# Research Papers — 2026-03 · bundle page 232 of 243

> Full-text markdown of papers #11551–11600 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/232
[← Previous bundle](https://aipod.dev/papers/archive/2026-03/page/231.md)
[Next bundle →](https://aipod.dev/papers/archive/2026-03/page/233.md)

### GenDB: The Next Generation of Query Processing -- Synthesized, Not Engineered

Authors: Jiale Lao, Immanuel Trummer
Published: 2026-03-02
Categories: cs.DB, cs.AI, cs.CL, cs.LG, cs.MA
arXiv: https://arxiv.org/abs/2603.02081v1

Traditional query processing relies on engines that are carefully optimized and engineered by many experts. However, new techniques and user requirements evolve rapidly, and existing systems often cannot keep pace. At the same time, these systems are difficult to extend due to their internal complexity, and developing new systems requires substantial engineering effort and cost. In this paper, we argue that recent advances in Large Language Models (LLMs) are starting to shape the next generation of query processing systems. We propose using LLMs to synthesize execution code for each incoming query, instead of continuously building, extending, and maintaining complex query processing engines. As a proof of concept, we present GenDB, an LLM-powered agentic system that generates instance-optimized and customized query execution code tailored to specific data, workloads, and hardware resources. We implemented an early prototype of GenDB that uses Claude Code Agent as the underlying component in the multi-agent system, and we evaluate it on OLAP workloads. We use queries from the well-known TPC-H benchmark and also construct a new benchmark designed to reduce potential data leakage from LLM training data. We compare GenDB with state-of-the-art query engines, including DuckDB, Umbra, MonetDB, ClickHouse, and PostgreSQL. GenDB achieves significantly better performance than these systems. Finally, we discuss the current limitations of GenDB and outline future extensions and related research challenges.

---

### From Pixels to Patches: Pooling Strategies for Earth Embeddings

Authors: Isaac Corley, Caleb Robinson, Inbal Becker-Reshef, Juan M. Lavista Ferres
Published: 2026-03-02
Categories: cs.CV, cs.LG
arXiv: https://arxiv.org/abs/2603.02080v2

Geospatial foundation models increasingly expose pixel-level embedding products that can be downloaded and reused without access to the underlying encoder. In this setting, downstream tasks with patch- or region-level labels require a post-hoc aggregation step that maps dense pixel embeddings to a single representation. The default choice, mean pooling, discards within-patch variability and can underperform under spatial distribution shift. To study this setting, we introduce EuroSAT-Embed: 81,000 embedding GeoTIFFs derived from three foundation models: AlphaEarth, OlmoEarth, and Tessera. Using these fixed embedding products, we benchmark 11 training-free pooling methods and 2 train-set-fitted baselines under both random and geographically disjoint test splits. Richer pooling schemes reduce the geographic generalization gap by over 50% relative to mean pooling and improve accuracy by up to 6% on spatial splits. We recommend a three-tier strategy: (1) mean as a baseline, (2) stats pooling (min/max/mean/std) as the default at 4x the embedding dimension, and (3) covariance pooling for peak accuracy. Across all three embedding products, simple distributional statistics improve spatial-split performance over mean pooling.

---

### MMNavAgent: Multi-Magnification WSI Navigation Agent for Clinically Consistent Whole-Slide Analysis

Authors: Zhengyang Xu, Han Li, Jingsong Liu, Linrui Xie, Xun Ma, Xin You, Shihui Zu, Ayako Ito, Xinyu Hao, Hongming Xu, Shaohua Kevin Zhou, Nassir Navab et al.
Published: 2026-03-02
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.02079v1

Recent AI navigation approaches aim to improve Whole-Slide Image (WSI) diagnosis by modeling spatial exploration and selecting diagnostically relevant regions, yet most operate at a single fixed magnification or rely on predefined magnification traversal. In clinical practice, pathologists examine slides across multiple magnifications and selectively inspect only necessary scales, dynamically integrating global and cellular evidence in a sequential manner. This mismatch prevents existing methods from modeling cross-magnification interactions and adaptive magnification selection inherent to real diagnostic workflows. To these, we propose a clinically consistent Multi-Magnification WSI Navigation Agent (MMNavAgent) that explicitly models multi magnification interaction and adaptive magnification selection. Specifically, we introduce a Cross-Magnification navigation Tool (CMT) that aggregates contextual information from adjacent magnifications to enhance discriminative representations along the navigation path. We further introduce a Magnification Selection Tool (MST) that leverages memory-driven reasoning within the agent framework to enable interactive and adaptive magnification selection, mimicking the sequential decision process of pathologists. Extensive experiments on a public dataset demonstrate improved diagnostic performance, with 1.45% gain of AUC and 2.93% gain of BACC over a non-agent baseline. Code will be public upon acceptance.

---

### Cognitive Prosthetic: An AI-Enabled Multimodal System for Episodic Recall in Knowledge Work

Authors: Lawrence Obiuwevwi, Krzysztof J. Rechowicz, Vikas Ashok, Sachin Shetty, Sampath Jayarathna
Published: 2026-03-02
Categories: cs.HC, cs.AI
arXiv: https://arxiv.org/abs/2603.02072v1

Modern knowledge workplaces increasingly strain human episodic memory as individuals navigate fragmented attention, overlapping meetings, and multimodal information streams. Existing workplace tools provide partial support through note-taking or analytics but rarely integrate cognitive, physiological, and attentional context into retrievable memory representations. This paper presents the Cognitive Prosthetic Multimodal System (CPMS) --an AI-enabled proof-of-concept designed to support episodic recall in knowledge work through structured episodic capture and natural language retrieval. CPMS synchronizes speech transcripts, physiological signals, and gaze behavior into temporally aligned, JSON-based episodic records processed locally for privacy. Beyond data logging, the system includes a web-based retrieval interface that allows users to query past workplace experiences using natural language, referencing semantic content, time, attentional focus, or physiological state. We present CPMS as a functional proof-of-concept demonstrating the technical feasibility of transforming heterogeneous sensor data into queryable episodic memories. The system is designed to be modular, supporting operation with partial sensor configurations, and incorporates privacy safeguards for workplace deployment. This work contributes an end-to-end, privacy-aware architecture for AI-enabled memory augmentation in workplace settings.

---

### Exploring Plan Space through Conversation: An Agentic Framework for LLM-Mediated Explanations in Planning

Authors: Guilhem Fouilhé, Rebecca Eifler, Antonin Poché, Sylvie Thiébaux, Nicholas Asher
Published: 2026-03-02
Categories: cs.AI, cs.CL, cs.HC, cs.MA
arXiv: https://arxiv.org/abs/2603.02070v3

When automating plan generation for a real-world sequential decision problem, the goal is often not to replace the human planner, but to facilitate an iterative reasoning and elicitation process, where the human's role is to guide the AI planner according to their preferences and expertise. In this context, explanations that respond to users' questions are crucial to improve their understanding of potential solutions and increase their trust in the system. To enable natural interaction with such a system, we present a multi-agent Large Language Model (LLM) architecture that is agnostic to the explanation framework and enables user- and context-dependent interactive explanations. We also describe an instantiation of this framework for goal-conflict explanations, which we use to conduct a user study comparing the LLM-powered interaction with a baseline template-based explanation interface.

---

### Scaling Laws of SignSGD in Linear Regression: When Does It Outperform SGD?

Authors: Jihwan Kim, Dogyoon Song, Chulhee Yun
Published: 2026-03-02
Categories: cs.LG, cs.AI, math.OC, stat.ML
arXiv: https://arxiv.org/abs/2603.02069v2

We study scaling laws of signSGD under a power-law random features (PLRF) model that accounts for both feature and target decay. We analyze the population risk of a linear model trained with one-pass signSGD on Gaussian-sketched features. We express the risk as a function of model size, training steps, learning rate, and the feature and target decay parameters. Comparing against the SGD risk analyzed by Paquette et al. (2024), we identify a drift-normalization effect and a noise-reshaping effect unique to signSGD. We then obtain compute-optimal scaling laws under the optimal choice of learning rate. Our analysis shows that the noise-reshaping effect can make the compute-optimal slope of signSGD steeper than that of SGD in regimes where noise is dominant. Finally, we observe that the widely used warmup-stable-decay (WSD) schedule further reduces the noise term and sharpens the compute-optimal slope, when feature decay is fast but target decay is slow.

---

### Scaling Laws of SignSGD in Linear Regression: When Does It Outperform SGD?

Authors: Jihwan Kim, Dogyoon Song, Chulhee Yun
Published: 2026-03-02
Categories: cs.LG, cs.AI, math.OC, stat.ML
arXiv: https://arxiv.org/abs/2603.02069v1

We study scaling laws of signSGD under a power-law random features (PLRF) model that accounts for both feature and target decay. We analyze the population risk of a linear model trained with one-pass signSGD on Gaussian-sketched features. We express the risk as a function of model size, training steps, learning rate, and the feature and target decay parameters. Comparing against the SGD risk analyzed by Paquette et al. (2024), we identify a drift-normalization effect and a noise-reshaping effect unique to signSGD. We then obtain compute-optimal scaling laws under the optimal choice of learning rate. Our analysis shows that the noise-reshaping effect can make the compute-optimal slope of signSGD steeper than that of SGD in regimes where noise is dominant. Finally, we observe that the widely used warmup-stable-decay (WSD) schedule further reduces the noise term and sharpens the compute-optimal slope, when feature decay is fast but target decay is slow.

---

### Accelerating PDE Surrogates via RL-Guided Mesh Optimization

Authors: Yang Meng, Ruoxi Jiang, Zhuokai Zhao, Chong Liu, Rebecca Willett, Yuxin Chen
Published: 2026-03-02
Categories: cs.LG
arXiv: https://arxiv.org/abs/2603.02066v1

Deep surrogate models for parametric partial differential equations (PDEs) can deliver high-fidelity approximations but remain prohibitively data-hungry: training often requires thousands of fine-grid simulations, each incurring substantial computational cost. To address this challenge, we introduce RLMesh, an end-to-end framework for efficient surrogate training under limited simulation budget. The key idea is to use reinforcement learning (RL) to adaptively allocate mesh grid points non-uniformly within each simulation domain, focusing numerical resolution in regions most critical for accurate PDE solutions. A lightweight proxy model further accelerates RL training by providing efficient reward estimates without full surrogate retraining. Experiments on PDE benchmarks demonstrate that RLMesh achieves competitive accuracy to baselines but with substantially fewer simulation queries. These results show that solver-level spatial adaptivity can dramatically improve the efficiency of surrogate training pipelines, enabling practical deployment of learning-based PDE surrogates across a wide range of problems.

---

### Never Saddle for Reparameterized Steepest Descent as Mirror Flow

Authors: Tom Jacobs, Chao Zhou, Rebekka Burkholz
Published: 2026-03-02
Categories: cs.LG
arXiv: https://arxiv.org/abs/2603.02064v1

How does the choice of optimization algorithm shape a model's ability to learn features? To address this question for steepest descent methods --including sign descent, which is closely related to Adam --we introduce steepest mirror flows as a unifying theoretical framework. This framework reveals how optimization geometry governs learning dynamics, implicit bias, and sparsity and it provides two explanations for why Adam and AdamW often outperform SGD in fine-tuning. Focusing on diagonal linear networks and deep diagonal linear reparameterizations (a simplified proxy for attention), we show that steeper descent facilitates both saddle-point escape and feature learning. In contrast, gradient descent requires unrealistically large learning rates to escape saddles, an uncommon regime in fine-tuning. Empirically, we confirm that saddle-point escape is a central challenge in fine-tuning. Furthermore, we demonstrate that decoupled weight decay, as in AdamW, stabilizes feature learning by enforcing novel balance equations. Together, these results highlight two mechanisms how steepest descent can aid modern optimization.

---

### ORGAN: Object-Centric Representation Learning using Cycle Consistent Generative Adversarial Networks

Authors: Joël Küchler, Ellen van Maren, Vaiva Vasiliauskaitė, Katarina Vulić, Reza Abbasi-Asl, Stephan J. Ihle
Published: 2026-03-02
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.02063v2

Although data generation is often straightforward, extracting information from data is more difficult. Object-centric representation learning can extract information from images in an unsupervised manner. It does so by segmenting an image into its subcomponents: the objects. Each object is then represented in a low-dimensional latent space that can be used for downstream processing. Object-centric representation learning is dominated by autoencoder architectures (AEs). Here, we present ORGAN, a novel approach for object-centric representation learning, which is based on cycle-consistent Generative Adversarial Networks instead. We show that it performs similarly to other state-of-the-art approaches on synthetic datasets, while at the same time being the only approach tested here capable of handling more challenging real-world datasets with many objects and low visual contrast. Complementing these results, ORGAN creates expressive latent space representations that allow for object manipulation. Finally, we show that ORGAN scales well both with respect to the number of objects and the size of the images, giving it a unique edge over current state-of-the-art approaches.

---

### OpenRad: a Curated Repository of Open-access AI models for Radiology

Authors: Konstantinos Vrettos, Galini Papadaki, Emmanouil Brilakis, Matthaios Triantafyllou, Dimitrios Leventis, Despina Staraki, Maria Mavroforou, Eleftherios Tzanis, Konstantina Giouroukou, Michail E. Klontzas
Published: 2026-03-02
Categories: cs.AI
arXiv: https://arxiv.org/abs/2603.02062v1

The rapid developments in artificial intelligence (AI) research in radiology have produced numerous models that are scattered across various platforms and sources, limiting discoverability, reproducibility and clinical translation. Herein, OpenRad was created, a curated, standardized, open-access repository that aggregates radiology AI models and providing details such as the availability of pretrained weights and interactive applications. Retrospective analysis of peer reviewed literature and preprints indexed in PubMed, arXiv and Scopus was performed until Dec 2025 (n = 5239 records). Model records were generated using a locally hosted LLM (gpt-oss:120b), based on the RSNA AI Roadmap JSON schema, and manually verified by ten expert reviewers. Stability of LLM outputs was assessed on 225 randomly selected papers using text similarity metrics. A total of 1694 articles were included after review. Included models span all imaging modalities (CT, MRI, X-ray, US) and radiology subspecialties. Automated extraction demonstrated high stability for structured fields (Levenshtein ratio > 90%), with 78.5% of record edits being characterized as minor during expert review. Statistical analysis of the repository revealed CNN and transformer architectures as dominant, while MRI was the most commonly used modality (in 621 neuroradiology AI models). Research output was mostly concentrated in China and the United States. The OpenRad web interface enables model discovery via keyword search and filters for modality, subspecialty, intended use, verification status and demo availability, alongside live statistics. The community can contribute new models through a dedicated portal. OpenRad contains approx. 1700 open access, curated radiology AI models with standardized metadata, supplemented with analysis of code repositories, thereby creating a comprehensive, searchable resource for the radiology community.

---

### TRAKNN: Efficient Trajectory Aware Spatiotemporal kNN for Rare Meteorological Trajectory Detection

Authors: Guillaume Coulaud, Davide Faranda
Published: 2026-03-02
Categories: stat.ML, cs.LG
arXiv: https://arxiv.org/abs/2603.02059v1

Extreme weather events, such as windstorms and heatwaves, are driven by persistent atmospheric circulation patterns that evolve over several consecutive days. While traditional circulation-based studies often focus on instantaneous atmospheric states, capturing the temporal evolution, or trajectory, of these spatial fields is essential for characterizing rare and potentially impactful atmospheric behavior. However, performing an exhaustive similarity search on multi-decadal, continental-scale gridded datasets presents significant computational and memory challenges. In this paper, we propose TRAKNN (TRajectory Aware KNN), a fully unsupervised and data-agnostic framework for detecting geometrically rare short trajectories in spatio-temporal data with an exact kNN approach. TRAKNN leverages a recurrence-based algorithm that decouples computational complexity from trajectory length and efficient batch operations, maximizing computational intensity. These optimizations enable exhaustive analysis on standard workstations, either on CPU or on GPU. We evaluate our approach on 75 years of daily European sea-level pressure data. Our results illustrate that rare trajectories identified by TRAKNN correspond to physically coherent atmospheric anomalies and align with independent extreme-event databases.

---

### A Resource-Rational Principle for Modeling Visual Attention Control

Authors: Yunpeng Bai
Published: 2026-03-02
Categories: cs.HC, cs.AI
arXiv: https://arxiv.org/abs/2603.02056v1

Understanding how people allocate visual attention is central to Human-Computer Interaction (HCI), yet existing computational models of attention are often either descriptive, task-specific, or difficult to interpret. My dissertation develops a resource-rational, simulation-based framework for modeling visual attention as a sequential decision-making process under perceptual, memory, and time constraints. I formalize visual tasks, such as reading and multitasking, as bounded-optimal control problems using Partially Observable Markov Decision Processes, enabling eye-movement behaviors such as fixation and attention switching to emerge from rational adaptation rather than being hand-coded or purely data-driven. These models are instantiated in simulation environments spanning traditional text reading and reading-while-walking with smart glasses, where they reproduce classic empirical effects, explain observed trade-offs between comprehension and safety, and generate novel predictions under time pressure and interface variation. Collectively, this work contributes a unified computational account of visual attention, offering new tools for theory-driven and resource-efficient HCI design.

---

### Strategic Advice in the Age of Personal AI

Authors: Yueyang Liu, Wichinpong Park Sinchaisri
Published: 2026-03-02
Categories: cs.LG, cs.GT, cs.HC
arXiv: https://arxiv.org/abs/2603.02055v2

Personal AI assistants are changing how individuals use advice. We study how an advisor should design its recommendation in anticipation of stochastic consultation with personal AI whose recommendation is predictable. Personal AI enters through two dimensions: consultation probability and relative trust, which captures the relative influence personal AI receives when consulted. In the baseline model, the advisor optimally counteracts the personal AI signal. Counteraction increases with consultation probability but is hump-shaped in relative trust. The advisor's minimized loss is hump-shaped in consultation probability, vanishing when personal AI is never or always consulted. Greater relative trust in personal AI increases the irreducible loss arising from stochastic consultation. We extend the analysis to partial predictability and costly recommendation adjustment, characterizing their effects on optimal recommendations and minimized loss. The framework also accommodates richer information structures, including settings in which personal AI is perceived as having access to private information relevant to the task. We introduce an online forecasting experiment that examines how participants obtain personal AI advice and combine it with an advisor's recommendation and their initial judgments. Participants place weight on all three inputs. When access requires an additional action, some participants do not seek personal AI advice, while some others attempt to obtain it without success. Together, these findings highlight two distinct aspects of personal AI use: whether advice is obtained and how much weight it receives when available.

---

### Strategic Advice in the Age of Personal AI

Authors: Yueyang Liu, Wichinpong Park Sinchaisri
Published: 2026-03-02
Categories: cs.LG, cs.GT, cs.HC
arXiv: https://arxiv.org/abs/2603.02055v1

Personal AI assistants have changed how people use institutional and professional advice. We study this new strategic setting in which individuals may stochastically consult a personal AI whose recommendation is predictable to the focal advisor. Personal AI enters this strategic environment along two dimensions: how often it is consulted and how much weight it receives in the human's decision when consulted. Anticipating this, the advisor responds by counteracting the personal AI recommendation. Counteraction becomes more aggressive as personal AI is consulted more often. Yet advisor performance is non-monotone: equilibrium loss is highest at intermediate levels of adoption and vanishes when personal AI is never used or always used. Trust affects performance through a single relative influence index, and greater relative influence of personal AI increases advisor vulnerability. Extending the framework to costly credibility building, we characterize how personal AI adoption reshapes incentives to invest in trust.

---

### "When to Hand Off, When to Work Together": Understanding Concurrent Human-Agent Interaction in Shared Co-Creative Workspaces

Authors: Kihoon Son, Hyewon Lee, DaEun Choi, Yoonsu Kim, Tae Soo Kim, Yoonjoo Lee, John Joon Young Chung, HyunJoon Jung, Juho Kim
Published: 2026-03-02
Categories: cs.HC, cs.AI
arXiv: https://arxiv.org/abs/2603.02050v5

Shared co-creative workspaces allow users to contribute while agents execute. Yet how users adjust their participation as their work and the agent's execution shape one another remains less understood. We conducted two design probe studies with professional designers. In Study 1 (N=10), participants identified opportunities to contribute by observing the agent's progress, but avoided intervention when the agent misinterpreted their edits. Therefore, we developed CLEO, a second probe that tracks user actions and selectively updates its plan. In Study 2 (N=12), we analyzed 255 interaction turns and stimulated recall interviews, identifying five action categories, ten codes, six triggers, and four enabling factors. Participants engaged in concurrent interaction during 33.73% of turns, expressing newly recognized design ideas or preferences through edits and changing the division of work by completing pending subtasks or independently developing copies of unfinished outputs. We present a decision model and five design implications for concurrent human-agent collaboration.

---

### "When to Hand Off, When to Work Together": Expanding Human-Agent Co-Creative Collaboration through Concurrent Interaction

Authors: Kihoon Son, Hyewon Lee, DaEun Choi, Yoonsu Kim, Tae Soo Kim, Yoonjoo Lee, John Joon Young Chung, HyunJoon Jung, Juho Kim
Published: 2026-03-02
Categories: cs.HC, cs.AI
arXiv: https://arxiv.org/abs/2603.02050v4

As agents move into shared workspaces and their execution becomes visible, human-agent collaboration faces a fundamental shift from sequential delegation to concurrent co-creation. This raises a new coordination problem: what interaction patterns emerge, and what agent capabilities are required to support them? Study 1 (N=10) revealed that process visibility naturally prompted concurrent intervention, but exposed a critical capability gap: agents lacked the collaborative context awareness needed to distinguish user feedback from independent parallel work. This motivated CLEO, a design probe that embodies this capability, interpreting concurrent user actions as feedback or independent work and adapting execution accordingly. Study 2 (N=10) analyzed 214 turn-level interactions, identifying a taxonomy of five action patterns and ten codes, along with six triggers and four enabling factors explaining when and why users shift between collaboration modes. Concurrent interaction appeared in 31.8% of turns. We present a decision model, design implications, and an annotated dataset, positioning concurrent interaction as what makes delegation work better.

---

### WorldStereo: Bridging Camera-Guided Video Generation and Scene Reconstruction via 3D Geometric Memories

Authors: Yisu Zhang, Chenjie Cao, Tengfei Wang, Xuhui Zuo, Junta Wu, Jianke Zhu, Chunchao Guo
Published: 2026-03-02
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.02049v1

Recent advances in foundational Video Diffusion Models (VDMs) have yielded significant progress. Yet, despite the remarkable visual quality of generated videos, reconstructing consistent 3D scenes from these outputs remains challenging, due to limited camera controllability and inconsistent generated content when viewed from distinct camera trajectories. In this paper, we propose WorldStereo, a novel framework that bridges camera-guided video generation and 3D reconstruction via two dedicated geometric memory modules. Formally, the global-geometric memory enables precise camera control while injecting coarse structural priors through incrementally updated point clouds. Moreover, the spatial-stereo memory constrains the model's attention receptive fields with 3D correspondence to focus on fine-grained details from the memory bank. These components enable WorldStereo to generate multi-view-consistent videos under precise camera control, facilitating high-quality 3D reconstruction. Furthermore, the flexible control branch-based WorldStereo shows impressive efficiency, benefiting from the distribution matching distilled VDM backbone without joint training. Extensive experiments across both camera-guided video generation and 3D reconstruction benchmarks demonstrate the effectiveness of our approach. Notably, we show that WorldStereo acts as a powerful world model, tackling diverse scene generation tasks (whether starting from perspective or panoramic images) with high-fidelity 3D results. Models will be released.

---

### NICO-RAG: Multimodal Hypergraph Retrieval-Augmented Generation for Understanding the Nicotine Public Health Crisis

Authors: Manuel Serna-Aguilera, Raegan Anderes, Page Dobbs, Khoa Luu
Published: 2026-03-02
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.02047v1

The nicotine addiction public health crisis continues to be pervasive. In this century alone, the tobacco industry has released and marketed new products in an aggressive effort to lure new and young customers for life. Such innovations and product development, namely flavored nicotine or tobacco such as nicotine pouches, have undone years of anti-tobacco campaign work. Past work is limited both in scope and in its ability to connect large-scale data points. Thus, we introduce the Nicotine Innovation Counter-Offensive (NICO) Dataset to provide public health researchers with over 200,000 multimodal samples, including images and text descriptions, on 55 tobacco and nicotine product brands. In addition, to provide public health researchers with factual connections across a large-scale dataset, we propose NICO-RAG, a retrieval-augmented generation (RAG) framework that can retrieve image features without incurring the high-cost of language models, as well as the added cost of processing image tokens with large-scale datasets such as NICO. At construction time, NICO-RAG organizes image- and text-extracted entities and relations into hypergraphs to produce as factual responses as possible. This joint multimodal knowledge representation enables NICO-RAG to retrieve images for query answering not only by visual similarity but also by the semantic similarity of image descriptions. Experimentals show that without needing to process additional tokens from images for over 100 questions, NICO-RAG performs comparably to the state-of-the-art RAG method adapted for images.

---

### Expanding LLM Agent Boundaries with Strategy-Guided Exploration

Authors: Andrew Szot, Michael Kirchhof, Omar Attia, Alexander Toshev
Published: 2026-03-02
Categories: cs.LG
arXiv: https://arxiv.org/abs/2603.02045v1

Reinforcement learning (RL) has demonstrated notable success in post-training large language models (LLMs) as agents for tasks such as computer use, tool calling, and coding. However, exploration remains a central challenge in RL for LLM agents, especially as they operate in language-action spaces with complex observations and sparse outcome rewards. In this work, we address exploration for LLM agents by leveraging the ability of LLMs to plan and reason in language about the environment to shift exploration from low-level actions to higher-level language strategies. We thus propose Strategy-Guided Exploration (SGE), which first generates a concise natural-language strategy that describes what to do to make progress toward the goal, and then generates environment actions conditioned on that strategy. By exploring in the space of strategies rather than the space of actions, SGE induces structured and diverse exploration that targets different environment outcomes. To increase strategy diversity during RL, SGE introduces mixed-temperature sampling, which explores diverse strategies in parallel, along with a strategy reflection process that grounds strategy generation on the outcomes of previous strategies in the environment. Across UI interaction, tool-calling, coding, and embodied agent environments, SGE consistently outperforms exploration-focused RL baselines, improving both learning efficiency and final performance. We show that SGE enables the agent to learn to solve tasks too difficult for the base model.

---

### Multiplicative Oracle Inequalities for Transductive Learning via Level-Set Aggregation

Authors: Jian Qian, Jiachen Xu
Published: 2026-03-02
Categories: cs.LG, stat.ML
arXiv: https://arxiv.org/abs/2603.02043v2

We revisit transductive learning where predictions are made with the set of all covariates known in advance. In the leave-one-out (LOO) setting, the prediction is made with labels of the remaining sample points and evaluated by the average error. In particular, we study multiplicative oracle inequalities for agnostic transductive LOO prediction for a variety of tasks, including classification with 0-1 loss, squared loss regression, density estimation, and logistic regression. Specifically, we introduce \emph{Median of Level-Set Aggregation} (MLSA), an aggregation procedure built on near-ERM level sets (i.e., empirical-risk level sets around the ERM). We prove a general multiplicative oracle inequality for the LOO error of the form \( LOO_S(MLSA) \;\le\; C \left( \frac{1}{n} \min_{h\in H} L_S(h) \;+\; \frac{\log |H|}{n}\right), \qquad C>1, \) where $H$ is the hypothesis/function class. This inequality holds for hypothesis classes under a local level-set growth condition together with losses satisfying a mild monotonicity assumption. For classification with VC classes under the $0$--$1$ loss, the $\log |H|$ factor can be improved to be $d\log n$, where $d$ is the VC dimension, recovering Long (1998) up to a $\log n$ factor. For logistic regression with bounded covariates and parameters, the $\log |H|$ factor can be improved to be $d\log n$ up to problem-dependent factors, where $d$ is the ambient dimension.

---

### EstLLM: Enhancing Estonian Capabilities in Multilingual LLMs via Continued Pretraining and Post-Training

Authors: Aleksei Dorkin, Taido Purason, Emil Kalbaliyev, Hele-Andra Kuulmets, Marii Ojastu, Mark Fišel, Tanel Alumäe, Eleri Aedmaa, Krister Kruusmaa, Kairit Sirts
Published: 2026-03-02
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2603.02041v2

Large language models (LLMs) are predominantly trained on English-centric data, resulting in uneven performance for smaller languages. We study whether continued pretraining (CPT) can improve Estonian capabilities in multilingual LLMs while preserving English and general reasoning performance. Using Llama 3.1 8B and Apertus 8B as base models, we apply CPT with Estonian-enriched multilingual replay, followed by mostly English supervised fine-tuning, preference optimization, and chat vector merging. Evaluation on Estonian benchmarks, targeted pairwise human evaluation, and an Estonian Chatbot Arena-style setup shows consistent improvements in Estonian language competence, reasoning, translation, and instruction-following. Although Apertus exhibits stronger Estonian capabilities before adaptation, the more English-centric Llama model achieves substantially larger gains after adaptation. While some English capabilities regress relative to the original instruction-tuned models, chat vector merging substantially restores English instruction-following and reasoning performance. These findings suggest that CPT with balanced multilingual replay and lightweight post-training alignment can substantially improve single-language capabilities in multilingual LLMs.

---

### EstLLM: Enhancing Estonian Capabilities in Multilingual LLMs via Continued Pretraining and Post-Training

Authors: Aleksei Dorkin, Taido Purason, Emil Kalbaliyev, Hele-Andra Kuulmets, Marii Ojastu, Mark Fišel, Tanel Alumäe, Eleri Aedmaa, Krister Kruusmaa, Kairit Sirts
Published: 2026-03-02
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2603.02041v1

Large language models (LLMs) are predominantly trained on English-centric data, resulting in uneven performance for smaller languages. We study whether continued pretraining (CPT) can substantially improve Estonian capabilities in a pretrained multilingual LLM while preserving its English and general reasoning performance. Using Llama 3.1 8B as the main base model, we perform CPT on a mixture that increases Estonian exposure while approximating the original training distribution through English replay and the inclusion of code, mathematics, and instruction-like data. We subsequently apply supervised fine-tuning, preference optimization, and chat vector merging to introduce robust instruction-following behavior. Evaluation on a comprehensive suite of Estonian benchmarks shows consistent gains in linguistic competence, knowledge, reasoning, translation quality, and instruction-following compared to the original base model and its instruction-tuned variant, while maintaining competitive performance on English benchmarks. These findings indicate that CPT, with an appropriately balanced data mixture, together with post-training alignment, can substantially improve single-language capabilities in pretrained multilingual LLMs.

---

### Graph neural network force fields for adiabatic dynamics of lattice Hamiltonians

Authors: Yunhao Fan, Gia-Wei Chern
Published: 2026-03-02
Categories: cond-mat.str-el, cs.LG, physics.comp-ph
arXiv: https://arxiv.org/abs/2603.02039v1

Scalable and symmetry-consistent force-field models are essential for extending quantum-accurate simulations to large spatiotemporal scales. While descriptor-based neural networks can incorporate lattice symmetries through carefully engineered features, we show that graph neural networks (GNNs) provide a conceptually simpler and more unified alternative in which discrete lattice translation and point-group symmetries are enforced directly through local message passing and weight sharing. We develop a GNN-based force-field framework for the adiabatic dynamics of lattice Hamiltonians and demonstrate it for the semiclassical Holstein model. Trained on exact-diagonalization data, the GNN achieves high force accuracy, strict linear scaling with system size, and direct transferability to large lattices. Enabled by this scalability, we perform large-scale Langevin simulations of charge-density-wave ordering following thermal quenches, revealing dynamical scaling and anomalously slow sub--Allen--Cahn coarsening. These results establish GNNs as an elegant and efficient architecture for symmetry-aware, large-scale dynamical simulations of correlated lattice systems.

---

### LAD-Drive: Bridging Language and Trajectory with Action-Aware Diffusion Transformers

Authors: Fabian Schmidt, Karol Fedurko, Markus Enzweiler, Abhinav Valada
Published: 2026-03-02
Categories: cs.RO, cs.CV
arXiv: https://arxiv.org/abs/2603.02035v1

While multimodal large language models (MLLMs) provide advanced reasoning for autonomous driving, translating their discrete semantic knowledge into continuous trajectories remains a fundamental challenge. Existing methods often rely on unimodal planning heads that inherently limit their ability to represent multimodal driving behavior. Furthermore, most generative approaches frequently condition on one-hot encoded actions, discarding the nuanced navigational uncertainty critical for complex scenarios. To resolve these limitations, we introduce LAD-Drive, a generative framework that structurally disentangles high-level intention from low-level spatial planning. LAD-Drive employs an action decoder to infer a probabilistic meta-action distribution, establishing an explicit belief state that preserves the nuanced intent typically lost by one-hot encodings. This distribution, fused with the vehicle's kinematic state, conditions an action-aware diffusion decoder that utilizes a truncated denoising process to refine learned motion anchors into safe, kinematically feasible trajectories. Extensive evaluations on the LangAuto benchmark demonstrate that LAD-Drive achieves state-of-the-art results, outperforming competitive baselines by up to 59% in Driving Score while significantly reducing route deviations and collisions. We will publicly release the code and models on https://github.com/iis-esslingen/lad-drive.

---

### TCG CREST System Description for the DISPLACE-M Challenge

Authors: Nikhil Raghav, Md Sahidullah
Published: 2026-03-02
Categories: eess.AS, cs.LG
arXiv: https://arxiv.org/abs/2603.02030v4

This report presents the TCG CREST system description for Track 1 (Speaker Diarization) of the DISPLACE-M challenge, focusing on naturalistic medical conversations in noisy rural-healthcare scenarios. Our study evaluates the impact of various voice activity detection (VAD) methods and advanced clustering algorithms on overall speaker diarization (SD) performance. We compare and analyze two SD frameworks: a modular pipeline utilizing SpeechBrain with ECAPA-TDNN embeddings, and a state-of-the-art (SOTA) hybrid end-to-end neural diarization system, Diarizen, built on top of a pre-trained WavLM. With these frameworks, we explore diverse clustering techniques, including agglomerative hierarchical clustering (AHC), and multiple novel variants of spectral clustering, such as SC-adapt, SC-PNA, and SC-MK. Experimental results demonstrate that the Diarizen system provides an approximate $39\%$ relative improvement in the diarization error rate (DER) on the post-evaluation analysis of Phase~I compared to the SpeechBrain baseline. Our best-performing submitted system employing the Diarizen baseline with AHC employing a median filtering with a larger context window of $29$ achieved a DER of 10.37\% on the development and 9.21\% on the evaluation sets, respectively. Our team ranked fifth out of the 11 participating teams after the Phase~I evaluation.

---

### Rich Insights from Cheap Signals: Efficient Evaluations via Tensor Factorization

Authors: Felipe Maia Polo, Aida Nematzadeh, Virginia Aglietti, Adam Fisch, Isabela Albuquerque
Published: 2026-03-02
Categories: cs.AI, cs.LG, stat.ML
arXiv: https://arxiv.org/abs/2603.02029v2

Moving beyond evaluations that collapse performance across heterogeneous prompts toward fine-grained evaluation at the prompt level, or within relatively homogeneous subsets, is necessary to diagnose generative models' strengths and weaknesses. Such fine-grained evaluations, however, suffer from a data bottleneck: human gold-standard labels are too costly at this scale, while automated ratings are often misaligned with human judgment. To resolve this challenge, we propose a novel statistical model based on tensor factorization that merges cheap autorater data with a limited set of human gold-standard labels. Specifically, our approach uses autorater scores to pretrain latent representations of prompts and generative models, and then aligns those pretrained representations to human preferences using a small calibration set. This sample-efficient methodology is robust to autorater quality, more accurately predicts human preferences on a per-prompt basis than standard baselines, and provides tight confidence intervals for key statistical parameters of interest. We also showcase the practical utility of our method by constructing granular leaderboards based on prompt qualities and by estimating model performance solely from autorater scores, eliminating the need for additional human annotations.

---

### Latent attention on masked patches for flow reconstruction

Authors: Ben Eze, Luca Magri, Andrea Nóvoa
Published: 2026-03-02
Categories: cs.LG
arXiv: https://arxiv.org/abs/2603.02028v2

Vision transformers have shown outstanding performance in image generation, yet their adoption in fluid dynamics remains limited. We introduce the Latent Attention on Masked Patches (LAMP) model, an interpretable regression-based modified vision transformer designed for masked flow reconstruction. LAMP follows a three-fold strategy: (i) partition of each flow snapshot into patches, (ii) patch-wise dimensionality reduction via proper orthogonal decomposition, and (iii) reconstruction of the full field from a masked input using a single-layer transformer trained via closed-form linear regression. We test the method on two canonical 2D unsteady wakes: a laminar wake past a bluff body, and a chaotic wake past two cylinders. On the laminar case, LAMP accurately reconstructs the full flow field from a 90%-masked and noisy input, across signal-to-noise ratios between 10 and 30dB. Further, the learned attention matrix yields interpretable multi-fidelity optimal sensor-placement maps. LAMP's performance on the chaotic wake is limited, but outperforms other regression methods such as gappy POD. The modularity of the framework, however, naturally accommodates nonlinear compression and deep attention blocks, thereby providing an efficient baseline for nonlinear, high-dimensional masked flow reconstruction.

---

### Learning to Read Where to Look: Disease-Aware Vision-Language Pretraining for 3D CT

Authors: Simon Ging, Philipp Arnold, Sebastian Walter, Hani Alnahas, Hannah Bast, Elmar Kotter, Jiancheng Yang, Behzad Bozorgtabar, Thomas Brox
Published: 2026-03-02
Categories: cs.CV, cs.CL, cs.LG
arXiv: https://arxiv.org/abs/2603.02026v1

Recent 3D CT vision-language models align volumes with reports via contrastive pretraining, but typically rely on limited public data and provide only coarse global supervision. We train a 3D CT vision-language model on 98k report-volume pairs (50k patients) collected at a single hospital, combined with public datasets, using SigLIP-style contrastive pretraining together with prompt-based disease supervision in the shared vision-text embedding space. On CT-RATE, our model achieves state-of-the-art text-to-image retrieval (R@10 31.5 vs. 22.2) and competitive disease classification (AUC 83.8 vs. 83.8), with consistent results on Rad-ChestCT (AUC 77.0 vs. 77.3). We further observe that radiologists routinely reference specific images within their reports (e.g., ``series X, image Y''), linking textual descriptions to precise axial locations. We automatically mine 262k such snippet-slice pairs and introduce the task of intra-scan snippet localization -- predicting the axial depth referred to by a text snippet -- reducing mean absolute error to 36.3 mm at 12 mm feature resolution, compared with 67.0 mm for the best baseline. Adding this localization objective leaves retrieval and classification broadly unchanged within confidence bounds, yielding a single unified model for retrieval, classification, and intra-scan grounding.

---

### Revealing Combinatorial Reasoning of GNNs via Graph Concept Bottleneck Layer

Authors: Yue Niu, Zhaokai Sun, Jiayi Yang, Xiaofeng Cao, Rui Fan, Xin Sun, Hanli Wang, Wei Ye
Published: 2026-03-02
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2603.02025v1

Despite their success in various domains, the growing dependence on GNNs raises a critical concern about the nature of the combinatorial reasoning underlying their predictions, which is often hidden within their black-box architectures. Addressing this challenge requires understanding how GNNs translate topological patterns into logical rules. However, current works only uncover the hard logical rules over graph concepts, which cannot quantify the contribution of each concept to prediction. Moreover, they are post-hoc interpretable methods that generate explanations after model training and may not accurately reflect the true combinatorial reasoning of GNNs, since they approximate it with a surrogate. In this work, we develop a graph concept bottleneck layer that can be integrated into any GNN architectures to guide them to predict the selected discriminative global graph concepts. The predicted concept scores are further projected to class labels by a sparse linear layer. It enforces the combinatorial reasoning of GNNs' predictions to fit the soft logical rule over graph concepts and thus can quantify the contribution of each concept. To further improve the quality of the concept bottleneck, we treat concepts as "graph words" and graphs as "graph sentences", and leverage language models to learn graph concept embeddings. Extensive experiments on multiple datasets show that our method GCBMs achieve state-of-the-art performance both in classification and interpretability.

---

### MMR-Life: Piecing Together Real-life Scenes for Multimodal Multi-image Reasoning

Authors: Jiachun Li, Shaoping Huang, Zhuoran Jin, Chenlong Zhang, Pengfei Cao, Yubo Chen, Kang Liu, Jun Zhao
Published: 2026-03-02
Categories: cs.CL, cs.AI, cs.CV
arXiv: https://arxiv.org/abs/2603.02024v1

Recent progress in the reasoning capabilities of multimodal large language models (MLLMs) has empowered them to address more complex tasks such as scientific analysis and mathematical reasoning. Despite their promise, MLLMs' reasoning abilities across different scenarios in real life remain largely unexplored and lack standardized benchmarks for evaluation. To address this gap, we introduce MMR-Life, a comprehensive benchmark designed to evaluate the diverse multimodal multi-image reasoning capabilities of MLLMs across real-life scenarios. MMR-Life consists of 2,646 multiple-choice questions based on 19,108 images primarily sourced from real-world contexts, comprehensively covering seven reasoning types: abductive, analogical, causal, deductive, inductive, spatial, and temporal. Unlike existing reasoning benchmarks, MMR-Life does not rely on domain-specific expertise but instead requires models to integrate information across multiple images and apply diverse reasoning abilities. The evaluation of 37 advanced models highlights the substantial challenge posed by MMR-Life. Even top models like GPT-5 achieve only 58% accuracy and display considerable variance in performance across reasoning types. Moreover, we analyze the reasoning paradigms of existing MLLMs, exploring how factors such as thinking length, reasoning method, and reasoning type affect their performance. In summary, MMR-Life establishes a comprehensive foundation for evaluating, analyzing, and improving the next generation of multimodal reasoning systems.

---

### PonderLM-3: Adaptive Token-Wise Pondering with Differentiable Masking

Authors: He Li, Feichen Song, Boyi Zeng, Shixiang Song, Zhiqin John Xu, Ziwei He, Zhouhan Lin
Published: 2026-03-02
Categories: cs.CL
arXiv: https://arxiv.org/abs/2603.02023v2

Test-time scaling has shown that allocating more additional computation at inference can improve generation quality, motivating a natural follow-up question: where should this computation be spent? Building on this insight, we introduce PonderLM-3, a pretraining framework for token-wise adaptive pondering that learns to selectively allocate additional computation under purely self-supervised objectives, built on top of the PonderLM-2 backbone. This makes additional inference computation an allocatable per-token resource, so tokens receive more computation only when it is beneficial, rather than paying a uniform extra cost. To make this allocation learnable while maintaining train-inference consistency, PonderLM-3 injects a differentiable attention mask during pretraining and pairs it with a matching hard pruning rule at inference. PonderLM-3 defines a stronger Pareto frontier: compared with existing recursive or adaptive baselines, it achieves lower pretraining perplexity at equal inference FLOPs. On downstream benchmarks, PonderLM-3 attains comparable performance to fixed-step PonderLM-2 under the same maximum number of additional computation steps, while using fewer inference FLOPs in practice. Overall, PonderLM-3 provides an end-to-end differentiable and train-inference consistent framework for token-wise adaptive computation, enabling additional inference compute to be allocated where it is most useful rather than paid uniformly by every token.

---

### CodecFlow: Efficient Bandwidth Extension via Conditional Flow Matching in Neural Codec Latent Space

Authors: Bowen Zhang, Junchuan Zhao, Ian McLoughlin, Ye Wang, A S Madhukumar
Published: 2026-03-02
Categories: cs.SD, cs.AI
arXiv: https://arxiv.org/abs/2603.02022v2

Speech Bandwidth Extension improves clarity and intelligibility by restoring/inferring appropriate high-frequency content for low-bandwidth speech. Existing methods often rely on spectrogram or waveform modeling, which can incur higher computational cost and have limited high-frequency fidelity. Neural audio codecs offer compact latent representations that better preserve acoustic detail, yet accurately recovering high-resolution latent information remains challenging due to representation mismatch. We present CodecFlow, a neural codec-based BWE framework that performs efficient speech reconstruction in a compact latent space. CodecFlow employs a voicing-aware conditional flow converter on continuous codec embeddings and a structure-constrained residual vector quantizer to improve latent alignment stability. Optimized end-to-end, CodecFlow achieves strong spectral fidelity and enhanced perceptual quality on 8 kHz to 16 kHz and 44.1 kHz speech BWE tasks.

---

### Selection as Power: Constrained Reinforcement for Bounded Decision Authority

Authors: Jose Manuel de la Chica Rodriguez, Juan Manuel Vera Díaz
Published: 2026-03-02
Categories: cs.MA, cs.AI, cs.CE, cs.LG
arXiv: https://arxiv.org/abs/2603.02019v1

Selection as Power argued that upstream selection authority, rather than internal objective misalignment, constitutes a primary source of risk in high-stakes agentic systems. However, the original framework was static: governance constraints bounded selection power but did not adapt over time. In this work, we extend the framework to dynamic settings by introducing incentivized selection governance, where reinforcement updates are applied to scoring and reducer parameters under externally enforced sovereignty constraints. We formalize selection as a constrained reinforcement process in which parameter updates are projected onto governance-defined feasible sets, preventing concentration beyond prescribed bounds. Across multiple regulated financial scenarios, unconstrained reinforcement consistently collapses into deterministic dominance under repeated feedback, especially at higher learning rates. In contrast, incentivized governance enables adaptive improvement while maintaining bounded selection concentration. Projection-based constraints transform reinforcement from irreversible lock-in into controlled adaptation, with governance debt quantifying the tension between optimization pressure and authority bounds. These results demonstrate that learning dynamics can coexist with structural diversity when sovereignty constraints are enforced at every update step, offering a principled approach to integrating reinforcement into high-stakes agentic systems without surrendering bounded selection authority.

---

### CausalWrap: Model-Agnostic Causal Constraint Wrappers for Tabular Synthetic Data

Authors: Amir Asiaee, Zhuohui J. Liang, Chao Yan
Published: 2026-03-02
Categories: cs.LG
arXiv: https://arxiv.org/abs/2603.02015v1

Tabular synthetic data generators are typically trained to match observational distributions, which can yield high conventional utility (e.g., column correlations, predictive accuracy) yet poor preservation of structural relations relevant to causal analysis and out-of-distribution (OOD) reasoning. When the downstream use of synthetic data involves causal reasoning -- estimating treatment effects, evaluating policies, or testing mediation pathways -- merely matching the observational distribution is insufficient: structural fidelity and treatment-mechanism preservation become essential. We propose CausalWrap (CW), a model-agnostic wrapper that injects partial causal knowledge (PCK) -- trusted edges, forbidden edges, and qualitative/monotonic constraints -- into any pretrained base generator (GAN, VAE, or diffusion model), without requiring access to its internals. CW learns a lightweight, differentiable post-hoc correction map applied to samples from the base generator, optimized with causal penalty terms under an augmented-Lagrangian schedule. We provide theoretical results connecting penalty-based optimization to constraint satisfaction and relating approximate factorization to joint distributional control. We validate CW on simulated structural causal models (SCMs) with known ground-truth interventions, semi-synthetic causal benchmarks (IHDP and an ACIC-style suite), and a real-world ICU cohort (MIMIC-IV) with expert-elicited partial graphs. CW improves causal fidelity across diverse base generators -- e.g., reducing average treatment effect (ATE) error by up to 63% on ACIC and lifting ATE agreement from 0.00 to 0.38 on the intensive care unit (ICU) cohort -- while largely retaining conventional utility.

---

### MAP-Diff: Multi-Anchor Guided Diffusion for Progressive 3D Whole-Body Low-Dose PET Denoising

Authors: Peiyuan Jing, Chun-Wun Cheng, Liutao Yang, Zhenxuan Zhang, Thiago V. Lima, Klaus Strobel, Antoine Leimgruber, Angelica Aviles-Rivero, Guang Yang, Javier A. Montoya-Zegarra
Published: 2026-03-02
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2603.02012v1

Low-dose Positron Emission Tomography (PET) reduces radiation exposure but suffers from severe noise and quantitative degradation. Diffusion-based denoising models achieve strong final reconstructions, yet their reverse trajectories are typically unconstrained and not aligned with the progressive nature of PET dose formation. We propose MAP-Diff, a multi-anchor guided diffusion framework for progressive 3D whole-body PET denoising. MAP-Diff introduces clinically observed intermediate-dose scans as trajectory anchors and enforces timestep-dependent supervision to regularize the reverse process toward dose-aligned intermediate states. Anchor timesteps are calibrated via degradation matching between simulated diffusion corruption and real multi-dose PET pairs, and a timestep-weighted anchor loss stabilizes stage-wise learning. At inference, the model requires only ultra-low-dose input while enabling progressive, dose-consistent intermediate restoration. Experiments on internal (Siemens Biograph Vision Quadra) and cross-scanner (United Imaging uEXPLORER) datasets show consistent improvements over strong CNN-, Transformer-, GAN-, and diffusion-based baselines. On the internal dataset, MAP-Diff improves PSNR from 42.48 dB to 43.71 dB (+1.23 dB), increases SSIM to 0.986, and reduces NMAE from 0.115 to 0.103 (-0.012) compared to 3D DDPM. Performance gains generalize across scanners, achieving 34.42 dB PSNR and 0.141 NMAE on the external cohort, outperforming all competing methods.

---

### Noise-Calibrated Inference from Differentially Private Sufficient Statistics in Exponential Families

Authors: Amir Asiaee, Samhita Pal
Published: 2026-03-02
Categories: cs.LG, stat.ML
arXiv: https://arxiv.org/abs/2603.02010v1

Many differentially private (DP) data release systems either output DP synthetic data and leave analysts to perform inference as usual, which can lead to severe miscalibration, or output a DP point estimate without a principled way to do uncertainty quantification. This paper develops a clean and tractable middle ground for exponential families: release only DP sufficient statistics, then perform noise-calibrated likelihood-based inference and optional parametric synthetic data generation as post-processing. Our contributions are: (1) a general recipe for approximate-DP release of clipped sufficient statistics under the Gaussian mechanism; (2) asymptotic normality, explicit variance inflation, and valid Wald-style confidence intervals for the plug-in DP MLE; (3) a noise-aware likelihood correction that is first-order equivalent to the plug-in but supports bootstrap-based intervals; and (4) a matching minimax lower bound showing the privacy distortion rate is unavoidable. The resulting theory yields concrete design rules and a practical pipeline for releasing DP synthetic data with principled uncertainty quantification, validated on three exponential families and real census data.

---

### Temporal Representations for Exploration: Learning Complex Exploratory Behavior without Extrinsic Rewards

Authors: Faisal Mohamed, Catherine Ji, Benjamin Eysenbach, Glen Berseth
Published: 2026-03-02
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2603.02008v2

Effective exploration in reinforcement learning requires not only tracking where an agent has been, but also understanding how the agent perceives and represents the world. To learn powerful representations, an agent should actively explore states that contribute to its knowledge of the environment. Temporal representations can capture the information necessary to solve a wide range of potential tasks while avoiding the computational cost associated with full state reconstruction. In this paper, we propose an exploration method that leverages temporal contrastive representations to guide exploration, prioritizing states with unpredictable future outcomes. We demonstrate that such representations can enable the learning of complex exploratory x in locomotion, manipulation, and embodied-AI tasks, revealing capabilities and behaviors that traditionally require extrinsic rewards. Unlike approaches that rely on explicit distance learning or episodic memory mechanisms (e.g., quasimetric-based methods), our method builds directly on temporal similarities, yielding a simpler yet effective strategy for exploration.

---

### Mitigating topology biases in Graph Diffusion via Counterfactual Intervention

Authors: Wendi Wang, Jiaxi Yang, Yongkang Du, Lu Lin
Published: 2026-03-02
Categories: cs.LG, cs.AI, cs.SI
arXiv: https://arxiv.org/abs/2603.02005v1

Graph diffusion models have gained significant attention in graph generation tasks, but they often inherit and amplify topology biases from sensitive attributes (e.g. gender, age, region), leading to unfair synthetic graphs. Existing fair graph generation using diffusion models is limited to specific graph-based applications with complete labels or requires simultaneous updates for graph structure and node attributes, making them unsuitable for general usage. To relax these limitations by applying the debiasing method directly on graph topology, we propose Fair Graph Diffusion Model (FairGDiff), a counterfactual-based one-step solution that mitigates topology biases while balancing fairness and utility. In detail, we construct a causal model to capture the relationship between sensitive attributes, biased link formation, and the generated graph structure. By answering the counterfactual question "Would the graph structure change if the sensitive attribute were different?", we estimate an unbiased treatment and incorporate it into the diffusion process. FairGDiff integrates counterfactual learning into both forward diffusion and backward denoising, ensuring that the generated graphs are independent of sensitive attributes while preserving structural integrity. Extensive experiments on real-world datasets demonstrate that FairGDiff achieves a superior trade-off between fairness and utility, outperforming existing fair graph generation methods while maintaining scalability.

---

### MatRIS: Toward Reliable and Efficient Pretrained Machine Learning Interatomic Potentials

Authors: Yuanchang Zhou, Siyu Hu, Xiangyu Zhang, Hongyu Wang, Guangming Tan, Weile Jia
Published: 2026-03-02
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2603.02002v3

Foundation MLIPs demonstrate broad applicability across diverse material systems and have emerged as a powerful and transformative paradigm in chemical and computational materials science. Equivariant MLIPs achieve state-of-the-art accuracy in a wide range of benchmarks by incorporating equivariant inductive bias. However, the reliance on tensor products and high-degree representations makes them computationally costly. This raises a fundamental question: as quantum mechanical-based datasets continue to expand, can we develop a more compact model to thoroughly exploit high-dimensional atomic interactions? In this work, we present MatRIS (\textbf{Mat}erials \textbf{R}epresentation and \textbf{I}nteraction \textbf{S}imulation), an invariant MLIP that introduces attention-based modeling of three-body interactions. MatRIS leverages a novel separable attention mechanism with linear complexity $O(N)$, enabling both scalability and expressiveness. MatRIS delivers accuracy comparable to that of leading equivariant models on a wide range of popular benchmarks (Matbench-Discovery, MatPES, MDR phonon, Molecular dataset, etc). Taking Matbench-Discovery as an example, MatRIS achieves an F1 score of up to 0.847 and attains comparable accuracy at a lower training cost. The work indicates that our carefully designed invariant models can match or exceed the accuracy of equivariant models at a fraction of the cost, shedding light on the development of accurate and efficient MLIPs.

---

### Learning Vision-Based Omnidirectional Navigation: A Teacher-Student Approach Using Monocular Depth Estimation

Authors: Jan Finke, Wayne Paul Martis, Adrian Schmelter, Lars Erbach, Christian Jestel, Marvin Wiedemann
Published: 2026-03-02
Categories: cs.RO, cs.CV, cs.LG
arXiv: https://arxiv.org/abs/2603.01999v3

Reliable obstacle avoidance in industrial settings demands 3D scene understanding, but widely used 2D LiDAR sensors perceive only a single horizontal slice of the environment, missing critical obstacles above or below the scan plane. We present a teacher-student framework for vision-based mobile robot navigation that eliminates the need for LiDAR sensors. A teacher policy trained via Proximal Policy Optimization (PPO) in NVIDIA Isaac Lab leverages privileged 2D LiDAR observations that account for the full robot footprint to learn robust navigation. The learned behavior is distilled into a student policy that relies solely on monocular depth maps predicted by a fine-tuned Depth Anything V2 model from four RGB cameras. The complete inference pipeline, comprising monocular depth estimation (MDE), policy execution, and motor control, runs entirely onboard an NVIDIA Jetson Orin AGX mounted on a DJI RoboMaster platform, requiring no external computation for inference. In simulation, the student achieves success rates of 82-96.5%, consistently outperforming the standard 2D LiDAR teacher (50-89%). In real-world experiments, the MDE-based student outperforms the 2D LiDAR teacher when navigating around obstacles with complex 3D geometries, such as overhanging structures and low-profile objects, that fall outside the single scan plane of a 2D LiDAR.

---

### Event-Only Drone Trajectory Forecasting with RPM-Modulated Kalman Filtering

Authors: Hari Prasanth S. M., Pejman Habibiroudkenar, Eerik Alamikkotervo, Dimitrios Bouzoulas, Risto Ojala
Published: 2026-03-02
Categories: cs.CV, cs.RO, eess.IV
arXiv: https://arxiv.org/abs/2603.01997v1

Event cameras provide high-temporal-resolution visual sensing that is well suited for observing fast-moving aerial objects; however, their use for drone trajectory prediction remains limited. This work introduces an event-only drone forecasting method that exploits propeller-induced motion cues. Propeller rotational speed are extracted directly from raw event data and fused within an RPM-aware Kalman filtering framework. Evaluations on the FRED dataset show that the proposed method outperforms learning-based approaches and vanilla kalman filter in terms of average distance error and final distance error at 0.4s and 0.8s forecasting horizons. The results demonstrate robust and accurate short- and medium-horizon trajectory forecasting without reliance on RGB imagery or training data.

---

### Cultivating Forensic Reasoning for Generalizable Multimodal Manipulation Detection

Authors: Yuchen Zhang, Yaxiong Wang, Kecheng Han, Yujiao Wu, Lianwei Wu, Li Zhu, Zhedong Zheng
Published: 2026-03-02
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.01993v2

Recent advances in generative AI have significantly enhanced the realism of multimodal media manipulation, thereby posing substantial challenges to manipulation detection. Existing manipulation detection and grounding approaches predominantly focus on manipulation type classification under result-oriented supervision, which not only lacks interpretability but also tends to overfit superficial artifacts. In this paper, we argue that generalizable detection requires incorporating explicit forensic reasoning, rather than merely classifying a limited set of manipulation types, which fails to generalize to unseen manipulation patterns. To this end, we propose REFORM, a reasoning-driven framework that shifts learning from outcome fitting to process modeling. REFORM adopts a three-stage curriculum that first induces forensic rationales, then aligns reasoning with final judgments, and finally refines logical consistency via reinforcement learning. To support this paradigm, we introduce ROM, a large-scale dataset with rich reasoning annotations. Extensive experiments show that REFORM establishes new state-of-the-art performance with superior generalization, achieving 81.52% ACC on ROM, 76.65% ACC on DGM4, and 74.9 F1 on MMFakeBench.

---

### According to Me: Long-Term Personalized Referential Memory QA

Authors: Jingbiao Mei, Jinghong Chen, Guangyu Yang, Xinyu Hou, Margaret Li, Bill Byrne
Published: 2026-03-02
Categories: cs.AI, cs.CL, cs.CV
arXiv: https://arxiv.org/abs/2603.01990v1

Personalized AI assistants must recall and reason over long-term user memory, which naturally spans multiple modalities and sources such as images, videos, and emails. However, existing Long-term Memory benchmarks focus primarily on dialogue history, failing to capture realistic personalized references grounded in lived experience. We introduce ATM-Bench, the first benchmark for multimodal, multi-source personalized referential Memory QA. ATM-Bench contains approximately four years of privacy-preserving personal memory data and human-annotated question-answer pairs with ground-truth memory evidence, including queries that require resolving personal references, multi-evidence reasoning from multi-source and handling conflicting evidence. We propose Schema-Guided Memory (SGM) to structurally represent memory items originated from different sources. In experiments, we implement 5 state-of-the-art memory systems along with a standard RAG baseline and evaluate variants with different memory ingestion, retrieval, and answer generation techniques. We find poor performance (under 20\% accuracy) on the ATM-Bench-Hard set, and that SGM improves performance over Descriptive Memory commonly adopted in prior works. Code available at: https://github.com/JingbiaoMei/ATM-Bench

---

### Accurate, private, secure, federated U-statistics with higher degree

Authors: Quentin Sinh, Jan Ramon
Published: 2026-03-02
Categories: cs.CR, cs.LG
arXiv: https://arxiv.org/abs/2603.01986v1

We study the problem of computing a U-statistic with a kernel function f of degree k $\ge$ 2, i.e., the average of some function f over all k-tuples of instances, in a federated learning setting. Ustatistics of degree 2 include several useful statistics such as Kendall's $τ$ coefficient, the Area under the Receiver-Operator Curve and the Gini mean difference. Existing methods provide solutions only under the lower-utility local differential privacy model and/or scale poorly in the size of the domain discretization. In this work, we propose a protocol that securely computes U-statistics of degree k $\ge$ 2 under central differential privacy by leveraging Multi Party Computation (MPC). Our method substantially improves accuracy when compared to prior solutions. We provide a detailed theoretical analysis of its accuracy, communication and computational properties. We evaluate its performance empirically, obtaining favorable results, e.g., for Kendall's $τ$ coefficient, our approach reduces the Mean Squared Error by up to four orders of magnitude over existing baselines.

---

### Quantitative Convergence of Wasserstein Gradient Flows of Kernel Mean Discrepancies

Authors: Lénaïc Chizat, Maria Colombo, Roberto Colombo, Xavier Fernández-Real
Published: 2026-03-02
Categories: math.AP, cs.LG, math.OC
arXiv: https://arxiv.org/abs/2603.01977v1

We study the quantitative convergence of Wasserstein gradient flows of Kernel Mean Discrepancy (KMD) (also known as Maximum Mean Discrepancy (MMD)) functionals. Our setting covers in particular the training dynamics of shallow neural networks in the infinite-width and continuous time limit, as well as interacting particle systems with pairwise Riesz kernel interaction in the mean-field and overdamped limit. Our main analysis concerns the model case of KMD functionals given by the squared Sobolev distance $ \mathscr{E}^ν_{s}(μ)= \frac{1}{2}\lVert μ-ν\rVert_{\dot H^{-s}}^{2}$ for any $s\geq 1 $ and $ν$ a fixed probability measure on the $d$-dimensional torus. First, inspired by Yudovich theory for the $2d$-Euler equation, we establish existence and uniqueness in natural weak regularity classes. Next, we show that for $s=1$ the flow converges globally at an exponential rate under minimal assumptions, while for $s>1$ we prove local convergence at polynomial rates that depend explicitly on $s$ and on the Sobolev regularity of $μ$ and $ν$. These rates hold both at the energy level and in higher regularity classes and are tight for $ν$ uniform. We then consider the gradient flow of the population loss for shallow neural networks with ReLU activation, which can be cast as a Wasserstein--Fisher--Rao gradient flow on the space of nonnegative measures on the sphere $\mathbb{S}^d$. Exploiting a correspondence with the Sobolev energy case with $s=(d+3)/2$, we derive an explicit polynomial local convergence rate for this dynamics. Except for the special case $s=1$, even non-quantitative convergence was previously open in all these settings. We also include numerical experiments in dimension $d=1$ using both PDE and particle methods which illustrate our analysis.

---

### Robust White Blood Cell Classification with Stain-Normalized Decoupled Learning and Ensembling

Authors: Luu Le, Hoang-Loc Cao, Ha-Hieu Pham, Thanh-Huy Nguyen, Ulas Bagci
Published: 2026-03-02
Categories: cs.CV
arXiv: https://arxiv.org/abs/2603.01976v1

White blood cell (WBC) classification is fundamental for hematology applications such as infection assessment, leukemia screening, and treatment monitoring. However, real-world WBC datasets present substantial appearance variations caused by staining and scanning conditions, as well as severe class imbalance in which common cell types dominate while rare but clinically important categories are underrepresented. To address these challenges, we propose a stain-normalized, decoupled training framework that first learns transferable representations using instance-balanced sampling, and then rebalances the classifier with class-aware sampling and a hybrid loss combining effective-number weighting and focal modulation. In inference stage, we further enhance robustness by ensembling various trained backbones with test-time augmentation. Our approach achieved the top rank on the leaderboard of the WBCBench 2026: Robust White Blood Cell Classification Challenge at ISBI 2026.

---

### Density-Matrix Spectral Embeddings for Categorical Data: Operator Structure and Stability

Authors: Raquel Bosch-Romeu, Antonio Falcó, osé-Antonio Rodríguez-Gallego
Published: 2026-03-02
Categories: stat.ML, math.NA
arXiv: https://arxiv.org/abs/2603.01975v1

We introduce a supervised dimensionality reduction methodology for categorical (and discretized mixed-type) data based on a density-matrix construction induced by class-conditional frequencies. Given a labeled dataset encoded in a one-hot survey space, we assemble a frequency matrix whose columns aggregate feature occurrences within each class, and define a normalized Gram-type operator that satisfies the axioms of a density matrix. The resulting representation admits an intrinsic rank bound controlled by the number of classes, enabling low-dimensional spectral embeddings via dominant eigenmodes. Classification is performed in the reduced space through class-conditional kernel density estimation and a maximum-likelihood decision rule. We establish structural invariances, provide complexity estimates, and validate the approach on synthetic benchmarks probing high cardinality, sparsity, noise, and class imbalance.

---

### CharacterFlywheel: Scaling Iterative Improvement of Engaging and Steerable LLMs in Production

Authors: Yixin Nie, Lin Guan, Zhongyao Ma, Anchit Gupta, Yipin Zhou, Xiao Li, Zhengping Zhou, Raymond Zeng, Gelin Zhou, Shigan Chu, Ajay Thampi, Wancen Mu et al.
Published: 2026-03-02
Categories: cs.CL, cs.AI, cs.SI
arXiv: https://arxiv.org/abs/2603.01973v1

This report presents CharacterFlywheel, an iterative flywheel process for improving large language models (LLMs) in production social chat applications across Instagram, WhatsApp, and Messenger. Starting from LLaMA 3.1, we refined models across 15 generations using data from both internal and external real-user traffic. Through continuous deployments from July 2024 to April 2025, we conducted controlled 7-day A/B tests showing consistent engagement improvements: 7 of 8 newly deployed models demonstrated positive lift over the baseline, with the strongest performers achieving up to 8.8% improvement in engagement breadth and 19.4% in engagement depth. We also observed substantial gains in steerability, with instruction following increasing from 59.2% to 84.8% and instruction violations decreasing from 26.6% to 5.8%. We detail the CharacterFlywheel process which integrates data curation, reward modeling to estimate and interpolate the landscape of engagement metrics, supervised fine-tuning (SFT), reinforcement learning (RL), and both offline and online evaluation to ensure reliable progress at each optimization step. We also discuss our methods for overfitting prevention and navigating production dynamics at scale. These contributions advance the scientific rigor and understanding of LLMs in social applications serving millions of users.

---

### LOCUS: A Distribution-Free Loss-Quantile Score for Risk-Aware Predictions

Authors: Matheus Barreto, Mário de Castro, Thiago R. Ramos, Denis Valle, Rafael Izbicki
Published: 2026-03-02
Categories: stat.ML, cs.LG
arXiv: https://arxiv.org/abs/2603.01971v1

Modern machine learning models can be accurate on average yet still make mistakes that dominate deployment cost. We introduce Locus, a distribution-free wrapper that produces a per-input loss-scale reliability score for a fixed prediction function. Rather than quantifying uncertainty about the label, Locus models the realized loss of the prediction function using any engine that outputs a predictive distribution for the loss given an input. A simple split-calibration step turns this function into a distribution-free interpretable score that is comparable across inputs and can be read as an upper loss level. The score is useful on its own for ranking, and it can optionally be thresholded to obtain a transparent flagging rule with distribution-free control of large-loss events. Experiments across 13 regression benchmarks show that Locus yields effective risk ranking and reduces large-loss frequency compared to standard heuristics.

---