# Research Papers — 2026-03 · bundle page 146 of 243

> Full-text markdown of papers #7251–7300 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/146
[← Previous bundle](https://aipod.dev/papers/archive/2026-03/page/145.md)
[Next bundle →](https://aipod.dev/papers/archive/2026-03/page/147.md)

### A Prediction-as-Perception Framework for 3D Object Detection

Authors: Song Zhang, Haoyu Chen, Ruibo Wang
Published: 2026-03-13
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.12599v1

Humans combine prediction and perception to observe the world. When faced with rapidly moving birds or insects, we can only perceive them clearly by predicting their next position and focusing our gaze there. Inspired by this, this paper proposes the Prediction-As-Perception (PAP) framework, integrating a prediction-perception architecture into 3D object perception tasks to enhance the model's perceptual accuracy. The PAP framework consists of two main modules: prediction and perception, primarily utilizing continuous frame information as input. Firstly, the prediction module forecasts the potential future positions of ego vehicles and surrounding traffic participants based on the perception results of the current frame. These predicted positions are then passed as queries to the perception module of the subsequent frame. The perceived results are iteratively fed back into the prediction module. We evaluated the PAP structure using the end-to-end model UniAD on the nuScenes dataset. The results demonstrate that the PAP structure improves UniAD's target tracking accuracy by 10% and increases the inference speed by 15%. This indicates that such a biomimetic design significantly enhances the efficiency and accuracy of perception models while reducing computational resource consumption.

---

### Neural Gate: Mitigating Privacy Risks in LVLMs via Neuron-Level Gradient Gating

Authors: Xiangkui Cao, Jie Zhang, Meina Kan, Shiguang Shan, Xilin Chen
Published: 2026-03-13
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.12598v2

Large Vision-Language Models (LVLMs) have shown remarkable potential across a wide array of vision-language tasks, leading to their adoption in critical domains such as finance and healthcare. However, their growing deployment also introduces significant security and privacy risks. Malicious actors could potentially exploit these models to extract sensitive information, highlighting a critical vulnerability. Recent studies show that LVLMs often fail to consistently refuse instructions designed to compromise user privacy. While existing work on privacy protection has made meaningful progress in preventing the leakage of sensitive data, they are constrained by limitations in both generalization and non-destructiveness. They often struggle to robustly handle unseen privacy-related queries and may inadvertently degrade a model's performance on standard tasks. To address these challenges, we introduce Neural Gate, a novel method for mitigating privacy risks through neuron-level model editing. Our method improves a model's privacy safeguards by increasing its rate of refusal for privacy-related questions, crucially extending this protective behavior to novel sensitive queries not encountered during the editing process. Neural Gate operates by learning a feature vector to identify neurons associated with privacy-related concepts within the model's representation of a subject. This localization then precisely guides the update of model parameters. Through comprehensive experiments on MiniGPT and LLaVA, we demonstrate that our method significantly boosts the model's privacy protection while preserving its original utility. The code is available at https://github.com/Xiangkui-Cao/Neural-Gate.

---

### Feynman: Knowledge-Infused Diagramming Agent for Scalable Visual Designs

Authors: Zixin Wen, Yifu Cai, Kyle Lee, Sam Estep, Josh Sunshine, Aarti Singh, Yuejie Chi, Wode Ni
Published: 2026-03-13
Categories: cs.LG, cs.AI, cs.HC, cs.MA, cs.SE
arXiv: https://arxiv.org/abs/2603.12597v1

Visual design is an essential application of state-of-the-art multi-modal AI systems. Improving these systems requires high-quality vision-language data at scale. Despite the abundance of internet image and text data, knowledge-rich and well-aligned image-text pairs are rare. In this paper, we present a scalable diagram generation pipeline built with our agent, Feynman. To create diagrams, Feynman first enumerates domain-specific knowledge components (''ideas'') and performs code planning based on the ideas. Given the plan, Feynman translates ideas into simple declarative programs and iterates to receives feedback and visually refine diagrams. Finally, the declarative programs are rendered by the Penrose diagramming system. The optimization-based rendering of Penrose preserves the visual semantics while injecting fresh randomness into the layout, thereby producing diagrams with visual consistency and diversity. As a result, Feynman can author diagrams along with grounded captions with very little cost and time. Using Feynman, we synthesized a dataset with more than 100k well-aligned diagram-caption pairs. We also curate a visual-language benchmark, Diagramma, from freshly generated data. Diagramma can be used for evaluating the visual reasoning capabilities of vision-language models. We plan to release the dataset, benchmark, and the full agent pipeline as an open-source project.

---

### Optimize Wider, Not Deeper: Consensus Aggregation for Policy Optimization

Authors: Zelal Su, Mustafaoglu, Sungyoung Lee, Eshan Balachandar, Risto Miikkulainen, Keshav Pingali
Published: 2026-03-13
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2603.12596v1

Proximal policy optimization (PPO) approximates the trust region update using multiple epochs of clipped SGD. Each epoch may drift further from the natural gradient direction, creating path-dependent noise. To understand this drift, we can use Fisher information geometry to decompose policy updates into signal (the natural gradient projection) and waste (the Fisher-orthogonal residual that consumes trust region budget without first-order surrogate improvement). Empirically, signal saturates but waste grows with additional epochs, creating an optimization-depth dilemma. We propose Consensus Aggregation for Policy Optimization (CAPO), which redirects compute from depth to width: $K$ PPO replicates are optimized on the same batch, differing only in minibatch shuffling order, and then aggregated into a consensus. We study aggregation in two spaces: Euclidean parameter space, and the natural parameter space of the policy distribution via the logarithmic opinion pool. In natural parameter space, the consensus provably achieves higher KL-penalized surrogate and tighter trust region compliance than the mean expert; parameter averaging inherits these guarantees approximately. On continuous control tasks, CAPO outperforms PPO and compute-matched deeper baselines under fixed sample budgets by up to 8.6x. CAPO demonstrates that policy optimization can be improved by optimizing wider, rather than deeper, without additional environment interactions.

---

### Swap-guided Preference Learning for Personalized Reinforcement Learning from Human Feedback

Authors: Gihoon Kim, Euntai Kim
Published: 2026-03-13
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2603.12595v1

Reinforcement Learning from Human Feedback (RLHF) is a widely used approach to align large-scale AI systems with human values. However, RLHF typically assumes a single, universal reward, which overlooks diverse preferences and limits personalization. Variational Preference Learning (VPL) seeks to address this by introducing user-specific latent variables. Despite its promise, we found that VPL suffers from posterior collapse. While this phenomenon is well known in VAEs, it has not previously been identified in preference learning frameworks. Under sparse preference data and with overly expressive decoders, VPL may cause latent variables to be ignored, reverting to a single-reward model. To overcome this limitation, we propose Swap-guided Preference Learning (SPL). The key idea is to construct fictitious swap annotators and use the mirroring property of their preferences to guide the encoder. SPL introduces three components: (1) swap-guided base regularization, (2) Preferential Inverse Autoregressive Flow (P-IAF), and (3) adaptive latent conditioning. Experiments show that SPL mitigates collapse, enriches user-specific latents, and improves preference prediction. Our code and data are available at https://github.com/cobang0111/SPL

---

### Maximizing Incremental Information Entropy for Contrastive Learning

Authors: Jiansong Zhang, Zhuoqin Yang, Xu Wu, Xiaoling Luo, Peizhong Liu, Linlin Shen
Published: 2026-03-13
Categories: cs.LG
arXiv: https://arxiv.org/abs/2603.12594v1

Contrastive learning has achieved remarkable success in self-supervised representation learning, often guided by information-theoretic objectives such as mutual information maximization. Motivated by the limitations of static augmentations and rigid invariance constraints, we propose IE-CL (Incremental-Entropy Contrastive Learning), a framework that explicitly optimizes the entropy gain between augmented views while preserving semantic consistency. Our theoretical framework reframes the challenge by identifying the encoder as an information bottleneck and proposes a joint optimization of two components: a learnable transformation for entropy generation and an encoder regularizer for its preservation. Experiments on CIFAR-10/100, STL-10, and ImageNet demonstrate that IE-CL consistently improves performance under small-batch settings. Moreover, our core modules can be seamlessly integrated into existing frameworks. This work bridges theoretical principles and practice, offering a new perspective in contrastive learning.

---

### Early Pruning for Public Transport Routing

Authors: Andrii Rohovyi, Abdallah Abuaisha, Toby Walsh
Published: 2026-03-13
Categories: cs.DS, cs.AI, cs.RO
arXiv: https://arxiv.org/abs/2603.12592v4

Routing algorithms for public transport, particularly the widely used RAPTOR and its variants, often face performance bottlenecks during the transfer relaxation phase, especially on dense transfer graphs, when supporting unlimited transfers. This inefficiency arises from iterating over many potential inter-stop connections (walks, bikes, e-scooters, etc.). To maintain acceptable performance, practitioners often limit transfer distances or exclude certain transfer options, which can reduce path optimality and restrict the multimodal options presented to travellers. This paper introduces Early Pruning, a low-overhead technique that accelerates routing algorithms without compromising optimality. By pre-sorting transfer connections by duration and applying a pruning rule within the transfer loop, the method discards longer transfers at a stop once they cannot yield an earlier arrival than the current best solution. Early Pruning can be integrated with minimal changes to existing codebases and requires only a one-time preprocessing step. The technique preserves Pareto-optimality in extended-criteria settings whenever the additional optimization criteria are monotonically non-decreasing in transfer duration. Across multiple state-of-the-art RAPTOR-based solutions, including RAPTOR, ULTRA-RAPTOR, McRAPTOR, BM-RAPTOR, ULTRA-McRAPTOR, and UBM-RAPTOR and tested on the Switzerland and London transit networks, we achieved query time reductions of up to 57\%. This approach provides a generalizable improvement to the efficiency of transit pathfinding algorithms.

---

### CA-HFP: Curvature-Aware Heterogeneous Federated Pruning with Model Reconstruction

Authors: Gang Hu, Yinglei Teng, Pengfei Wu, Shijun Ma
Published: 2026-03-13
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2603.12591v1

Federated learning on heterogeneous edge devices requires personalized compression while preserving aggregation compatibility and stable convergence. We present Curvature-Aware Heterogeneous Federated Pruning (CA-HFP), a practical framework that enables each client perform structured, device-specific pruning guided by a curvature-informed significance score, and subsequently maps its compact submodel back into a common global parameter space via a lightweight reconstruction. We derive a convergence bound for federated optimization with multiple local SGD steps that explicitly accounts for local computation, data heterogeneity, and pruning-induced perturbations; from which a principled loss-based pruning criterion is derived. Extensive experiments on FMNIST, CIFAR-10, and CIFAR-100 using VGG and ResNet architectures under varying degrees of data heterogeneity demonstrate that CA-HFP preserves model accuracy while significantly reducing per-client computation and communication costs, outperforming standard federated training and existing pruning-based baselines.

---

### SDF-Net: Structure-Aware Disentangled Feature Learning for Opticall-SAR Ship Re-identification

Authors: Furui Chen, Han Wang, Yuhan Sun, Jianing You, Yixuan Lv, Zhuang Zhou, Hong Tan, Shengyang Li
Published: 2026-03-13
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.12588v2

Cross-modal ship re-identification (ReID) between optical and synthetic aperture radar (SAR) imagery is fundamentally challenged by the severe radiometric discrepancy between passive optical imaging and coherent active radar sensing. While existing approaches primarily rely on statistical distribution alignment or semantic matching, they often overlook a critical physical prior: ships are rigid objects whose geometric structures remain stable across sensing modalities, whereas texture appearance is highly modality-dependent. In this work, we propose SDF-Net, a Structure-Aware Disentangled Feature Learning Network that systematically incorporates geometric consistency into optical--SAR ship ReID. Built upon a ViT backbone, SDF-Net introduces a structure consistency constraint that extracts scale-invariant gradient energy statistics from intermediate layers to robustly anchor representations against radiometric variations. At the terminal stage, SDF-Net disentangles the learned representations into modality-invariant identity features and modality-specific characteristics. These decoupled cues are then integrated through a parameter-free additive residual fusion, effectively enhancing discriminative power. Extensive experiments on the HOSS-ReID dataset demonstrate that SDF-Net consistently outperforms existing state-of-the-art methods. The code and trained models are publicly available at https://github.com/cfrfree/SDF-Net.

---

### MRGeo: Robust Cross-View Geo-Localization of Corrupted Images via Spatial and Channel Feature Enhancement

Authors: Le Wu, Lv Bo, Songsong Ouyang, Yingying Zhu
Published: 2026-03-13
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.12587v1

Cross-view geo-localization (CVGL) aims to accurately localize street-view images through retrieval of corresponding geo-tagged satellite images. While prior works have achieved nearly perfect performance on certain standard datasets, their robustness in real-world corrupted environments remains under-explored. This oversight causes severe performance degradation or failure when images are affected by corruption such as blur or weather, significantly limiting practical deployment. To address this critical gap, we introduce MRGeo, the first systematic method designed for robust CVGL under corruption. MRGeo employs a hierarchical defense strategy that enhances the intrinsic quality of features and then enforces a robust geometric prior. Its core is the Spatial-Channel Enhancement Block, which contains: (1) a Spatial Adaptive Representation Module that models global and local features in parallel and uses a dynamic gating mechanism to arbitrate their fusion based on feature reliability; and (2) a Channel Calibration Module that performs compensatory adjustments by modeling multi-granularity channel dependencies to counteract information loss. To prevent spatial misalignment under severe corruption, a Region-level Geometric Alignment Module imposes a geometric structure on the final descriptors, ensuring coarse-grained consistency. Comprehensive experiments on both robustness benchmark and standard datasets demonstrate that MRGeo not only achieves an average R@1 improvement of 2.92\% across three comprehensive robustness benchmarks (CVUSA-C-ALL, CVACT\_val-C-ALL, and CVACT\_test-C-ALL) but also establishes superior performance in cross-area evaluation, thereby demonstrating its robustness and generalization capability.

---

### Deferred is Better: A Framework for Multi-Granularity Deferred Interaction of Heterogeneous Features

Authors: Yi Xu, Moyu Zhang, Chaofan Fan, Jinxin Hu, Yu Zhang, Xiaoyi Zeng
Published: 2026-03-13
Categories: cs.IR, cs.LG
arXiv: https://arxiv.org/abs/2603.12586v1

Click-through rate (CTR) prediction models estimates the probability of a user-item click by modeling interactions across a vast feature space. A fundamental yet often overlooked challenge is the inherent heterogeneity of these features: their sparsity and information content vary dramatically. For instance, categorical features like item IDs are extremely sparse, whereas numerical features like item price are relatively dense. Prevailing CTR models have largely ignored this heterogeneity, employing a uniform feature interaction strategy that inputs all features into the interaction layers simultaneously. This approach is suboptimal, as the premature introduction of low-information features can inject significant noise and mask the signals from information-rich features, which leads to model collapse and hinders the learning of robust representations. To address the above challenge, we propose a Multi-Granularity Information-Aware Deferred Interaction Network (MGDIN), which adaptively defers the introduction of features into the feature interaction process. MGDIN's core mechanism operates in two stages: First, it employs a multi-granularity feature grouping strategy to partition the raw features into distinct groups with more homogeneous information density in different granularities, thereby mitigating the effects of extreme individual feature sparsity and enabling the model to capture feature interactions from diverse perspectives. Second, a delayed interaction mechanism is implemented through a hierarchical masking strategy, which governs when and how each group participates by masking low-information groups in the early layers and progressively unmasking them as the network deepens. This deferred introduction allows the model to establish a robust understanding based on high-information features before gradually incorporating sparser information from other groups...

---

### RTD-Guard: A Black-Box Textual Adversarial Detection Framework via Replacement Token Detection

Authors: He Zhu, Yanshu Li, Wen Liu, Haitian Yang
Published: 2026-03-13
Categories: cs.CL, cs.CR
arXiv: https://arxiv.org/abs/2603.12582v1

Textual adversarial attacks pose a serious security threat to Natural Language Processing (NLP) systems by introducing imperceptible perturbations that mislead deep learning models. While adversarial example detection offers a lightweight alternative to robust training, existing methods typically rely on prior knowledge of attacks, white-box access to the victim model, or numerous queries, which severely limits their practical deployment. This paper introduces RTD-Guard, a novel black-box framework for detecting textual adversarial examples. Our key insight is that word-substitution perturbations in adversarial attacks closely resemble the "replaced tokens" that a Replaced Token Detection (RTD) discriminator is pre-trained to identify. Leveraging this, RTD-Guard employs an off-the-shelf RTD discriminator-without fine-tuning-to localize suspicious tokens, masks them, and detects adversarial examples by observing the prediction confidence shift of the victim model before and after intervention. The entire process requires no adversarial data, model tuning, or internal model access, and uses only two black-box queries. Comprehensive experiments on multiple benchmark datasets demonstrate that RTD-Guard effectively detects adversarial texts generated by diverse state-of-the-art attack methods. It surpasses existing detection baselines across multiple metrics, offering a highly efficient, practical, and resource-light defense mechanism-particularly suited for real-world deployment in resource-constrained or privacy-sensitive environments.

---

### Multiscale Structure-Guided Latent Diffusion for Multimodal MRI Translation

Authors: Jianqiang Lin, Zhiqiang Shen, Peng Cao, Jinzhu Yang, Osmar R. Zaiane, Xiaoli Liu
Published: 2026-03-13
Categories: eess.IV, cs.AI, cs.CV
arXiv: https://arxiv.org/abs/2603.12581v1

Although diffusion models have achieved remarkable progress in multi-modal magnetic resonance imaging (MRI) translation tasks, existing methods still tend to suffer from anatomical inconsistencies or degraded texture details when handling arbitrary missing-modality scenarios. To address these issues, we propose a latent diffusion-based multi-modal MRI translation framework, termed MSG-LDM. By leveraging the available modalities, the proposed method infers complete structural information, which preserves reliable boundary details. Specifically, we introduce a style--structure disentanglement mechanism in the latent space, which explicitly separates modality-specific style features from shared structural representations, and jointly models low-frequency anatomical layouts and high-frequency boundary details in a multi-scale feature space. During the structure disentanglement stage, high-frequency structural information is explicitly incorporated to enhance feature representations, guiding the model to focus on fine-grained structural cues while learning modality-invariant low-frequency anatomical representations. Furthermore, to reduce interference from modality-specific styles and improve the stability of structure representations, we design a style consistency loss and a structure-aware loss. Extensive experiments on the BraTS2020 and WMH datasets demonstrate that the proposed method outperforms existing MRI synthesis approaches, particularly in reconstructing complete structures. The source code is publicly available at https://github.com/ziyi-start/MSG-LDM.

---

### DINOLight: Robust Ambient Light Normalization with Self-supervised Visual Prior Integration

Authors: Youngjin Oh, Junhyeong Kwon, Nam Ik Cho
Published: 2026-03-13
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.12579v1

This paper presents a new ambient light normalization framework, DINOLight, that integrates the self-supervised model DINOv2's image understanding capability into the restoration process as a visual prior. Ambient light normalization aims to restore images degraded by non-uniform shadows and lighting caused by multiple light sources and complex scene geometries. We observe that DINOv2 can reliably extract both semantic and geometric information from a degraded image. Based on this observation, we develop a novel framework to utilize DINOv2 features for lighting normalization. First, we propose an adaptive feature fusion module that combines features from different DINOv2 layers using a point-wise softmax mask. Next, the fused features are integrated into our proposed restoration network in both spatial and frequency domains through an auxiliary cross-attention mechanism. Experiments show that DINOLight achieves superior performance on the Ambient6K dataset, and that DINOv2 features are effective for enhancing ambient light normalization. We also apply our method to shadow-removal benchmark datasets, achieving competitive results compared to methods that use mask priors. Codes will be released upon acceptance.

---

### Expert Pyramid Tuning: Efficient Parameter Fine-Tuning for Expertise-Driven Task Allocation

Authors: Jia-Chen Zhang, Zhen-Wei Yan, Yu-Jie Xiong, Chun-Ming Xia
Published: 2026-03-13
Categories: cs.CL, cs.CV
arXiv: https://arxiv.org/abs/2603.12577v1

Parameter-Efficient Fine-Tuning (PEFT) has become a dominant paradigm for deploying LLMs in multi-task scenarios due to its extreme parameter efficiency. While Mixture-of-Experts (MoE) based LoRA variants have achieved promising results by dynamically routing tokens to different low-rank experts, they largely overlook the hierarchical nature of task complexity. Existing methods typically employ experts with uniform architectures, limiting their ability to capture diverse feature granularities required by distinct tasks--where some tasks demand high-level semantic abstraction while others require fine-grained syntactic manipulation. To bridge this gap, we propose Expert Pyramid Tuning (EPT), a novel architecture that integrates the multi-scale feature pyramid concept from computer vision into the realm of PEFT. Unlike standard LoRA, EPT decomposes task adaptation into two stages: (1) A shared meta-knowledge Subspace that encodes universal linguistic patterns in low dimensions; (2) A Pyramid Projection Mechanism that utilizes learnable up-projection operators to reconstruct high-dimensional features at varying scales. A task-aware router then dynamically selects the optimal combination of these multi-scale features. Extensive experiments across multiple multi-task benchmarks demonstrate that EPT significantly outperforms SOTA MoE-LoRA variants. Crucially, thanks to the re-parameterization capability of our design, EPT achieves this performance improvement while simultaneously reducing the number of training parameters.

---

### A Spectral Revisit of the Distributional Bellman Operator under the Cramér Metric

Authors: Keru Wang, Yixin Deng, Yao Lyu, Stephen Redmond, Shengbo Eben Li
Published: 2026-03-13
Categories: cs.LG
arXiv: https://arxiv.org/abs/2603.12576v1

Distributional reinforcement learning (DRL) studies the evolution of full return distributions under Bellman updates rather than focusing on expected values. A classical result is that the distributional Bellman operator is contractive under the Cramér metric, which corresponds to an $L^2$ geometry on differences of cumulative distribution functions (CDFs). While this contraction ensures stability of policy evaluation, existing analyses remain largely metric, focusing on contraction properties without elucidating the structural action of the Bellman update on distributions. In this work, we analyse distributional Bellman dynamics directly at the level of CDFs, treating the Cramér geometry as the intrinsic analytical setting. At this level, the Bellman update acts affinely on CDFs and linearly on differences between CDFs, and its contraction property yields a uniform bound on this linear action. Building on this intrinsic formulation, we construct a family of regularised spectral Hilbert representations that realise the CDF-level geometry by exact conjugation, without modifying the underlying Bellman dynamics. The regularisation affects only the geometry and vanishes in the zero-regularisation limit, recovering the native Cramér metric. This framework clarifies the operator structure underlying distributional Bellman updates and provides a foundation for further functional and operator-theoretic analyses in DRL.

---

### AccelAes: Accelerating Diffusion Transformers for Training-Free Aesthetic-Enhanced Image Generation

Authors: Xuanhua Yin, Chuanzhi Xu, Haoxian Zhou, Boyu Wei, Weidong Cai
Published: 2026-03-13
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.12575v2

Diffusion Transformers (DiTs) are a dominant backbone for high-fidelity text-to-image generation due to strong scalability and alignment at high resolutions. However, quadratic self-attention over dense spatial tokens leads to high inference latency and limits deployment. We observe that denoising is spatially non-uniform with respect to aesthetic descriptors in the prompt. Regions associated with aesthetic tokens receive concentrated cross-attention and show larger temporal variation, while low-affinity regions evolve smoothly with redundant computation. Based on this insight, we propose AccelAes, a training-free framework that accelerates DiTs through aesthetics-aware spatio-temporal reduction while improving perceptual aesthetics. AccelAes builds AesMask, a one-shot aesthetic focus mask derived from prompt semantics and cross-attention signals. When localized computation is feasible, SkipSparse reallocates computation and guidance to masked regions. We further reduce temporal redundancy using a lightweight step-level prediction cache that periodically replaces full Transformer evaluations. Experiments on representative DiT families show consistent acceleration and improved aesthetics-oriented quality. On Lumina-Next, AccelAes achieves a 2.11$\times$ speedup and improves ImageReward by +11.9% over the dense baseline. Code is available at https://github.com/xuanhuayin/AccelAes.

---

### LMEB: Long-horizon Memory Embedding Benchmark

Authors: Xinping Zhao, Xinshuo Hu, Jiaxin Xu, Danyu Tang, Xin Zhang, Mengjia Zhou, Yan Zhong, Yao Zhou, Zifei Shan, Meishan Zhang, Baotian Hu, Min Zhang
Published: 2026-03-13
Categories: cs.CL
arXiv: https://arxiv.org/abs/2603.12572v6

Memory embeddings are crucial for memory-augmented systems, such as OpenClaw, but their evaluation is underexplored in current text embedding benchmarks, which narrowly focus on traditional passage retrieval and fail to assess models' ability to handle long-horizon memory retrieval tasks involving fragmented, context-dependent, and temporally distant information. To address this gap, we introduce the Long-horizon Memory Embedding Benchmark (LMEB), a comprehensive framework for evaluating embedding models on complex, long-horizon memory retrieval. LMEB comprises 22 datasets and 193 zero-shot retrieval tasks spanning four memory types: episodic, dialogue, semantic, and procedural. These memory types differ in terms of level of abstraction and temporal dependency, capturing distinct aspects of memory retrieval that reflect the diverse challenges of the real world. We evaluate 15 widely used embedding models, ranging from hundreds of millions to ten billion parameters. The results reveal that (1) LMEB provides a reasonable level of difficulty; (2) Larger models do not always perform better; (3) LMEB and MTEB measure orthogonal capabilities. This suggests that the field has yet to converge on a universal model capable of excelling across all memory retrieval tasks, and that strong performance on traditional passage retrieval does not necessarily transfer to long-horizon memory retrieval. LMEB provides a standardized and reproducible framework that fills a key gap in memory embedding evaluation and supports future advances in long-term, context-dependent retrieval.

---

### Foundation-Model Surrogates Enable Data-Efficient Active Learning for Materials Discovery

Authors: Jeffrey Hu, Rongzhi Dong, Ying Feng, Ming Hu, Jianjun Hu
Published: 2026-03-13
Categories: cond-mat.mtrl-sci, cs.LG
arXiv: https://arxiv.org/abs/2603.12567v3

Active learning (AL) has emerged as a powerful paradigm for accelerating materials discovery by iteratively steering experiments toward promising candidates, reducing the number of costly synthesis-and-characterization cycles needed to identify optimal materials. However, current AL relies predominantly on Gaussian Process (GP) and Random Forest (RF) surrogates, which suffer from complementary limitations: GP underfits complex composition-property landscapes due to rigid kernel assumptions, while RF produces unreliable heuristic uncertainty estimates in small-data regimes. This small-data challenge is pervasive in materials science, making reliable surrogate modeling extremely difficult with models trained from scratch on each new dataset. Here we propose In-Context Active Learning (ICAL), which addresses this bottleneck by replacing conventional surrogates with TabPFN, a transformer-based foundation model (FM) pre-trained on millions of synthetic regression tasks to meta-learn a universal prior over tabular data, upon which TabPFN performs principled Bayesian inference in a single forward pass without dataset-specific retraining, delivering strong small-data regression performance and well-calibrated predictive uncertainty (required for effective AL). We benchmark ICAL against GP and RF across 10 materials datasets and TabPFN wins on 8 out of 10 datasets, achieving a mean saving of 52% in extra evaluations relative to GP and 29.77% relative to RF. Cross-validation analysis confirms that TabPFN's advantage stems from superior uncertainty calibration, achieving the lowest Negative Log-Likelihood and Area Under the Sparsification Error curve among all surrogates. These results demonstrate that pre-trained FMs can serve as effective surrogates for active learning, enabling data-efficient discovery across diverse materials systems and small-data experimental sciences.

---

### Speech-Worthy Alignment for Japanese SpeechLLMs via Direct Preference Optimization

Authors: Mengjie Zhao, Lianbo Liu, Yusuke Fujita, Hao Shi, Yuan Gao, Roman Koshkin, Yui Sudo
Published: 2026-03-13
Categories: cs.SD, cs.CL
arXiv: https://arxiv.org/abs/2603.12565v1

SpeechLLMs typically combine ASR-trained encoders with text-based LLM backbones, leading them to inherit written-style output patterns unsuitable for text-to-speech synthesis. This mismatch is particularly pronounced in Japanese, where spoken and written registers differ substantially in politeness markers, sentence-final particles, and syntactic complexity. We propose a preference-based alignment approach to adapt Japanese SpeechLLMs for speech-worthy outputs: text that is concise, conversational, and readily synthesized as natural speech. To rigorously evaluate this task, we introduce SpokenElyza, a benchmark for Japanese speech-worthiness derived from ELYZA-tasks-100 with auditory verification by native experts. Experiments show that our approach achieves substantial improvement on SpokenElyza while largely preserving performance on the original written-style evaluation. We will release SpokenElyza to support future research on Japanese spoken dialog systems.

---

### Sell Me This Stock: Unsafe Recommendation Drift in LLM Agents

Authors: Zekun Wu, Adriano Koshiyama, Sahan Bulathwela, Maria Perez-Ortiz
Published: 2026-03-13
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2603.12564v8

People increasingly use LLM agents for multi-turn financial recommendations, where the agent pulls market data through tools and tracks user preferences across turns. When tool outputs are manipulated, the recommendations stop matching the user's stated risk profile, but because standard metrics like NDCG only score general relevance, risky and safe stocks score alike, so the metric says nothing went wrong. We call this gap evaluation blindness. We replay 23-turn financial advisory conversations across eight language models, running each dialogue twice with clean and manipulated tool data. Quality scores stay nearly identical to clean sessions while the agents produce risk-mismatched recommendations in 65-99% of turns, unanimous across all eight models. The mechanism is visible turn-by-turn: 80% of risk-score citations across 1,840 turns reproduce the manipulated value verbatim, not a single turn pushes back, and safe-language framing of high-risk stocks ranges from 14% (Qwen2.5-7B) to 69% (Claude Sonnet 4.6). The property that makes frontier models good agents, faithfully grounding their reasoning in tool outputs, also makes them follow manipulated ones. The damage is not memory-driven: contaminating only the current turn still produces 95% of the violations. The model internally distinguishes the manipulation (sparse autoencoder features separate adversarial from random perturbations), but this does not translate into safer output. Activation-level interventions recover under 6% of the safety gap, prompt-level self-verification fails because the self-check reads the same manipulated data, and a parametric cross-check that flags contamination at 99-100% per turn on a frontier model still leaves aggregate suitability unchanged: the agent identifies the tampering and recommends it anyway.

---

### Variational Garrote for Sparse Inverse Problems

Authors: Kanghun Lee, Hyungjoon Soh, Junghyo Jo
Published: 2026-03-13
Categories: stat.ML, cs.CV, cs.LG
arXiv: https://arxiv.org/abs/2603.12562v1

Sparse regularization plays a central role in solving inverse problems arising from incomplete or corrupted measurements. Different regularizers correspond to different prior assumptions about the structure of the unknown signal, and reconstruction performance depends on how well these priors match the intrinsic sparsity of the data. This work investigates the effect of sparsity priors in inverse problems by comparing conventional L1 regularization with the Variational Garrote (VG), a probabilistic method that approximates L0 sparsity through variational binary gating variables. A unified experimental framework is constructed across multiple reconstruction tasks including signal resampling, signal denoising, and sparse-view computed tomography. To enable consistent comparison across models with different parameterizations, regularization strength is swept across wide ranges and reconstruction behavior is analyzed through train-generalization error curves. Experiments reveal characteristic bias-variance tradeoff patterns across tasks and demonstrate that VG frequently achieves lower minimum generalization error and improved stability in strongly underdetermined regimes where accurate support recovery is critical. These results suggest that sparsity priors closer to spike-and-slab structure can provide advantages when the underlying coefficient distribution is strongly sparse. The study highlights the importance of prior-data alignment in sparse inverse problems and provides empirical insights into the behavior of variational L0-type methods across different information bottlenecks.

---

### Lyapunov Stable Graph Neural Flow

Authors: Haoyu Chu, Xiaotong Chen, Wei Zhou, Wenjun Cui, Kai Zhao, Shikui Wei, Qiyu Kang
Published: 2026-03-13
Categories: cs.LG, cs.CV
arXiv: https://arxiv.org/abs/2603.12557v1

Graph Neural Networks (GNNs) are highly vulnerable to adversarial perturbations in both topology and features, making the learning of robust representations a critical challenge. In this work, we bridge GNNs with control theory to introduce a novel defense framework grounded in integer- and fractional-order Lyapunov stability. Unlike conventional strategies that rely on resource-heavy adversarial training or data purification, our approach fundamentally constrains the underlying feature-update dynamics of the GNN. We propose an adaptive, learnable Lyapunov function paired with a novel projection mechanism that maps the network's state into a stable space, thereby offering theoretically provable stability guarantees. Notably, this mechanism is orthogonal to existing defenses, allowing for seamless integration with techniques like adversarial training to achieve cumulative robustness. Extensive experiments demonstrate that our Lyapunov-stable graph neural flows substantially outperform base neural flows and state-of-the-art baselines across standard benchmarks and various adversarial attack scenarios.

---

### Scaling Laws and Pathologies of Single-Layer PINNs: Network Width and PDE Nonlinearity

Authors: Faris Chaudhry
Published: 2026-03-13
Categories: cs.LG, math.NA, physics.comp-ph
arXiv: https://arxiv.org/abs/2603.12556v1

We establish empirical scaling laws for Single-Layer Physics-Informed Neural Networks on canonical nonlinear PDEs. We identify a dual optimization failure: (i) a baseline pathology, where the solution error fails to decrease with network width, even at fixed nonlinearity, falling short of theoretical approximation bounds, and (ii) a compounding pathology, where this failure is exacerbated by nonlinearity. We provide quantitative evidence that a simple separable power law is insufficient, and that the scaling behavior is governed by a more complex, non-separable relationship. This failure is consistent with the concept of spectral bias, where networks struggle to learn the high-frequency solution components that intensify with nonlinearity. We show that optimization, not approximation capacity, is the primary bottleneck, and propose a methodology to empirically measure these complex scaling effects.

---

### Reinforcement Learning for Diffusion LLMs with Entropy-Guided Step Selection and Stepwise Advantages

Authors: Vishnu Teja Kunde, Fatemeh Doudi, Mahdi Farahbakhsh, Dileep Kalathil, Krishna Narayanan, Jean-Francois Chamberland
Published: 2026-03-13
Categories: cs.LG, cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2603.12554v2

Reinforcement learning (RL) has been effective for post-training autoregressive (AR) language models, but extending these methods to diffusion language models (DLMs) is challenging due to intractable sequence-level likelihoods. Existing approaches therefore rely on surrogate likelihoods or heuristic approximations, which can introduce bias and obscure the sequential structure of denoising. We formulate diffusion-based sequence generation as a finite-horizon Markov decision process over the denoising trajectory and derive an exact, unbiased policy gradient that decomposes over denoising steps and is expressed in terms of intermediate advantages, without requiring explicit evaluation of the sequence likelihood. To obtain a practical and compute-efficient estimator, we (i) select denoising steps for policy updates via an entropy-guided approximation bound, and (ii) estimate intermediate advantages using a one-step denoising reward naturally provided by the diffusion model, avoiding costly multi-step rollouts. Experiments on coding and logical reasoning benchmarks demonstrate state-of-the-art results, with strong competitive performance on mathematical reasoning, outperforming existing RL post-training approaches for DLMs. Code is available at https://github.com/vishnutez/egspo-dllm-rl.

---

### Beyond Dense Futures: World Models as Structured Planners for Robotic Manipulation

Authors: Minghao Jin, Mozheng Liao, Mingfei Han, Zhihui Li, Xiaojun Chang
Published: 2026-03-13
Categories: cs.RO, cs.CV
arXiv: https://arxiv.org/abs/2603.12553v1

Recent world-model-based Vision-Language-Action (VLA) architectures have improved robotic manipulation through predictive visual foresight. However, dense future prediction introduces visual redundancy and accumulates errors, causing long-horizon plan drift. Meanwhile, recent sparse methods typically represent visual foresight using high-level semantic subtasks or implicit latent states. These representations often lack explicit kinematic grounding, weakening the alignment between planning and low-level execution. To address this, we propose StructVLA, which reformulates a generative world model into an explicit structured planner for reliable control. Instead of dense rollouts or semantic goals, StructVLA predicts sparse, physically meaningful structured frames. Derived from intrinsic kinematic cues (e.g., gripper transitions and kinematic turning points), these frames capture spatiotemporal milestones closely aligned with task progress. We implement this approach through a two-stage training paradigm with a unified discrete token vocabulary: the world model is first trained to predict structured frames and subsequently optimized to map the structured foresight into low-level actions. This approach provides clear physical guidance and bridges visual planning and motion control. In our experiments, StructVLA achieves strong average success rates of 75.0% on SimplerEnv-WidowX and 94.8% on LIBERO. Real-world deployments further demonstrate reliable task completion and robust generalization across both basic pick-and-place and complex long-horizon tasks.

---

### Asymptotic and Finite-Time Guarantees for Langevin-Based Temperature Annealing in InfoNCE

Authors: Faris Chaudhry
Published: 2026-03-13
Categories: cs.LG, math.OC, stat.ML
arXiv: https://arxiv.org/abs/2603.12552v1

The InfoNCE loss in contrastive learning depends critically on a temperature parameter, yet its dynamics under fixed versus annealed schedules remain poorly understood. We provide a theoretical analysis by modeling embedding evolution under Langevin dynamics on a compact Riemannian manifold. Under mild smoothness and energy-barrier assumptions, we show that classical simulated annealing guarantees extend to this setting: slow logarithmic inverse-temperature schedules ensure convergence in probability to a set of globally optimal representations, while faster schedules risk becoming trapped in suboptimal minima. Our results establish a link between contrastive learning and simulated annealing, providing a principled basis for understanding and tuning temperature schedules.

---

### CVGL: Causal Learning and Geometric Topology

Authors: Songsong Ouyang, Yingying Zhu
Published: 2026-03-13
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.12551v1

Cross-view geo-localization (CVGL) aims to estimate the geographic location of a street image by matching it with a corresponding aerial image. This is critical for autonomous navigation and mapping in complex real-world scenarios. However, the task remains challenging due to significant viewpoint differences and the influence of confounding factors. To tackle these issues, we propose the Causal Learning and Geometric Topology (CLGT) framework, which integrates two key components: a Causal Feature Extractor (CFE) that mitigates the influence of confounding factors by leveraging causal intervention to encourage the model to focus on stable, task-relevant semantics; and a Geometric Topology Fusion (GT Fusion) module that injects Bird's Eye View (BEV) road topology into street features to alleviate cross-view inconsistencies caused by extreme perspective changes. Additionally, we introduce a Data-Adaptive Pooling (DA Pooling) module to enhance the representation of semantically rich regions. Extensive experiments on CVUSA, CVACT, and their robustness-enhanced variants (CVUSA-C-ALL and CVACT-C-ALL) demonstrate that CLGT achieves state-of-the-art performance, particularly under challenging real-world corruptions. Our codes are available at https://github.com/oyss-szu/CLGT.

---

### Decoding Matters: Efficient Mamba-Based Decoder with Distribution-Aware Deep Supervision for Medical Image Segmentation

Authors: Fares Bougourzi, Fadi Dornaika, Abdenour Hadid
Published: 2026-03-13
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.12547v1

Deep learning has achieved remarkable success in medical image segmentation, often reaching expert-level accuracy in delineating tumors and tissues. However, most existing approaches remain task-specific, showing strong performance on individual datasets but limited generalization across diverse imaging modalities. Moreover, many methods focus primarily on the encoder, relying on large pretrained backbones that increase computational complexity. In this paper, we propose a decoder-centric approach for generalized 2D medical image segmentation. The proposed Deco-Mamba follows a U-Net-like structure with a Transformer-CNN-Mamba design. The encoder combines a CNN block and Transformer backbone for efficient feature extraction, while the decoder integrates our novel Co-Attention Gate (CAG), Vision State Space Module (VSSM), and deformable convolutional refinement block to enhance multi-scale contextual representation. Additionally, a windowed distribution-aware KL-divergence loss is introduced for deep supervision across multiple decoding stages. Extensive experiments on diverse medical image segmentation benchmarks yield state-of-the-art performance and strong generalization capability while maintaining moderate model complexity. The source code will be released upon acceptance.

---

### Spatial Reasoning is Not a Free Lunch: A Controlled Study on LLaVA

Authors: Nahid Alam, Leema Krishna Murali, Siddhant Bharadwaj, Patrick Liu, Timothy Chung, Drishti Sharma, Akshata A., Kranthi Kiran, Wesley Tam, Bala Krishna S Vegesna
Published: 2026-03-13
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.12545v2

Vision-language models (VLMs) have advanced rapidly, yet they still struggle with basic spatial reasoning. Despite strong performance on general benchmarks, modern VLMs remain brittle at understanding 2D spatial relationships such as relative position, layout, and counting. We argue that this failure is not merely a data problem, but is closely tied to dominant design choices in current VLM pipelines: reliance on CLIP-style image encoders and the flattening of images into 1D token sequences with 1D positional encoding. We present a controlled diagnostic study within the LLaVA framework to isolate how these choices affect spatial grounding. We evaluate frontier models and LLaVA variants on a suite of spatial benchmarks, comparing CLIP-based encoders against alternatives trained with denser or generative objectives, as well as variants augmented with 2D positional encoding. Our results show consistent spatial performance gaps across models, and indicate that encoder objectives and positional structure shape spatial behavior, but do not fully resolve it.

---

### Deep Distance Measurement Method for Unsupervised Multivariate Time Series Similarity Retrieval

Authors: Susumu Naito, Kouta Nakata, Yasunori Taguchi
Published: 2026-03-13
Categories: cs.LG
arXiv: https://arxiv.org/abs/2603.12544v1

We propose the Deep Distance Measurement Method (DDMM) to improve retrieval accuracy in unsupervised multivariate time series similarity retrieval. DDMM enables learning of minute differences within states in the entire time series and thereby recognition of minute differences between states, which are of interest to users in industrial plants. To achieve this, DDMM uses a learning algorithm that assigns a weight to each pair of an anchor and a positive sample, arbitrarily sampled from the entire time series, based on the Euclidean distance within the pair and learns the differences within the pairs weighted by the weights. This algorithm allows both learning minute differences within states and sampling pairs from the entire time series. Our empirical studies showed that DDMM significantly outperformed state-of-the-art time series representation learning methods on the Pulp-and-paper mill dataset and demonstrated the effectiveness of DDMM in industrial plants. Furthermore, we showed that accuracy can be further improved by linking DDMM with existing feature extraction methods through experiments with the combined model.

---

### CALF: Communication-Aware Learning Framework for Distributed Reinforcement Learning

Authors: Carlos Purves, Pietro Lio'
Published: 2026-03-13
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2603.12543v1

Distributed reinforcement learning policies face network delays, jitter, and packet loss when deployed across edge devices and cloud servers. Standard RL training assumes zero-latency interaction, causing severe performance degradation under realistic network conditions. We introduce CALF (Communication-Aware Learning Framework), which trains policies under realistic network models during simulation. Systematic experiments demonstrate that network-aware training substantially reduces deployment performance gaps compared to network-agnostic baselines. Distributed policy deployments across heterogeneous hardware validate that explicitly modelling communication constraints during training enables robust real-world execution. These findings establish network conditions as a major axis of sim-to-real transfer for Wi-Fi-like distributed deployments, complementing physics and visual domain randomisation.

---

### As Language Models Scale, Low-order Linear Depth Dynamics Emerge

Authors: Buddhika Nettasinghe, Geethu Joseph
Published: 2026-03-13
Categories: cs.LG, eess.SY
arXiv: https://arxiv.org/abs/2603.12541v1

Large language models are often viewed as high-dimensional nonlinear systems and treated as black boxes. Here, we show that transformer depth dynamics admit accurate low-order linear surrogates within context. Across tasks including toxicity, irony, hate speech and sentiment, a 32-dimensional linear surrogate reproduces the layerwise sensitivity profile of GPT-2-large with near-perfect agreement, capturing how the final output shifts under additive injections at each layer. We then uncover a surprising scaling principle: for a fixed-order linear surrogate, agreement with the full model improves monotonically with model size across the GPT-2 family. This linear surrogate also enables principled multi-layer interventions that require less energy than standard heuristic schedules when applied to the full model. Together, our results reveal that as language models scale, low-order linear depth dynamics emerge within contexts, offering a systems-theoretic foundation for analyzing and controlling them.

---

### Embedded Quantum Machine Learning in Embedded Systems: Feasibility, Hybrid Architectures, and Quantum Co-Processors

Authors: Somdip Dey, Syed Muhammad Raza
Published: 2026-03-13
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2603.12540v1

Embedded quantum machine learning (EQML) seeks to bring quantum machine learning (QML) capabilities to resource-constrained edge platforms such as IoT nodes, wearables, drones, and cyber-physical controllers. In 2026, EQML is technically feasible only in limited and highly experimental forms: (i) hybrid workflows where an embedded device performs sensing and classical processing while offloading a narrowly scoped quantum subroutine to a remote quantum processing unit (QPU) or nearby quantum appliance, and (ii) early-stage "embedded QPU" concepts in which a compact quantum co-processor is integrated with classical control hardware. A practical bridge is quantum-inspired machine learning and optimisation on classical embedded processors and FPGAs. This paper analyses feasibility from a circuits-and-systems perspective aligned with the academic community, formalises two implementation pathways, identifies the dominant barriers (latency, data encoding overhead, NISQ noise, tooling mismatch, and energy), and maps them to concrete engineering directions in interface design, control electronics, power management, verification, and security. We also argue that responsible deployment requires adversarial evaluation and governance practices that are increasingly necessary for edge AI systems.

---

### Spatio-Semantic Expert Routing Architecture with Mixture-of-Experts for Referring Image Segmentation

Authors: Alaa Dalaq, Muzammil Behzad
Published: 2026-03-13
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2603.12538v1

Referring image segmentation aims to produce a pixel-level mask for the image region described by a natural-language expression. Although pretrained vision-language models have improved semantic grounding, many existing methods still rely on uniform refinement strategies that do not fully match the diverse reasoning requirements of referring expressions. Because of this mismatch, predictions often contain fragmented regions, inaccurate boundaries, or even the wrong object, especially when pretrained backbones are frozen for computational efficiency. To address these limitations, we propose SERA, a Spatio-Semantic Expert Routing Architecture for referring image segmentation. SERA introduces lightweight, expression-aware expert refinement at two complementary stages within a vision-language framework. First, we design SERA-Adapter, which inserts an expression-conditioned adapter into selected backbone blocks to improve spatial coherence and boundary precision through expert-guided refinement and cross-modal attention. We then introduce SERA-Fusion, which strengthens intermediate visual representations by reshaping token features into spatial grids and applying geometry-preserving expert transformations before multimodal interaction. In addition, a lightweight routing mechanism adaptively weights expert contributions while remaining compatible with pretrained representations. To make this routing stable under frozen encoders, SERA uses a parameter-efficient tuning strategy that updates only normalization and bias terms, affecting less than 1% of the backbone parameters. Experiments on standard referring image segmentation benchmarks show that SERA consistently outperforms strong baselines, with especially clear gains on expressions that require accurate spatial localization and precise boundary delineation.

---

### Do You See What I Am Pointing At? Gesture-Based Egocentric Video Question Answering

Authors: Yura Choi, Roy Miles, Rolandos Alexandros Potamias, Ismail Elezi, Jiankang Deng, Stefanos Zafeiriou
Published: 2026-03-13
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.12533v2

Understanding and answering questions based on a user's pointing gesture is essential for next-generation egocentric AI assistants. However, current Multimodal Large Language Models (MLLMs) struggle with such tasks due to the lack of gesture-rich data and their limited ability to infer fine-grained pointing intent from egocentric video. To address this, we introduce EgoPointVQA, a dataset and benchmark for gesture-grounded egocentric question answering, comprising 4000 synthetic and 400 real-world videos across multiple deictic reasoning tasks. Built upon it, we further propose Hand Intent Tokens (HINT), which encodes tokens derived from 3D hand keypoints using an off-the-shelf reconstruction model and interleaves them with the model input to provide explicit spatial and temporal context for interpreting pointing intent. We show that our model outperforms others in different backbones and model sizes. In particular, HINT-14B achieves 68.1% accuracy, on average over 6 tasks, surpassing the state-of-the-art, InternVL3-14B, by 6.6%. To further facilitate the open research, we will release the code, model, and dataset. Project page: https://yuuraa.github.io/papers/choi2026egovqa

---

### Mixing Makes Markovian Contexts Cheap for Linear Bandits

Authors: Kaan Buyukkalayci, Osama Hanna, Christina Fragouli
Published: 2026-03-13
Categories: cs.LG
arXiv: https://arxiv.org/abs/2603.12530v3

Recent work shows that when contexts are drawn i.i.d., linear contextual bandits can be reduced to single-context linear bandits. This ``contexts are cheap'' perspective is highly advantageous, as it allows for sharper finite-time analyses and leverages mature techniques from the linear bandit literature, such as those for misspecification and adversarial corruption. However, this reduction crucially relies on the independence of contexts and does not extend to settings with temporally correlated (e.g., Markovian) contexts, which arise frequently in practice. Motivated by applications with temporally correlated availability, we extend this perspective to linear bandits with Markovian context processes, where the action set evolves via an exogenous Markov chain. Our main contribution is a reduction that applies under uniform geometric ergodicity. We construct a stationary surrogate action set to solve the problem using a standard linear bandit oracle, employing a delayed-update scheme to control the bias induced by the nonstationary conditional context distributions. We further provide a phased algorithm for unknown stationary distributions that learns the surrogate mapping online. In both settings, we obtain a high-probability worst-case regret bound matching that of the underlying linear bandit oracle in sufficiently fast mixing regimes. We then validate our results on a real-world instance, where we show practical gains over a LinUCB baseline.

---

### Mixing Makes Markovian Contexts Cheap for Linear Bandits

Authors: Kaan Buyukkalayci, Osama Hanna, Christina Fragouli
Published: 2026-03-13
Categories: cs.LG
arXiv: https://arxiv.org/abs/2603.12530v2

Recent work shows that when contexts are drawn i.i.d., linear contextual bandits can be reduced to single-context linear bandits. This ``contexts are cheap'' perspective is highly advantageous, as it allows for sharper finite-time analyses and leverages mature techniques from the linear bandit literature, such as those for misspecification and adversarial corruption. However, this reduction crucially relies on the independence of contexts and does not extend to settings with temporally correlated (e.g., Markovian) contexts, which arise frequently in practice. Motivated by applications with temporally correlated availability, we extend this perspective to linear bandits with Markovian context processes, where the action set evolves via an exogenous Markov chain. Our main contribution is a reduction that applies under uniform geometric ergodicity. We construct a stationary surrogate action set to solve the problem using a standard linear bandit oracle, employing a delayed-update scheme to control the bias induced by the nonstationary conditional context distributions. We further provide a phased algorithm for unknown stationary distributions that learns the surrogate mapping online. In both settings, we obtain a high-probability worst-case regret bound matching that of the underlying linear bandit oracle in sufficiently fast mixing regimes. We then validate our results on a real-world instance, where we show practical gains over a LinUCB baseline.

---

### TERMINATOR: Learning Optimal Exit Points for Early Stopping in Chain-of-Thought Reasoning

Authors: Alliot Nagle, Jakhongir Saydaliev, Dhia Garbaya, Michael Gastpar, Ashok Vardhan Makkuva, Hyeji Kim
Published: 2026-03-13
Categories: cs.LG, cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2603.12529v2

Large Reasoning Models (LRMs) achieve impressive performance on complex reasoning tasks via Chain-of-Thought (CoT) reasoning, which enables them to generate intermediate thinking tokens before arriving at the final answer. However, LRMs often suffer from significant overthinking, spending excessive compute time even after the answer is generated early on. Prior work has identified the existence of an optimal reasoning length such that truncating reasoning at this point significantly shortens CoT outputs with virtually no change in performance. However, determining optimal CoT lengths for practical datasets is highly non-trivial as they are fully task and model-dependent. In this paper, we precisely address this and design Terminator, an early-exit strategy for LRMs at inference to mitigate overthinking. The central idea underpinning Terminator is that the first arrival of an LRM's final answer is often predictable, and we leverage these first answer positions to create a novel dataset of optimal reasoning lengths to train Terminator. Powered by this approach, Terminator achieves significant reductions in CoT lengths of 14%-55% on average across four challenging practical datasets: MATH-500, AIME 2025, HumanEval, and GPQA, while outperforming current state-of-the-art methods and reducing inference latency by more than 2x compared to the original LRM.

---

### Mixed Integer Goal Programming for Personalized Meal Optimization with User-Defined Serving Granularity

Authors: Francisco Aguilera Moreno
Published: 2026-03-12
Categories: cs.AI
arXiv: https://arxiv.org/abs/2605.13849v1

Determining what to eat to satisfy nutritional requirements is one of the oldest optimization problems in operations research, yet existing formulations have two persistent limitations: continuous variables produce impractical fractional servings (1.7 eggs, 0.37 bananas), and hard nutrient constraints cause infeasibility when targets conflict. A systematic review of 56 diet optimization papers found that none combine integer programming with goal programming to address both issues. We propose Mixed Integer Goal Programming (MIGP) for personalized meal optimization. The formulation uses integer variables for practical serving counts and goal programming deviations for soft nutrient targets, with inverse-target normalization to balance multi-nutrient optimization. Per-food serving granularity allows natural units (one egg, one tablespoon of oil) without post-hoc rounding. We characterize the integrality gap in the goal programming context and identify a deviation absorption property: GP deviation variables buffer the cost of requiring integer servings, making the gap structurally smaller than in hard-constraint MIP. For meals with 15+ foods, the integer solution matches the continuous optimum in every benchmark instance. A computational evaluation across 810 instances (30 USDA foods, 9 configurations, 3 methods) shows MIGP finds strictly better solutions than GP with post-hoc rounding in 66% of cases (never worse) while maintaining 100% feasibility; hard-constraint IP achieves only 48%. Solve times stay under 100 ms for typical meal sizes using the open-source HiGHS solver. The implementation is available as an open-source Python module integrated into an interactive meal planning application.

---

### Behavioral Intelligence Platforms: From Event Streams to Autonomous Insight via Probabilistic Journey Graphs, Behavioral Knowledge Extraction, and Grounded Language Generation

Authors: Arun Patra, Bhushan Vadgave
Published: 2026-03-12
Categories: cs.IR, cs.AI
arXiv: https://arxiv.org/abs/2604.22762v2

Contemporary product analytics systems require users to pose explicit queries, such as writing SQL, configuring dashboards, or constructing funnels, before insights can surface. This pull-based paradigm creates a bottleneck: it requires both domain knowledge and technical fluency, and assumes practitioners know in advance which questions to ask. We argue that behavioral analytics should move from passive systems that answer queries to active systems that continuously detect and explain behavioral phenomena. We present the Behavioral Intelligence Platform (BIP), a system architecture that transforms raw event streams into automatically generated insights. BIP consists of four layers. First, Normalization and State Derivation (NSD) standardizes events and maps them to a semantic state hierarchy. Second, a Behavioral Graph Engine (BGE) models user journeys as absorbing Markov chains and computes transition probabilities, removal effects, and path quality metrics. Third, a Behavioral Knowledge Graph (BKG) and Detector System convert graph outputs into grounded behavioral facts and identify behavioral phenomena. Finally, a Grounded Language Layer constrains large language model outputs to verified facts, producing reliable narrative insights. We formalize the Behavioral Intelligence Problem, introduce a taxonomy of detectors for autonomous insight generation, and propose an interestingness score to prioritize insights under limited attention.

---

### Preventing overfitting in deep learning using differential privacy

Authors: Alizishaan Anwar Hussein Khatri
Published: 2026-03-12
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2604.16334v1

The use of Deep Neural Network based systems in the real world is growing. They have achieved state-of-the-art performance on many image, speech and text datasets. They have been shown to be powerful systems that are capable of learning detailed relationships and abstractions from the data. This is a double-edged sword which makes such systems vulnerable to learning the noise in the training set, thereby negatively impacting performance. This is also known as the problem of \emph{overfitting} or \emph{poor generalization}. In a practical setting, analysts typically have limited data to build models that must generalize to unseen data. In this work, we explore the use of a differential-privacy based approach to improve generalization in Deep Neural Networks.

---

### A Discordance-Aware Multimodal Framework with Multi-Agent Clinical Reasoning

Authors: Pegah Ahadian, Mingrui Yang, Sixu Chen, Xiaojuan Li, Qiang Guan
Published: 2026-03-12
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2604.16333v1

Knee osteoarthritis frequently exhibits discordance between structural damage observed in imaging and patient-reported symptoms such as pain. This mismatch complicates clinical interpretation and patient stratification and remains insufficiently modeled in existing decision support systems. We propose a discordance aware multimodal framework that combines machine learning prediction models with a tool grounded multi agent reasoning system. Using baseline data from the FNIH Osteoarthritis Biomarkers Consortium, we trained multimodal models to predict two progression tasks, joint space loss only progression versus non progression, and pain only progression versus non progression. The predictive system integrates three modality specific experts: a CatBoost tabular model using demographic, radiographic, MRI-derived scalar, and biomarker features; MRI image embeddings extracted using a ResNet18 backbone; and Xray embeddings derived from the same architecture. Expert predictions are fused using a stacking ensemble. Residual based models estimate expected pain from structural features, enabling the computation of a pain structure discordance score between observed and expected symptoms. A multi-agent reasoning layer interprets these signals to assign clinically interpretable OA phenotypes and generate phenotype specific management recommendations.

---

### Annotation Entropy Predicts Per-Example Learning Dynamics in LoRA Fine-Tuning

Authors: Brady Steele
Published: 2026-03-12
Categories: cs.LG, cs.CL
arXiv: https://arxiv.org/abs/2604.16332v1

We find that LoRA fine-tuning exhibits un-learning on contested examples: items with high annotator disagreement show increasing loss during training, a qualitatively distinct pattern largely absent under full fine-tuning and consistent across all six models tested (four encoder, two decoder-only). This discovery emerges from correlating annotation entropy, computed from ChaosNLI's 100 labels per example, with per-example area under the loss curve (AULC) on SNLI and MNLI. The correlation is positive in all 25 conditions tested (Spearman $ρ= 0.06$-$0.43$), with decoder-only models showing stronger correlations than encoders at matched LoRA rank. The effect survives partial-correlation controls and replicates across seeds and datasets. A preliminary noise-injection experiment is consistent with these findings.

---

### BrainMem: Brain-Inspired Evolving Memory for Embodied Agent Task Planning

Authors: Xiaoyu Ma, Lianyu Hu, Wenbing Tang, Zixuan Hu, Zeqin Liao, Zhizhen Wu, Yang Liu
Published: 2026-03-12
Categories: cs.RO, cs.AI, cs.CV, cs.MA
arXiv: https://arxiv.org/abs/2604.16331v1

Embodied task planning requires agents to execute long-horizon, goal-directed actions in complex 3D environments, where success depends on both immediate perception and accumulated experience across tasks. However, most existing LLM-based planners are stateless and reactive, operating without persistent memory and therefore repeating errors and struggling with spatial or temporal dependencies. We propose BrainMem(Brain-Inspired Evolving Memory), a training-free hierarchical memory system that equips embodied agents with working, episodic, and semantic memory inspired by human cognition. BrainMem continuously transforms interaction histories into structured knowledge graphs and distilled symbolic guidelines, enabling planners to retrieve, reason over, and adapt behaviors from past experience without any model fine-tuning or additional training. This plug-and-play design integrates seamlessly with arbitrary multi-modal LLMs and greatly reduces reliance on task-specific prompt engineering. Extensive experiments on four representative benchmarks, including EB-ALFRED, EB-Navigation, EB-Manipulation, and EB-Habitat, demonstrate that BrainMem significantly enhances task success rates across diverse models and difficulty subsets, with the largest gains observed on long-horizon and spatially complex tasks. These results highlight evolving memory as a promising and scalable mechanism for generalizable embodied intelligence.

---

### MRGEN: A Conceptual Framework for LLM-Powered Mixed Reality Authoring Tools for Education

Authors: Mohammed Oussama Seddini, Mohamed Ez-Zaouia, Ngoc Luyen Le, Iza Marfisi
Published: 2026-03-12
Categories: cs.HC, cs.AI
arXiv: https://arxiv.org/abs/2604.15341v1

Mixed Reality (MR) offers immersive and multimodal opportunities for education but remains difficult for teachers to author without technical expertise. We propose MRGEN, a conceptual framework for LLM-powered authoring tools to support teachers in creating MR learning activities that work on mobile devices (tablets and smartphones). MRGEN articulates three axes: Learning Objectives, MR Modality, and GAI Assistance. To validate our framework, we implemented a prototype based on the open-source MIXAP authoring platform and conducted a user study with 24 participants. Results show that LLM-powered authoring reduced task duration by 36% on average, and that over 90% of participants found the AI support helpful for brainstorming, structuring, and aligning content with their learning goals. These findings yielded very promising results for future AI-assisted MR authoring tools.

---

### Beyond Theory of Mind in Robotics

Authors: Malte F. Jung
Published: 2026-03-12
Categories: cs.AI, cs.HC
arXiv: https://arxiv.org/abs/2604.09612v1

Theory of Mind, the capacity to explain and predict behavior by inferring hidden mental states, has become the dominant paradigm for social interaction in robotics. Yet ToM rests on three assumptions that poorly capture how most social interaction actually unfolds: that meaning travels inside-out from hidden states to observable behavior; that understanding requires detached inference rather than participation; and that the meaning of behavior is fixed and available to a passive observer. Drawing on ethnomethodology, conversation analysis, and participatory sense-making, I argue that social meaning is not decoded from behavior but produced through moment-to-moment coordination between agents. This interactional foundation has direct implications for robot design: shifting from internal state modeling toward policies for sustaining coordination, from observer-based inference toward active participation, and from fixed behavioral meaning toward meaning potential stabilized through response.

---

### Characterizing Performance-Energy Trade-offs of Large Language Models in Multi-Request Workflows

Authors: Md. Monzurul Amin Ifath, Israat Haque
Published: 2026-03-12
Categories: cs.DC, cs.AI
arXiv: https://arxiv.org/abs/2604.09611v1

Large language models (LLMs) are increasingly used in applications forming multi-request workflows like document summarization, search-based copilots, and multi-agent programming. While these workflows unlock richer functionality, they also amplify latency and energy demand during inference. Existing measurement and benchmarking efforts either focus on assessing LLM inference systems or consider single-request evaluations, overlooking workflow dependencies and cross-request interactions unique to multi-request workflows. Moreover, the energy usage of such interdependent LLM calls remains underexplored. To address these gaps, this paper presents the first systematic characterization of performance-energy trade-offs in multi-request LLM inference. We develop four representative workloads capturing sequential, interactive, agentic, and composite patterns common in modern deployments. Using an NVIDIA A100 testbed with state-of-the-art serving systems (vLLM and Parrot), we analyze how key energy knobs affect latency, throughput, and component-level energy use. Our findings reveal batch size as the most impactful lever, though benefits are workload dependent. While optimal batching benefits workloads with large shared prompts, it is ineffective for sequential summarization and only partially effective for multi-agent coding. GPU power capping provides modest but predictable savings, while output length induces linear energy scaling with limited efficiency gains. We further show that engine-level optimizations in vLLM maintain higher GPU utilization and efficiency, especially for decode-heavy workloads, while Parrot's workflow-aware scheduling achieves lower energy consumption under strict power constraints. These findings offer actionable guidelines for developers and system operators designing performance- and energy-aware LLM serving systems in emerging multi-request workflows.

---

### Temporally Phenotyping GLP-1RA Case Reports with Large Language Models: A Textual Time Series Corpus and Risk Modeling

Authors: Sayantan Kumar, Jeremy C. Weiss
Published: 2026-03-12
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2604.06197v1

Type 2 diabetes case reports describe complex clinical courses, but their timelines are often expressed in language that is difficult to reuse in longitudinal modeling. To address this gap, we developed a textual time-series corpus of 136 PubMed Open Access single-patient case reports involving glucagon-like peptide 1 receptor agonists, with clinical events associated with their most probable reference times. We evaluated automated LLM timeline extraction against gold-standard timelines annotated by clinical domain experts, assessing how well systems recovered clinical events and their timings. The best-performing LLM produced high event coverage (GPT5; 0.871) and reliable temporal sequencing across symptoms (GPT5; 0.843), diagnoses, treatments, laboratory tests, and outcomes. As a downstream demonstration, time-to-event analyses in diabetes suggested lower risk of respiratory sequelae among GLP-1 users versus non-users (HR=0.259, p<0.05), consistent with prior reports of improved respiratory outcomes. Temporal annotations and code will be released upon acceptance.

---

### Compositional Consistency-Guided Decoding for Three-Way Logical Question Answering

Authors: Tianyi Huang, Ming Hou, Jiaheng Su, Yutong Zhang, Ziling Zhang
Published: 2026-03-12
Categories: cs.CL, cs.AI, cs.LO
arXiv: https://arxiv.org/abs/2604.06196v2

Three-way logical question answering (QA) assigns one of $\text{True}$, $\text{False}$, or $\text{Unknown}$ to a hypothesis $H$ given a premise set $S$. We study this task as a compact compositional inference problem: predictions for $H$ and for a mechanically negated hypothesis $\neg H$ should agree under a deterministic negation map. Despite this simple structure, large language models (LLMs) can exhibit two practical failure modes: (i) negation inconsistency, where answers to $H$ and $\neg H$ violate the required label mapping, and (ii) epistemic $\text{Unknown}$, where the model abstains even when one side is entailed. We introduce CGD-PD, a lightweight, training-free test-time layer that combines neural 3-way classification, symbolic negation-consistency projection, and targeted binary entailment probes. On one validation split of FOLIO's first-order logic fields, CGD-PD improves accuracy by 4.4 points on GPT-5.2 and 6.8 points on Claude Sonnet 4.5, while reducing $\text{Unknown}$ predictions and epistemic abstention. These results provide a controlled proof of concept that simple logical composition at inference time can help evaluate and improve LLM reasoning reliability; they do not, by themselves, establish robustness beyond this formal benchmark setting.

---