, rather than an inherent mismatch between ViTs and edge dense prediction. To address this issue, we introduce EdgeCrafter, a unified compact ViT framework for edge dense prediction centered on ECDet, a detection model built from a distilled compact backbone and an edge-friendly encoder decoder design. On the COCO dataset, ECDet-S achieves 51.7 AP with fewer than 10M parameters using only COCO annotations. For instance segmentation, ECInsSeg achieves performance comparable to RF-DETR while using substantially fewer parameters. For pose estimation, ECPose-X reaches 74.8 AP, significantly outperforming YOLO26Pose-X (71.6 AP). These results show that compact ViTs, when paired with task-specialized distillation and edge-aware design, can be a practical and competitive option for edge dense prediction. Code is available at: https://intellindust-ai-lab.github.io/projects/EdgeCrafter/

---

### CausalRM: Causal-Theoretic Reward Modeling for RLHF from Observational User Feedbacks

Authors: Hao Wang, Licheng Pan, Zhichao Chen, Chunyuan Zheng, Zhixuan Chu, Xiaoxi Li, Yuan Lu, Xinggao Liu, Haoxuan Li, Zhouchen Lin
Published: 2026-03-19
Categories: cs.LG, cs.AI, cs.CL, stat.ML
arXiv: https://arxiv.org/abs/2603.18736v1

Despite the success of reinforcement learning from human feedback (RLHF) in aligning language models, current reward modeling heavily relies on experimental feedback data collected from human annotators under controlled and costly conditions. In this work, we introduce observational reward modeling -- learning reward models with observational user feedback (e.g., clicks, copies, and upvotes) -- as a scalable and cost-effective alternative. We identify two fundamental challenges in this setting: (1) observational feedback is noisy due to annotation errors, which deviates it from true user preference; (2) observational feedback is biased by user preference, where users preferentially provide feedback on responses they feel strongly about, which creats a distribution shift between training and inference data. To address these challenges, we propose CausalRM, a causal-theoretic reward modeling framework that aims to learn unbiased reward models from observational feedback. To tackle challenge (1), CausalRM introduces a noise-aware surrogate loss term that is provably equivalent to the primal loss under noise-free conditions by explicitly modeling the annotation error generation process. To tackle challenge (2), CausalRM uses propensity scores -- the probability of a user providing feedback for a given response -- to reweight training samples, yielding a loss function that eliminates user preference bias. Extensive experiments across diverse LLM backbones and benchmark datasets validate that CausalRM effectively learns accurate reward signals from noisy and biased observational feedback and delivers substantial performance improvements on downstream RLHF tasks -- including a 49.2% gain on WildGuardMix and a 32.7% improvement on HarmBench. Code is available on our project website.

---

### Analysis Of Linguistic Stereotypes in Single and Multi-Agent Generative AI Architectures

Authors: Martina Ullasci, Marco Rondina, Riccardo Coppola, Flavio Giobergia, Riccardo Bellanca, Gabriele Mancari Pasi, Luca Prato, Federico Spinoso, Silvia Tagliente
Published: 2026-03-19
Categories: cs.AI
arXiv: https://arxiv.org/abs/2603.18729v1

Many works in the literature show that LLM outputs exhibit discriminatory behaviour, triggering stereotype-based inferences based on the dialect in which the inputs are written. This bias has been shown to be particularly pronounced when the same inputs are provided to LLMs in Standard American English (SAE) and African-American English (AAE). In this paper, we replicate existing analyses of dialect-sensitive stereotype generation in LLM outputs and investigate the effects of mitigation strategies, including prompt engineering (role-based and Chain-Of-Thought prompting) and multi-agent architectures composed of generate-critique-revise models. We define eight prompt templates to analyse different ways in which dialect bias can manifest, such as suggested names, jobs, and adjectives for SAE or AAE speakers. We use an LLM-as-judge approach to evaluate the bias in the results. Our results show that stereotype-bearing differences emerge between SAE- and AAE-related outputs across all template categories, with the strongest effects observed in adjective and job attribution. Baseline disparities vary substantially by model, with the largest SAE-AAE differential observed in Claude Haiku and the smallest in Phi-4 Mini. Chain-Of-Thought prompting proved to be an effective mitigation strategy for Claude Haiku, whereas the use of a multi-agent architecture ensured consistent mitigation across all the models. These findings suggest that for intersectionality-informed software engineering, fairness evaluation should include model-specific validation of mitigation strategies, and workflow-level controls (e.g., agentic architectures involving critique models) in high-impact LLM deployments. The current results are exploratory in nature and limited in scope, but can lead to extensions and replications by increasing the dataset size and applying the procedure to different languages or dialects.

---

### Ontology-Guided Diffusion for Zero-Shot Visual Sim2Real Transfer

Authors: Mohamed Youssef, Mayar Elfares, Anna-Maria Meer, Matteo Bortoletto, Andreas Bulling
Published: 2026-03-19
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2603.18719v2

Bridging the simulation-to-reality (sim2real) gap remains challenging as labelled real-world data is scarce. Existing diffusion-based approaches rely on unstructured prompts or statistical alignment, which do not capture the structured factors that make images look real. We introduce Ontology- Guided Diffusion (OGD), a neuro-symbolic zero-shot sim2real image translation framework that represents realism as structured knowledge. OGD decomposes realism into an ontology of interpretable traits -- such as lighting and material properties -- and encodes their relationships in a knowledge graph. From a synthetic image, OGD infers trait activations and uses a graph neural network to produce a global embedding. In parallel, a symbolic planner uses the ontology traits to compute a consistent sequence of visual edits needed to narrow the realism gap. The graph embedding conditions a pretrained instruction-guided diffusion model via cross-attention, while the planned edits are converted into a structured instruction prompt. Across benchmarks, our graph-based embeddings better distinguish real from synthetic imagery than baselines, and OGD outperforms state-of-the-art diffusion methods in sim2real image translations. Overall, OGD shows that explicitly encoding realism structure enables interpretable, data-efficient, and generalisable zero-shot sim2real transfer.

---

### MemMA: Coordinating the Memory Cycle through Multi-Agent Reasoning and In-Situ Self-Evolution

Authors: Minhua Lin, Zhiwei Zhang, Hanqing Lu, Hui Liu, Xianfeng Tang, Qi He, Xiang Zhang, Suhang Wang
Published: 2026-03-19
Categories: cs.AI
arXiv: https://arxiv.org/abs/2603.18718v2

Memory-augmented LLM agents maintain external memory banks to support long-horizon interaction, yet most existing systems treat construction, retrieval, and utilization as isolated subroutines. This creates two coupled challenges: strategic blindness on the forward path of the memory cycle, where construction and retrieval are driven by local heuristics rather than explicit strategic reasoning, and sparse, delayed supervision on the backward path, where downstream failures rarely translate into direct repairs of the memory bank. To address these challenges, we propose MemMA, a plug-and-play multi-agent framework that coordinates the memory cycle along both the forward and backward paths. On the forward path, a Meta-Thinker produces structured guidance that steers a Memory Manager during construction and directs a Query Reasoner during iterative retrieval. On the backward path, MemMA introduces in-situ self-evolving memory construction, which synthesizes probe QA pairs, verifies the current memory, and converts failures into repair actions before the memory is finalized. Extensive experiments on LoCoMo show that MemMA consistently outperforms existing baselines across multiple LLM backbones and improves three different storage backends in a plug-and-play manner. Our code is publicly available at https://github.com/ventr1c/memma.

---

### MemMA: Coordinating the Memory Cycle through Multi-Agent Reasoning and In-Situ Self-Evolution

Authors: Minhua Lin, Zhiwei Zhang, Hanqing Lu, Hui Liu, Xianfeng Tang, Qi He, Xiang Zhang, Suhang Wang
Published: 2026-03-19
Categories: cs.AI
arXiv: https://arxiv.org/abs/2603.18718v1

Memory-augmented LLM agents maintain external memory banks to support long-horizon interaction, yet most existing systems treat construction, retrieval, and utilization as isolated subroutines. This creates two coupled challenges: strategic blindness on the forward path of the memory cycle, where construction and retrieval are driven by local heuristics rather than explicit strategic reasoning, and sparse, delayed supervision on the backward path, where downstream failures rarely translate into direct repairs of the memory bank. To address these challenges, we propose MemMA, a plug-and-play multi-agent framework that coordinates the memory cycle along both the forward and backward paths. On the forward path, a Meta-Thinker produces structured guidance that steers a Memory Manager during construction and directs a Query Reasoner during iterative retrieval. On the backward path, MemMA introduces in-situ self-evolving memory construction, which synthesizes probe QA pairs, verifies the current memory, and converts failures into repair actions before the memory is finalized. Extensive experiments on LoCoMo show that MemMA consistently outperforms existing baselines across multiple LLM backbones and improves three different storage backends in a plug-and-play manner. Our code is publicly available at https://github.com/ventr1c/memma.

---

### Holter-to-Sleep: AI-Enabled Repurposing of Single-Lead ECG for Sleep Phenotyping

Authors: Donglin Xie, Qingshuo Zhao, Jingyu Wang, Shijia Geng, Jiarui Jin, Jun Li, Rongrong Guo, Guangkun Nie, Gongzheng Tang, Yuxi Zhou, Thomas Penzel, Shenda Hong
Published: 2026-03-19
Categories: eess.SP, cs.LG
arXiv: https://arxiv.org/abs/2603.18714v1

Sleep disturbances are tightly linked to cardiovascular risk, yet polysomnography (PSG)-the clinical reference standard-remains resource-intensive and poorly suited for multi-night, home-based, and large-scale screening. Single-lead electrocardiography (ECG), already ubiquitous in Holter and patch-based devices, enables comfortable long-term acquisition and encodes sleep-relevant physiology through autonomic modulation and cardiorespiratory coupling. Here, we present a proof-of-concept Holter-to-Sleep framework that, using single-lead ECG as the sole input, jointly supports overnight sleep phenotyping and Holter-grade cardiac phenotyping within the same recording, and further provides an explicit analytic pathway for scalable cardio-sleep association studies. The framework is developed and validated on a pooled multi-center PSG sample of 10,439 studies spanning four public cohorts, with independent external evaluation to assess cross-cohort generalizability, and additional real-world feasibility assessment using overnight patch-ECG recordings via objective-subjective consistency analysis. This integrated design enables robust extraction of clinically meaningful overnight sleep phenotypes under heterogeneous populations and acquisition conditions, and facilitates systematic linkage between ECG-derived sleep metrics and arrhythmia-related Holter phenotypes. Collectively, the Holter-to-Sleep paradigm offers a practical foundation for low-burden, home-deployable, and scalable cardio-sleep monitoring and research beyond traditional PSG-centric workflows.

---

### Accurate and Efficient Multi-Channel Time Series Forecasting via Sparse Attention Mechanism

Authors: Lei Gao, Hengda Bao, Jingfei Fang, Guangzheng Wu, Weihua Zhou, Yun Zhou
Published: 2026-03-19
Categories: cs.AI
arXiv: https://arxiv.org/abs/2603.18712v1

The task of multi-channel time series forecasting is ubiquitous in numerous fields such as finance, supply chain management, and energy planning. It is critical to effectively capture complex dynamic dependencies within and between channels for accurate predictions. However, traditional method paid few attentions on learning the interaction among channels. This paper proposes Linear-Network (Li-Net), a novel architecture designed for multi-channel time series forecasting that captures the linear and non-linear dependencies among channels. Li-Net dynamically compresses representations across sequence and channel dimensions, processes the information through a configurable non-linear module and subsequently reconstructs the forecasts. Moreover, Li-Net integrates a sparse Top-K Softmax attention mechanism within a multi-scale projection framework to address these challenges. A core innovation is its ability to seamlessly incorporate and fuse multi-modal embeddings, guiding the sparse attention process to focus on the most informative time steps and feature channels. Through the experiment results on multiple real-world benchmark datasets demonstrate that Li-Net achieves competitive performance compared to state-of-the-art baseline methods. Furthermore, Li-Net provides a superior balance between prediction accuracy and computational burden, exhibiting significantly lower memory usage and faster inference times. Detailed ablation studies and parameter sensitivity analyses validate the effectiveness of each key component in our proposed architecture. Keywords: Multivariate Time Series Forecasting, Sparse Attention Mechanism, Multimodal Information Fusion, Non-linear relationship

---