# Research Papers — 2026-02 · bundle page 52 of 210

> Full-text markdown of papers #2551–2600 of 10493 published in 2026-02, newest first. AI-readable mirrors of the full 10493-paper 2026-02 archive (50 papers per file).

Canonical page: https://aipod.dev/papers/archive/2026-02/page/52
[← Previous bundle](https://aipod.dev/papers/archive/2026-02/page/51.md)
[Next bundle →](https://aipod.dev/papers/archive/2026-02/page/53.md)

### LoMime: Query-Efficient Membership Inference using Model Extraction in Label-Only Settings

Authors: Abdullah Caglar Oksuz, Anisa Halimi, Erman Ayday
Published: 2026-02-21
Categories: cs.LG, cs.CR
arXiv: https://arxiv.org/abs/2602.18934v2

Membership inference attacks (MIAs) threaten the privacy of machine learning models by revealing whether a specific data point was used during training. Existing MIAs often rely on impractical assumptions, such as access to public datasets, shadow models, confidence scores, or knowledge of the training data distribution, making them vulnerable to defenses like confidence masking and adversarial regularization. Label-only MIAs, even under strict constraints, suffer from high query requirements per sample. We propose a cost-effective label-only MIA framework based on transferability and model extraction. By querying the target model $M$ using active sampling, perturbation-based selection, and synthetic data, we extract a functionally similar surrogate model $S$ on which membership inference is performed. This shifts the query overhead to a one-time extraction phase, eliminating repeated queries to $M$. Our method matches the performance of state-of-the-art label-only MIAs while significantly reducing query costs and operating under strict black-box constraints. On benchmark tabular datasets, we show that a query budget equivalent to testing the membership of approximately $1%$ of the training samples is sufficient to extract $S$ and achieve membership inference accuracy within $\pm 1%$ of that obtained when attacking $M$ directly. We also evaluate the effectiveness of standard defenses, including DP-SGD and regularization, proposed for label-only MIAs against our attack. Finally, we present preliminary results extending our framework to deep neural networks trained on image datasets, demonstrating promising transferability and membership inference performance under label-only access while highlighting directions for further optimization.

---

### Give Users the Wheel: Towards Promptable Recommendation Paradigm

Authors: Fuyuan Lyu, Chenglin Luo, Qiyuan Zhang, Yupeng Hou, Haolun Wu, Xing Tang, Xue Liu, Jin L. C. Guo, Xiuqiang He
Published: 2026-02-21
Categories: cs.IR, cs.AI
arXiv: https://arxiv.org/abs/2602.18929v2

Conventional sequential recommendation models have achieved remarkable success in mining implicit behavioral patterns. However, these architectures remain structurally blind to explicit user intent: they struggle to adapt when a user's immediate goal (e.g., expressed via a natural language prompt) deviates from their historical habits. While Large Language Models (LLMs) offer the semantic reasoning to interpret such intent, existing integration paradigms force a dilemma: LLM-as-a-recommender paradigm sacrifices the efficiency and collaborative precision of ID-based retrieval, while Reranking methods are inherently bottlenecked by the recall capabilities of the underlying model. In this paper, we propose Decoupled Promptable Sequential Recommendation (DPR), a model-agnostic framework that empowers conventional sequential backbones to natively support Promptable Recommendation, the ability to dynamically steer the retrieval process using natural language without abandoning collaborative signals. DPR modulates the latent user representation directly within the retrieval space. To achieve this, we introduce a Fusion module to align the collaborative and semantic signals, a Mixture-of-Experts (MoE) architecture that disentangles the conflicting gradients from positive and negative steering, and a three-stage training strategy that progressively aligns the semantic space of prompts with the collaborative space. Extensive experiments on real-world datasets demonstrate that DPR significantly outperforms state-of-the-art baselines in prompt-guided tasks while maintaining competitive performance in standard sequential recommendation scenarios.

---

### Why Agent Caching Fails and How to Fix It: Structured Intent Canonicalization with Few-Shot Learning

Authors: Abhinaba Basu
Published: 2026-02-21
Categories: cs.CL, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2602.18922v2

Personal AI agents incur substantial cost via repeated LLM calls. We show existing caching methods fail: GPTCache achieves 37.9% accuracy on real benchmarks; APC achieves 0-12%. The root cause is optimizing for the wrong property -- cache effectiveness requires key consistency and precision, not classification accuracy. We observe cache-key evaluation reduces to clustering evaluation and apply V-measure decomposition to separate these on n=8,682 points across MASSIVE, BANKING77, CLINC150, and NyayaBench v2, our new 8,514-entry multilingual agentic dataset (528 intents, 20 W5H2 classes, 63 languages). We introduce W5H2, a structured intent decomposition framework. Using SetFit with 8 examples per class, W5H2 achieves 91.1%+/-1.7% on MASSIVE in ~2ms -- vs 37.9% for GPTCache and 68.8% for a 20B-parameter LLM at 3,447ms. On NyayaBench v2 (20 classes), SetFit achieves 55.3%, with cross-lingual transfer across 30 languages. Our five-tier cascade handles 85% of interactions locally, projecting 97.5% cost reduction. We provide risk-controlled selective prediction guarantees via RCPS with nine bound families.

---

### DeepInnovator: Triggering the Innovative Capabilities of LLMs

Authors: Tianyu Fan, Fengji Zhang, Yuxiang Zheng, Bei Chen, Xinyao Niu, Chengen Huang, Junyang Lin, Chao Huang
Published: 2026-02-21
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2602.18920v1

The application of Large Language Models (LLMs) in accelerating scientific discovery has garnered increasing attention, with a key focus on constructing research agents endowed with innovative capability, i.e., the ability to autonomously generate novel and significant research ideas. Existing approaches predominantly rely on sophisticated prompt engineering and lack a systematic training paradigm. To address this, we propose DeepInnovator, a training framework designed to trigger the innovative capability of LLMs. Our approach comprises two core components. (1) ``Standing on the shoulders of giants''. We construct an automated data extraction pipeline to extract and organize structured research knowledge from a vast corpus of unlabeled scientific literature. (2) ``Conjectures and refutations''. We introduce a ``Next Idea Prediction'' training paradigm, which models the generation of research ideas as an iterative process of continuously predicting, evaluating, and refining plausible and novel next idea. Both automatic and expert evaluations demonstrate that our DeepInnovator-14B significantly outperforms untrained baselines, achieving win rates of 80.53\%-93.81\%, and attains performance comparable to that of current leading LLMs. This work provides a scalable training pathway toward building research agents with genuine, originative innovative capability, and will open-source the dataset to foster community advancement. Source code and data are available at: https://github.com/HKUDS/DeepInnovator.

---

### Early Evidence of Vibe-Proving with Consumer LLMs: A Case Study on Spectral Region Characterization with ChatGPT-5.2 (Thinking)

Authors: Brecht Verbeken, Brando Vagenende, Marie-Anne Guerry, Andres Algaba, Vincent Ginis
Published: 2026-02-21
Categories: cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2602.18918v1

Large Language Models (LLMs) are increasingly used as scientific copilots, but evidence on their role in research-level mathematics remains limited, especially for workflows accessible to individual researchers. We present early evidence for vibe-proving with a consumer subscription LLM through an auditable case study that resolves Conjecture 20 of Ran and Teng (2024) on the exact nonreal spectral region of a 4-cycle row-stochastic nonnegative matrix family. We analyze seven shareable ChatGPT-5.2 (Thinking) threads and four versioned proof drafts, documenting an iterative pipeline of generate, referee, and repair. The model is most useful for high-level proof search, while human experts remain essential for correctness-critical closure. The final theorem provides necessary and sufficient region conditions and explicit boundary attainment constructions. Beyond the mathematical result, we contribute a process-level characterization of where LLM assistance materially helps and where verification bottlenecks persist, with implications for evaluation of AI-assisted research workflows and for designing human-in-the-loop theorem proving systems.

---

### Adaptive Collaboration of Arena-Based Argumentative LLMs for Explainable and Contestable Legal Reasoning

Authors: Hoang-Loc Cao, Phuc Ho, Truong Thanh Hung Nguyen, Phuc Truong Loc Nguyen, Dinh Thien Loc Nguyen, Hung Cao
Published: 2026-02-21
Categories: cs.MA, cs.AI, cs.SC
arXiv: https://arxiv.org/abs/2602.18916v1

Legal reasoning requires not only high accuracy but also the ability to justify decisions through verifiable and contestable arguments. However, existing Large Language Model (LLM) approaches, such as Chain-of-Thought (CoT) and Retrieval-Augmented Generation (RAG), often produce unstructured explanations that lack a formal mechanism for verification or user intervention. To address this limitation, we propose Adaptive Collaboration of Argumentative LLMs (ACAL), a neuro-symbolic framework that integrates adaptive multi-agent collaboration with an Arena-based Quantitative Bipolar Argumentation Framework (A-QBAF). ACAL dynamically deploys expert agent teams to construct arguments, employs a clash resolution mechanism to adjudicate conflicting claims, and utilizes uncertainty-aware escalation for borderline cases. Crucially, our framework supports a Human-in-the-Loop (HITL) contestability workflow, enabling users to directly audit and modify the underlying reasoning graph to influence the final judgment. Empirical evaluations on the LegalBench benchmark demonstrate that ACAL outperforms strong baselines across Gemini-2.5-Flash-Lite and Gemini-2.5-Flash architectures, effectively balancing efficient predictive performance with structured transparency and contestability. Our implementation is available at: https://github.com/loc110504/ACAL.

---

### AAVGen: Precision Engineering of Adeno-associated Viral Capsids for Renal Selective Targeting

Authors: Mohammadreza Ghaffarzadeh-Esfahani, Yousof Gheisari
Published: 2026-02-21
Categories: q-bio.QM, cs.AI, cs.CL, cs.LG
arXiv: https://arxiv.org/abs/2602.18915v1

Adeno-associated viruses (AAVs) are promising vectors for gene therapy, but their native serotypes face limitations in tissue tropism, immune evasion, and production efficiency. Engineering capsids to overcome these hurdles is challenging due to the vast sequence space and the difficulty of simultaneously optimizing multiple functional properties. The complexity also adds when it comes to the kidney, which presents unique anatomical barriers and cellular targets that require precise and efficient vector engineering. Here, we present AAVGen, a generative artificial intelligence framework for de novo design of AAV capsids with enhanced multi-trait profiles. AAVGen integrates a protein language model (PLM) with supervised fine-tuning (SFT) and a reinforcement learning technique termed Group Sequence Policy Optimization (GSPO). The model is guided by a composite reward signal derived from three ESM-2-based regression predictors, each trained to predict a key property: production fitness, kidney tropism, and thermostability. Our results demonstrate that AAVGen produces a diverse library of novel VP1 protein sequences. In silico validations revealed that the majority of the generated variants have superior performance across all three employed indices, indicating successful multi-objective optimization. Furthermore, structural analysis via AlphaFold3 confirms that the generated sequences preserve the canonical capsid folding despite sequence diversification. AAVGen establishes a foundation for data-driven viral vector engineering, accelerating the development of next-generation AAV vectors with tailored functional characteristics.

---

### From Human-Level AI Tales to AI Leveling Human Scales

Authors: Peter Romero, Fernando Martínez-Plumed, Zachary R. Tidler, Matthieu Téhénan, Sipeng Chen, Álvaro David Gómez Antón, Luning Sun, Manuel Cebrian, Lexin Zhou, Yael Moros Daval, Daniel Romero-Alvarado, Félix Martí Pérez et al.
Published: 2026-02-21
Categories: cs.LG
arXiv: https://arxiv.org/abs/2602.18911v2

Comparing AI models to "human level" is often misleading when benchmark scores are incommensurate or human baselines are drawn from a narrow population. To address this, we propose a framework that calibrates items against the 'world population' and report performance on a common, human-anchored scale. Concretely, we build on a set of multi-level scales for different capabilities where each level should represent a probability of success of the whole world population on a logarithmic scale with a base $B$. We calibrate each scale for each capability (reasoning, comprehension, knowledge, volume, etc.) by compiling publicly released human test data spanning education and reasoning benchmarks (PISA, TIMSS, ICAR, UKBioBank, and ReliabilityBench). The base $B$ is estimated by extrapolating between samples with two demographic profiles using LLMs, with the hypothesis that they condense rich information about human populations. We evaluate the quality of different mappings using group slicing and post-stratification. The new techniques allow for the recalibration and standardization of scales relative to the whole-world population.

---

### SLDP: Semi-Local Differential Privacy for Density-Adaptive Analytics

Authors: Alexey Kroshnin, Alexandra Suvorikova
Published: 2026-02-21
Categories: cs.LG
arXiv: https://arxiv.org/abs/2602.18910v1

Density-adaptive domain discretization is essential for high-utility privacy-preserving analytics but remains challenging under Local Differential Privacy (LDP) due to the privacy-budget costs associated with iterative refinement. We propose a novel framework, Semi-Local Differential Privacy (SLDP), that assigns a privacy region to each user based on local density and defines adjacency by the potential movement of a point within its privacy region. We present an interactive $(\varepsilon, δ)$-SLDP protocol, orchestrated by an honest-but-curious server over a public channel, to estimate these regions privately. Crucially, our framework decouples the privacy cost from the number of refinement iterations, allowing for high-resolution grids without additional privacy budget cost. We experimentally demonstrate the framework's effectiveness on estimation tasks across synthetic and real-world datasets.

---

### DeepInterestGR: Mining Deep Multi-Interest Using Multi-Modal LLMs for Generative Recommendation

Authors: Yangchen Zeng, Zhenyu Yu, Zhiyuan Hu, Wenxin Zhang, Jinze Wang, Rongfeng Guo
Published: 2026-02-21
Categories: cs.LG, cs.CV, cs.CY
arXiv: https://arxiv.org/abs/2602.18907v2

We introduce DeepInterestGR, a novel framework that integrates deep interest mining into the generative recommendation pipeline. This addresses the "Shallow Interest" problem - existing generative methods rely on surface-level textual features and fail to capture latent user motivations, limiting personalization depth and recommendation interpretability. Our approach leverages Multi-LLM Interest Mining (MLIM) via structured reasoning prompting, Reward-Labeled Deep Interest (RLDI) for quality control, and Interest-Enhanced Item Discretization (IEID) via RQ-VAE, combined with a two-stage SFT-GRPO training pipeline guided by an Interest-Aware Reward. We validate DeepInterestGR on three Amazon Review benchmarks (Beauty, Sports, Instruments), comparing against 14 state-of-the-art baselines including SASRec, BERT4Rec, TIGER, LC-Rec, and S-DPO. Our method achieves 5.8%-8.3% relative improvements on HR@10 and 7.7%-9.9% on NDCG@10 over the strongest baseline, with cross-domain generalization gains of +24.8%. These results provide evidence that incorporating deep semantic interests can effectively improve SID-based generative recommendation.

---

### Marginalized Bundle Adjustment: Multi-View Camera Pose from Monocular Depth Estimates

Authors: Shengjie Zhu, Ahmed Abdelkader, Mark J. Matthews, Xiaoming Liu, Wen-Sheng Chu
Published: 2026-02-21
Categories: cs.CV
arXiv: https://arxiv.org/abs/2602.18906v1

Structure-from-Motion (SfM) is a fundamental 3D vision task for recovering camera parameters and scene geometry from multi-view images. While recent deep learning advances enable accurate Monocular Depth Estimation (MDE) from single images without depending on camera motion, integrating MDE into SfM remains a challenge. Unlike conventional triangulated sparse point clouds, MDE produces dense depth maps with significantly higher error variance. Inspired by modern RANSAC estimators, we propose Marginalized Bundle Adjustment (MBA) to mitigate MDE error variance leveraging its density. With MBA, we show that MDE depth maps are sufficiently accurate to yield SoTA or competitive results in SfM and camera relocalization tasks. Through extensive evaluations, we demonstrate consistently robust performance across varying scales, ranging from few-frame setups to large multi-view systems with thousands of images. Our method highlights the significant potential of MDE in multi-view 3D vision.

---

### TRUE: A Trustworthy Unified Explanation Framework for Large Language Model Reasoning

Authors: Yujiao Yang
Published: 2026-02-21
Categories: cs.LG, cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2602.18905v2

Large language models (LLMs) have demonstrated strong capabilities in complex reasoning tasks, yet their decision-making processes remain difficult to interpret. Existing explanation methods often lack trustworthy structural insight and are limited to single-instance analysis, failing to reveal reasoning stability and systematic failure mechanisms. To address these limitations, we propose the Trustworthy Unified Explanation Framework (TRUE), which integrates executable reasoning verification, feasible-region directed acyclic graph (DAG) modeling, and causal failure mode analysis. At the instance level, we redefine reasoning traces as executable process specifications and introduce blind execution verification to assess operational validity. At the local structural level, we construct feasible-region DAGs via structure-consistent perturbations, enabling explicit characterization of reasoning stability and the executable region in the local input space. At the class level, we introduce a causal failure mode analysis method that identifies recurring structural failure patterns and quantifies their causal influence using Shapley values. Extensive experiments across multiple reasoning benchmarks demonstrate that the proposed framework provides multi-level, verifiable explanations, including executable reasoning structures for individual instances, feasible-region representations for neighboring inputs, and interpretable failure modes with quantified importance at the class level. These results establish a unified and principled paradigm for improving the interpretability and reliability of LLM reasoning systems.

---

### PCA-VAE: Differentiable Subspace Quantization without Codebook Collapse

Authors: Hao Lu, Onur C. Koyun, Yongxin Guo, Zhengjie Zhu, Abbas Alili, Metin Nafi Gurcan
Published: 2026-02-21
Categories: cs.LG, cs.CV
arXiv: https://arxiv.org/abs/2602.18904v1

Vector-quantized autoencoders deliver high-fidelity latents but suffer inherent flaws: the quantizer is non-differentiable, requires straight-through hacks, and is prone to collapse. We address these issues at the root by replacing VQ with a simple, principled, and fully differentiable alternative: an online PCA bottleneck trained via Oja's rule. The resulting model, PCA-VAE, learns an orthogonal, variance-ordered latent basis without codebooks, commitment losses, or lookup noise. Despite its simplicity, PCA-VAE exceeds VQ-GAN and SimVQ in reconstruction quality on CelebAHQ while using 10-100x fewer latent bits. It also produces naturally interpretable dimensions (e.g., pose, lighting, gender cues) without adversarial regularization or disentanglement objectives. These results suggest that PCA is a viable replacement for VQ: mathematically grounded, stable, bit-efficient, and semantically structured, offering a new direction for generative models beyond vector quantization.

---

### SCHEMA for Gemini 3 Pro Image: A Structured Methodology for Controlled AI Image Generation on Google's Native Multimodal Model

Authors: Luca Cazzaniga
Published: 2026-02-21
Categories: cs.CV, cs.HC
arXiv: https://arxiv.org/abs/2602.18903v1

This paper presents SCHEMA (Structured Components for Harmonized Engineered Modular Architecture), a structured prompt engineering methodology specifically developed for Google Gemini 3 Pro Image. Unlike generic prompt guidelines or model-agnostic tips, SCHEMA is an engineered framework built on systematic professional practice encompassing 850 verified API predictions within an estimated corpus of approximately 4,800 generated images, spanning six professional domains: real estate photography, commercial product photography, editorial content, storyboards, commercial campaigns, and information design. The methodology introduces a three-tier progressive system (BASE, MEDIO, AVANZATO) that scales practitioner control from exploratory (approximately 5%) to directive (approximately 95%), a modular label architecture with 7 core and 5 optional structured components, a decision tree with explicit routing rules to alternative tools, and systematically documented model limitations with corresponding workarounds. Key findings include an observed 91% Mandatory compliance rate and 94% Prohibitions compliance rate across 621 structured prompts, a comparative batch consistency test demonstrating substantially higher inter-generation coherence for structured prompts, independent practitioner validation (n=40), and a dedicated Information Design validation demonstrating >95% first-generation compliance for spatial and typographical control across approximately 300 publicly verifiable infographics. Previously published on Zenodo (doi:10.5281/zenodo.18721380).

---

### PrivacyBench: Privacy Isn't Free in Hybrid Privacy-Preserving Vision Systems

Authors: Nnaemeka Obiefuna, Samuel Oyeneye, Similoluwa Odunaiya, Iremide Oyelaja, Steven Kolawole
Published: 2026-02-21
Categories: cs.CR, cs.CV
arXiv: https://arxiv.org/abs/2602.18900v2

Privacy preserving machine learning deployments in sensitive deep learning applications; from medical imaging to autonomous systems; increasingly require combining multiple techniques. Yet, practitioners lack systematic guidance to assess the synergistic and non-additive interactions of these hybrid configurations, relying instead on isolated technique analysis that misses critical system level interactions. We introduce PrivacyBench, a benchmarking framework that reveals striking failures in privacy technique combinations with severe deployment implications. Through systematic evaluation across ResNet18 and ViT models on medical datasets, we uncover that FL + DP combinations exhibit severe convergence failure, with accuracy dropping from 98% to 13% while compute costs and energy consumption substantially increase. In contrast, FL + SMPC maintains near-baseline performance with modest overhead. Our framework provides the first systematic platform for evaluating privacy-utility-cost trade-offs through automated YAML configuration, resource monitoring, and reproducible experimental protocols. PrivacyBench enables practitioners to identify problematic technique interactions before deployment, moving privacy-preserving computer vision from ad-hoc evaluation toward principled systems design. These findings demonstrate that privacy techniques cannot be composed arbitrarily and provide critical guidance for robust deployment in resource-constrained environments.

---

### (b)=(d)-(t)+(p): Self-supervised Speech Models Discover Phonological Vector Arithmetic

Authors: Kwanghee Choi, Eunjung Yeo, Cheol Jun Cho, David Harwath, David R. Mortensen
Published: 2026-02-21
Categories: eess.AS, cs.CL, cs.LG, cs.SD
arXiv: https://arxiv.org/abs/2602.18899v3

Self-supervised speech models (S3Ms) are known to encode rich phonetic information, yet how this information is structured remains underexplored. We conduct a comprehensive study across 96 languages to analyze the underlying structure of S3M representations, with particular attention to phonological vectors. We first show that there exist linear directions within the model's representation space that correspond to phonological features. We further demonstrate that the scale of these phonological vectors correlate to the degree of acoustic realization of their corresponding phonological features in a continuous manner. For example, the difference between (d) and (t) yields a voicing vector: adding this vector to (p) produces (b), while scaling it results in a continuum of voicing. Together, these findings indicate that S3Ms encode speech using phonologically interpretable and compositional vectors, demonstrating phonological vector arithmetic. All code and interactive demos are available at https://github.com/juice500ml/phonetic-arithmetic .

---

### HEHRGNN: A Unified Embedding Model for Knowledge Graphs with Hyperedges and Hyper-Relational Edges

Authors: Rajesh Rajagopalamenon, Unnikrishnan Cheramangalath
Published: 2026-02-21
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2602.18897v1

Knowledge Graph(KG) has gained traction as a machine-readable organization of real-world knowledge for analytics using artificial intelligence systems. Graph Neural Network(GNN), is proven to be an effective KG embedding technique that enables various downstream tasks like link prediction, node classification, and graph classification. The focus of research in both KG embedding and GNNs has been mostly oriented towards simple graphs with binary relations. However, real-world knowledge bases have a significant share of complex and n-ary facts that cannot be represented by binary edges. More specifically, real-world knowledge bases are often a mix of two types of n-ary facts - (i) that require hyperedges and (ii) that require hyper-relational edges. Though there are research efforts catering to these n-ary fact types, they are pursued independently for each type. We propose $H$yper$E$dge $H$yper-$R$elational edge $GNN$(HEHRGNN), a unified embedding model for n-ary relational KGs with both hyperedges and hyper-relational edges. The two main components of the model are i)HEHR unified fact representation format, and ii)HEHRGNN encoder, a GNN-based encoder with a novel message propagation model capable of capturing complex graph structures comprising both hyperedges and hyper-relational edges. The experimental results of HEHRGNN on link prediction tasks show its effectiveness as a unified embedding model, with inductive prediction capability, for link prediction across real-world datasets having different types of n-ary facts. The model also shows improved link prediction performance over baseline models for hyperedge and hyper-relational datasets.

---

### Beyond Stationarity: Rethinking Codebook Collapse in Vector Quantization

Authors: Hao Lu, Onur C. Koyun, Yongxin Guo, Zhengjie Zhu, Abbas Alili, Metin Nafi Gurcan
Published: 2026-02-21
Categories: cs.CV
arXiv: https://arxiv.org/abs/2602.18896v1

Vector Quantization (VQ) underpins many modern generative frameworks such as VQ-VAE, VQ-GAN, and latent diffusion models. Yet, it suffers from the persistent problem of codebook collapse, where a large fraction of code vectors remains unused during training. This work provides a new theoretical explanation by identifying the nonstationary nature of encoder updates as the fundamental cause of this phenomenon. We show that as the encoder drifts, unselected code vectors fail to receive updates and gradually become inactive. To address this, we propose two new methods: Non-Stationary Vector Quantization (NSVQ), which propagates encoder drift to non-selected codes through a kernel-based rule, and Transformer-based Vector Quantization (TransVQ), which employs a lightweight mapping to adaptively transform the entire codebook while preserving convergence to the k-means solution. Experiments on the CelebA-HQ dataset demonstrate that both methods achieve near-complete codebook utilization and superior reconstruction quality compared to baseline VQ variants, providing a principled and scalable foundation for future VQ-based generative models. The code is available at: https://github.com/CAIR- LAB- WFUSM/NSVQ-TransVQ.git

---

### Could Large Language Models work as Post-hoc Explainability Tools in Credit Risk Models?

Authors: Wenxi Geng, Dingyuan Liu, Liya Li, Yiqing Wang
Published: 2026-02-21
Categories: q-fin.RM, cs.LG
arXiv: https://arxiv.org/abs/2602.18895v2

Large language models (LLMs) have shown promise in translating model-based explanations into human-readable narratives. This study evaluates whether LLMs can serve as post-hoc explainability interfaces for credit risk models, focusing on their ability to preserve feature-importance rankings and generate autonomous explanations. Using a LendingClub dataset, we compare LLM outputs with SHAP and coefficient-based attributions on three major LLMs, including GPT-4-turbo, Claude-Sonnet-4.5, and Gemini-2.5-Flash. Results indicate that LLMs reliably reproduce reference rankings under controlled prompts but show limited alignment when generating explanations autonomously. These findings suggest that LLMs are best deployed as narrative interfaces rather than substitutes for formal attribution methods in credit risk governance.

---

### Orchestrating LLM Agents for Scientific Research: A Pilot Study of Multiple Choice Question (MCQ) Generation and Evaluation

Authors: Yuan An
Published: 2026-02-21
Categories: cs.CY, cs.AI, cs.HC
arXiv: https://arxiv.org/abs/2602.18891v1

Advances in large language models (LLMs) are rapidly transforming scientific work, yet empirical evidence on how these systems reshape research activities remains limited. We report a mixed-methods pilot evaluation of an AI-orchestrated research workflow in which a human researcher coordinated multiple LLM-based agents to perform data extraction, corpus construction, artifact generation, and artifact evaluation. Using the generation and assessment of multiple-choice questions (MCQs) as a testbed, we collected 1,071 SAT Math MCQs and employed LLM agents to extract questions from PDFs, retrieve and convert open textbooks into structured representations, align each MCQ with relevant textbook content, generate new MCQs under specified difficulty and cognitive levels, and evaluate both original and generated MCQs using a 24-criterion quality framework. Across all evaluations, average MCQ quality was high. However, criterion-level analysis and equivalence testing show that generated MCQs are not fully comparable to expert-vetted baseline questions. Strict similarity (24/24 criteria equivalent) was never achieved. Persistent gaps concentrated in skill\ depth, cognitive engagement, difficulty calibration, and metadata alignment, while surface-level qualities, such as {grammar fluency}, {clarity options}, {no duplicates}, were consistently strong. Beyond MCQ outcomes, the study documents a labor shift. The researcher's work moved from ``authoring items'' toward {specification, orchestration, verification}, and {governance}. Formalizing constraints, designing rubrics, building validation loops, recovering from tool failures, and auditing provenance constituted the primary activities. We discuss implications for the future of scientific work, including emerging ``AI research operations'' skills required for AI-empowered research pipelines.

---

### SafeDrive: Fine-Grained Safety Reasoning for End-to-End Driving in a Sparse World

Authors: Jungho Kim, Jiyong Oh, Seunghoon Yu, Hongjae Shin, Donghyuk Kwak, Jun Won Choi
Published: 2026-02-21
Categories: cs.CV
arXiv: https://arxiv.org/abs/2602.18887v2

The end-to-end (E2E) paradigm, which maps sensor inputs directly to driving decisions, has recently attracted significant attention due to its unified modeling capability and scalability. However, ensuring safety in this unified framework remains one of the most critical challenges. In this work, we propose SafeDrive, an E2E planning framework designed to perform explicit and interpretable safety reasoning through a trajectory-conditioned Sparse World Model. SafeDrive comprises two complementary networks: the Sparse World Network (SWNet) and the Fine-grained Reasoning Network (FRNet). SWNet constructs trajectory-conditioned sparse worlds that simulate the future behaviors of critical dynamic agents and road entities, providing interaction-centric representations for downstream reasoning. FRNet then evaluates agent-specific collision risks and temporal adherence to drivable regions, enabling precise identification of safety-critical events across future timesteps. SafeDrive achieves state-of-the-art performance on both open-loop and closed-loop benchmarks. On NAVSIM, it records a PDMS of 91.6 and an EPDMS of 87.5, with only 61 collisions out of 12,146 scenarios (0.5%). On Bench2Drive, SafeDrive attains a 66.8% driving score.

---

### PhysConvex: Physics-Informed 3D Dynamic Convex Radiance Fields for Reconstruction and Simulation

Authors: Dan Wang, Xinrui Cui, Serge Belongie, Ravi Ramamoorthi
Published: 2026-02-21
Categories: cs.CV, cs.GR
arXiv: https://arxiv.org/abs/2602.18886v1

Reconstructing and simulating dynamic 3D scenes with both visual realism and physical consistency remains a fundamental challenge. Existing neural representations, such as NeRFs and 3DGS, excel in appearance reconstruction but struggle to capture complex material deformation and dynamics. We propose PhysConvex, a Physics-informed 3D Dynamic Convex Radiance Field that unifies visual rendering and physical simulation. PhysConvex represents deformable radiance fields using physically grounded convex primitives governed by continuum mechanics. We introduce a boundary-driven dynamic convex representation that models deformation through vertex and surface dynamics, capturing spatially adaptive, non-uniform deformation, and evolving boundaries. To efficiently simulate complex geometries and heterogeneous materials, we further develop a reduced-order convex simulation that advects dynamic convex fields using neural skinning eigenmodes as shape- and material-aware deformation bases with time-varying reduced DOFs under Newtonian dynamics. Convex dynamics also offers compact, gap-free volumetric coverage, enhancing both geometric efficiency and simulation fidelity. Experiments demonstrate that PhysConvex achieves high-fidelity reconstruction of geometry, appearance, and physical properties from videos, outperforming existing methods.

---

### TPRU: Advancing Temporal and Procedural Understanding in Large Multimodal Models

Authors: Zhenkun Gao, Xuhong Wang, Xin Tan, Yuan Xie
Published: 2026-02-21
Categories: cs.AI
arXiv: https://arxiv.org/abs/2602.18884v1

Multimodal Large Language Models (MLLMs), particularly smaller, deployable variants, exhibit a critical deficiency in understanding temporal and procedural visual data, a bottleneck hindering their application in real-world embodied AI. This gap is largely caused by a systemic failure in training paradigms, which lack large-scale, procedurally coherent data. To address this problem, we introduce TPRU, a large-scale dataset sourced from diverse embodied scenarios such as robotic manipulation and GUI navigation. TPRU is systematically designed to cultivate temporal reasoning through three complementary tasks: Temporal Reordering, Next-Frame Prediction, and Previous-Frame Review. A key feature is the inclusion of challenging negative samples, compelling models to transition from passive observation to active, cross-modal validation. We leverage TPRU with a reinforcement learning (RL) fine-tuning methodology, specifically targeting the enhancement of resource-efficient models. Experiments show our approach yields dramatic gains: on our manually curated TPRU-Test, the accuracy of TPRU-7B soars from 50.33\% to 75.70\%, a state-of-the-art result that significantly outperforms vastly larger baselines, including GPT-4o. Crucially, these capabilities generalize effectively, demonstrating substantial improvements on established benchmarks. The codebase is available at https://github.com/Stephen-gzk/TPRU/ .

---

### Characterization of Residual Morphological Substructure Using Supervised and Unsupervised Deep Learning

Authors: Kameswara Bharadwaj Mantha, Daniel H. McIntosh, Cody Ciaschi, Rubyet Evan, Luther Landry, Henry C. Ferguson, Camilla Pacifici, Joel Primack, Nimish Hathi, Anton Koekemoer, Yicheng Guo, The CANDELS Collaboration
Published: 2026-02-21
Categories: astro-ph.GA, cs.CV
arXiv: https://arxiv.org/abs/2602.18883v1

Automated characterization of galactic substructure is an essential step in understanding the transformative physical processes driving galaxy evolution. In this study, we investigate the application of deep learning (DL) frameworks to characterize different galactic substructures hosted within parametric light-profile subtracted ``residual'' images of a large sample galaxies from the CANDELS survey. We develop a supervised Convolutional Neural Network (CNN) and unsupervised Convolutional Variational Autoencoder (CvAE) and train it on the single-Sérsic profile fitting based residual images of $10,046$ bright and massive galaxies ($H<24.5\,{\rm mag}$ and $M_{\rm stellar} \geq 10^{9.5}\,M_{\odot}$) spanning $1<z<3$, in conjunction with their visual-based classification labels indicating the nature of residual substructures hosted within them. Using our unique data preprocessing approach, we prepare our residual images such that the inputs to our DL networks comprise only ``galaxy of interest'', and augment them such that our sample span uniformly across different residual characteristics. We assess the latent space of the CNN and CvAE using Principle Component Analysis (PCA) along with independently quantified metrics of residual strength (significant pixel flux $SPF$, Bumpiness, and Residual Flux Fraction). We also employ an unsupervised Gaussian Mixture Modeling (GMM) based clustering scheme with Support Vector Classification (SVC) to identify groupings in PCA space that correspond to similar residual substructure. We find that our supervised CNN latent features in PCA space correlate with the $SPF$ values and distinguish between qualitatively strong and weak residual substructures. While our unsupervised CvAE latent space also correlates with visual and quantitative residual characteristics, but lacks clear discriminatory power when characterizing different residual substructures.

---

### SceneTok: A Compressed, Diffusable Token Space for 3D Scenes

Authors: Mohammad Asim, Christopher Wewer, Jan Eric Lenssen
Published: 2026-02-21
Categories: cs.CV, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2602.18882v2

We present SceneTok, a novel tokenizer for encoding view sets of scenes into a compressed and diffusable set of unstructured tokens. Existing approaches for 3D scene representation and generation commonly use 3D data structures or view-aligned fields. In contrast, we introduce the first method that encodes scene information into a small set of permutation-invariant tokens that is disentangled from the spatial grid. The scene tokens are predicted by a multi-view tokenizer given many context views and rendered into novel views by employing a light-weight rectified flow decoder. We show that the compression is 1-3 orders of magnitude stronger than for other representations while still reaching state-of-the-art reconstruction quality. Further, our representation can be rendered from novel trajectories, including ones deviating from the input trajectory, and we show that the decoder gracefully handles uncertainty. Finally, the highly-compressed set of unstructured latent scene tokens enables simple and efficient scene generation in 5 seconds, achieving a much better quality-speed trade-off than previous paradigms.

---

### FOCA: Frequency-Oriented Cross-Domain Forgery Detection, Localization and Explanation via Multi-Modal Large Language Model

Authors: Zhou Liu, Tonghua Su, Hongshi Zhang, Fuxiang Yang, Donglin Di, Yang Song, Lei Fan
Published: 2026-02-21
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2602.18880v1

Advances in image tampering techniques, particularly generative models, pose significant challenges to media verification, digital forensics, and public trust. Existing image forgery detection and localization (IFDL) methods suffer from two key limitations: over-reliance on semantic content while neglecting textural cues, and limited interpretability of subtle low-level tampering traces. To address these issues, we propose FOCA, a multimodal large language model-based framework that integrates discriminative features from both the RGB spatial and frequency domains via a cross-attention fusion module. This design enables accurate forgery detection and localization while providing explicit, human-interpretable cross-domain explanations. We further introduce FSE-Set, a large-scale dataset with diverse authentic and tampered images, pixel-level masks, and dual-domain annotations. Extensive experiments show that FOCA outperforms state-of-the-art methods in detection performance and interpretability across both spatial and frequency domains.

---

### Structure-Level Disentangled Diffusion for Few-Shot Chinese Font Generation

Authors: Jie Li, Suorong Yang, Jian Zhao, Furao Shen
Published: 2026-02-21
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2602.18874v1

Few-shot Chinese font generation aims to synthesize new characters in a target style using only a handful of reference images. Achieving accurate content rendering and faithful style transfer requires effective disentanglement between content and style. However, existing approaches achieve only feature-level disentanglement, allowing the generator to re-entangle these features, leading to content distortion and degraded style fidelity. We propose the Structure-Level Disentangled Diffusion Model (SLD-Font), which receives content and style information from two separate channels. SimSun-style images are used as content templates and concatenated with noisy latent features as the input. Style features extracted by a CLIP model from target-style images are integrated via cross-attention. Additionally, we train a Background Noise Removal module in the pixel space to remove background noise in complex stroke regions. Based on theoretical validation of disentanglement effectiveness, we introduce a parameter-efficient fine-tuning strategy that updates only the style-related modules. This allows the model to better adapt to new styles while avoiding overfitting to the reference images' content. We further introduce the Grey and OCR metrics to evaluate the content quality of generated characters. Experimental results show that SLD-Font achieves significantly higher style fidelity while maintaining comparable content accuracy to existing state-of-the-art methods.

---

### BiMotion: B-spline Motion for Text-guided Dynamic 3D Character Generation

Authors: Miaowei Wang, Qingxuan Yan, Zhi Cao, Yayuan Li, Oisin Mac Aodha, Jason J. Corso, Amir Vaxman
Published: 2026-02-21
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2602.18873v2

Text-guided dynamic 3D character generation has advanced rapidly, yet producing high-quality motion that faithfully reflects rich textual descriptions remains challenging. Existing methods tend to generate limited sub-actions or incoherent motion due to fixed-length temporal inputs and discrete frame-wise representations that fail to capture rich motion semantics. We address these limitations by representing motion with continuous differentiable B-spline curves, enabling more effective motion generation without modifying the capabilities of the underlying generative model. Specifically, our closed-form, Laplacian-regularized B-spline solver efficiently compresses variable-length motion sequences into compact representations with a fixed number of control points. Further, we introduce a normal-fusion strategy for input shape adherence along with correspondence-aware and local-rigidity losses for motion-restoration quality. To train our model, we collate BIMO, a new dataset containing diverse variable-length 3D motion sequences with rich, high-quality text annotations. Extensive evaluations show that our feed-forward framework BiMotion generates more expressive, higher-quality, and better prompt-aligned motions than existing state-of-the-art methods, while also achieving faster generation. Our project page is at: https://wangmiaowei.github.io/BiMotion.github.io/.

---

### Federated Measurement of Demographic Disparities from Quantile Sketches

Authors: Arthur Charpentier, Agathe Fernandes Machado, Olivier Côté, François Hu
Published: 2026-02-21
Categories: stat.ML, cs.LG
arXiv: https://arxiv.org/abs/2602.18870v1

Many fairness goals are defined at a population level that misaligns with siloed data collection, which remains unsharable due to privacy regulations. Horizontal federated learning (FL) enables collaborative modeling across clients with aligned features without sharing raw data. We study federated auditing of demographic parity through score distributions, measuring disparity as a Wasserstein--Frechet variance between sensitive-group score laws, and expressing the population metric in federated form that makes explicit how silo-specific selection drives local-global mismatch. For the squared Wasserstein distance, we prove an ANOVA-style decomposition that separates (i) selection-induced mixture effects from (ii) cross-silo heterogeneity, yielding tight bounds linking local and global metrics. We then propose a one-shot, communication-efficient protocol in which each silo shares only group counts and a quantile summary of its local score distributions, enabling the server to estimate global disparity and its decomposition, with $O(1/k)$ discretization bias ($k$ quantiles) and finite-sample guarantees. Experiments on synthetic data and COMPAS show that a few dozen quantiles suffice to recover global disparity and diagnose its sources.

---

### Enhancing 3D LiDAR Segmentation by Shaping Dense and Accurate 2D Semantic Predictions

Authors: Xiaoyu Dong, Tiankui Xian, Wanshui Gan, Naoto Yokoya
Published: 2026-02-21
Categories: cs.CV
arXiv: https://arxiv.org/abs/2602.18869v1

Semantic segmentation of 3D LiDAR point clouds is important in urban remote sensing for understanding real-world street environments. This task, by projecting LiDAR point clouds and 3D semantic labels as sparse maps, can be reformulated as a 2D problem. However, the intrinsic sparsity of the projected LiDAR and label maps can result in sparse and inaccurate intermediate 2D semantic predictions, which in return limits the final 3D accuracy. To address this issue, we enhance this task by shaping dense and accurate 2D predictions. Specifically, we develop a multi-modal segmentation model, MM2D3D. By leveraging camera images as auxiliary data, we introduce cross-modal guided filtering to overcome label map sparsity by constraining intermediate 2D semantic predictions with dense semantic relations derived from the camera images; and we introduce dynamic cross pseudo supervision to overcome LiDAR map sparsity by encouraging the 2D predictions to emulate the dense distribution of the semantic predictions from the camera images. Experiments show that our techniques enable our model to achieve intermediate 2D semantic predictions with dense distribution and higher accuracy, which effectively enhances the final 3D accuracy. Comparisons with previous methods demonstrate our superior performance in both 2D and 3D spaces.

---

### Limits of Convergence-Rate Control for Open-Weight Safety

Authors: Domenic Rosati, Xijie Zeng, Hong Huang, Sebastian Dionicio, Subhabrata Majumdar, Frank Rudzicz, Hassan Sajjad
Published: 2026-02-21
Categories: math.OC, cs.LG
arXiv: https://arxiv.org/abs/2602.18868v1

Open-weight foundation models can be fine-tuned for harmful purposes after release, yet no existing training resistance methods provide theoretical guarantees. Treating these interventions as convergence-rate control problems allows us to connect optimization speed to the spectral structure of model weights. We leverage this insight to develop a novel understanding of convergence rate control through spectral reparameterization and derive an algorithm, SpecDef, that can both provably and empirically slow first- and second-order optimization in non-adversarial settings. In adversarial settings, we establish a fundamental limit on a broad class of convergence rate control methods including our own: an attacker with sufficient knowledge can restore fast convergence at a linear increase in model size. In order to overcome this limitation, future works will need to investigate methods that are not equivalent to controlling convergence rate.

---

### Similarity-as-Evidence: Calibrating Overconfident VLMs for Interpretable and Label-Efficient Medical Active Learning

Authors: Zhuofan Xie, Zishan Lin, Jinliang Lin, Jie Qi, Shaohua Hong, Shuo Li
Published: 2026-02-21
Categories: cs.CV
arXiv: https://arxiv.org/abs/2602.18867v2

Active Learning (AL) reduces annotation costs in medical imaging by selecting only the most informative samples for labeling, but suffers from cold-start when labeled data are scarce. Vision-Language Models (VLMs) address the cold-start problem via zero-shot predictions, yet their temperature-scaled softmax outputs treat text-image similarities as deterministic scores while ignoring inherent uncertainty, leading to overconfidence. This overconfidence misleads sample selection, wasting annotation budgets on uninformative cases. To overcome these limitations, the Similarity-as-Evidence (SaE) framework calibrates text-image similarities by introducing a Similarity Evidence Head (SEH), which reinterprets the similarity vector as evidence and parameterizes a Dirichlet distribution over labels. In contrast to a standard softmax that enforces confident predictions even under weak signals, the Dirichlet formulation explicitly quantifies lack of evidence (vacuity) and conflicting evidence (dissonance), thereby mitigating overconfidence caused by rigid softmax normalization. Building on this, SaE employs a dual-factor acquisition strategy: high-vacuity samples (e.g., rare diseases) are prioritized in early rounds to ensure coverage, while high-dissonance samples (e.g., ambiguous diagnoses) are prioritized later to refine boundaries, providing clinically interpretable selection rationales. Experiments on ten public medical imaging datasets with a 20% label budget show that SaE attains state-of-the-art macro-averaged accuracy of 82.57%. On the representative BTMRI dataset, SaE also achieves superior calibration, with a negative log-likelihood (NLL) of 0.425.

---

### $(α,β)$-Stability for Boosting Vector-Valued Prediction

Authors: Jian Qian, Shu Ge
Published: 2026-02-21
Categories: cs.LG, stat.ML
arXiv: https://arxiv.org/abs/2602.18866v2

Despite the widespread use of boosting in structured prediction, a general theoretical understanding of aggregation beyond scalar prediction remains incomplete. We study vector-valued prediction under a target divergence and identify a geometric stability property under which aggregation amplifies weak guarantees into strong ones. We formalize this property as $(α,β)$-stability by geometric median and show how it supports a boosting framework based on exponential reweighting and geometric-median aggregation. For vector-valued prediction, we characterize this stability property under several natural divergences: $\ell_1$ and $\ell_2$ distances for unconstrained vector-valued prediction, and TV, Hellinger, and KL for density estimation over finite probability vectors. Building on these results, we propose a generic boosting framework \geomedboost. Under a weak learner condition and $(α,β)$-stability, we obtain exponential decay of the empirical divergence error, which then yields population guarantees through a generalization bound.

---

### TIACam: Text-Anchored Invariant Feature Learning with Auto-Augmentation for Camera-Robust Zero-Watermarking

Authors: Abdullah All Tanvir, Agnibh Dasgupta, Xin Zhong
Published: 2026-02-21
Categories: eess.IV, cs.CV, cs.LG, cs.MM
arXiv: https://arxiv.org/abs/2602.18863v1

Camera recapture introduces complex optical degradations, such as perspective warping, illumination shifts, and Moiré interference, that remain challenging for deep watermarking systems. We present TIACam, a text-anchored invariant feature learning framework with auto-augmentation for camera-robust zero-watermarking. The method integrates three key innovations: (1) a learnable auto-augmentor that discovers camera-like distortions through differentiable geometric, photometric, and Moiré operators; (2) a text-anchored invariant feature learner that enforces semantic consistency via cross-modal adversarial alignment between image and text; and (3) a zero-watermarking head that binds binary messages in the invariant feature space without modifying image pixels. This unified formulation jointly optimizes invariance, semantic alignment, and watermark recoverability. Extensive experiments on both synthetic and real-world camera captures demonstrate that TIACam achieves state-of-the-art feature stability and watermark extraction accuracy, establishing a principled bridge between multimodal invariance learning and physically robust zero-watermarking.

---

### Joint Post-Training Quantization of Vision Transformers with Learned Prompt-Guided Data Generation

Authors: Shile Li, Markus Karmann, Onay Urfalioglu
Published: 2026-02-21
Categories: cs.CV
arXiv: https://arxiv.org/abs/2602.18861v1

We present a framework for end-to-end joint quantization of Vision Transformers trained on ImageNet for the purpose of image classification. Unlike prior post-training or block-wise reconstruction methods, we jointly optimize over the entire set of all layers and inter-block dependencies without any labeled data, scaling effectively with the number of samples and completing in just one hour on a single GPU for ViT-small. We achieve state-of-the-art W4A4 and W3A3 accuracies on ImageNet and, to the best of our knowledge, the first PTQ results that maintain strong accuracy on ViT, DeiT, and Swin-T models under extremely low-bit settings (W1.58A8), demonstrating the potential for efficient edge deployment. Furthermore, we introduce a data-free calibration strategy that synthesizes diverse, label-free samples using Stable Diffusion Turbo guided by learned multi-mode prompts. By encouraging diversity in both the learned prompt embeddings and the generated image features, our data-free approach achieves performance on par with real-data ImageNet calibration and surpasses simple text-prompt baselines such as "a <adjective> photo of <adjective> <cls>".

---

### Hyperbolic Busemann Neural Networks

Authors: Ziheng Chen, Bernhard Schölkopf, Nicu Sebe
Published: 2026-02-21
Categories: cs.LG, cs.AI, cs.CV
arXiv: https://arxiv.org/abs/2602.18858v2

Hyperbolic spaces provide a natural geometry for representing hierarchical and tree-structured data due to their exponential volume growth. To leverage these benefits, neural networks require intrinsic and efficient components that operate directly in hyperbolic space. In this work, we lift two core components of neural networks, Multinomial Logistic Regression (MLR) and Fully Connected (FC) layers, into hyperbolic space via Busemann functions, resulting in Busemann MLR (BMLR) and Busemann FC (BFC) layers with a unified mathematical interpretation. BMLR provides compact parameters, a point-to-horosphere distance interpretation, batch-efficient computation, and a Euclidean limit, while BFC generalizes FC and activation layers with comparable complexity. Experiments on image classification, genome sequence learning, node classification, and link prediction demonstrate improvements in effectiveness and efficiency over prior hyperbolic layers. The code is available at https://github.com/GitZH-Chen/HBNN.

---

### VariBASed: Variational Bayes-Adaptive Sequential Monte-Carlo Planning for Deep Reinforcement Learning

Authors: Joery A. de Vries, Jinke He, Yaniv Oren, Pascal R. van der Vaart, Mathijs M. de Weerdt, Matthijs T. J. Spaan
Published: 2026-02-21
Categories: cs.LG
arXiv: https://arxiv.org/abs/2602.18857v1

Optimally trading-off exploration and exploitation is the holy grail of reinforcement learning as it promises maximal data-efficiency for solving any task. Bayes-optimal agents achieve this, but obtaining the belief-state and performing planning are both typically intractable. Although deep learning methods can greatly help in scaling this computation, existing methods are still costly to train. To accelerate this, this paper proposes a variational framework for learning and planning in Bayes-adaptive Markov decision processes that coalesces variational belief learning, sequential Monte-Carlo planning, and meta-reinforcement learning. In a single-GPU setup, our new method VariBASeD exhibits favorable scaling to larger planning budgets, improving sample- and runtime-efficiency over prior methods.

---

### Issues with Measuring Task Complexity via Random Policies in Robotic Tasks

Authors: Reabetswe M. Nkhumise, Mohamed S. Talamali, Aditya Gilra
Published: 2026-02-21
Categories: cs.LG
arXiv: https://arxiv.org/abs/2602.18856v1

Reinforcement learning (RL) has enabled major advances in fields such as robotics and natural language processing. A key challenge in RL is measuring task complexity, which is essential for creating meaningful benchmarks and designing effective curricula. While there are numerous well-established metrics for assessing task complexity in tabular settings, relatively few exist in non-tabular domains. These include (i) Statistical analysis of the performance of random policies via Random Weight Guessing (RWG), and (ii) information-theoretic metrics Policy Information Capacity (PIC) and Policy-Optimal Information Capacity (POIC), which are reliant on RWG. In this paper, we evaluate these methods using progressively difficult robotic manipulation setups, with known relative complexity, with both dense and sparse reward formulations. Our empirical results reveal that measuring complexity is still nuanced. Specifically, under the same reward formulation, PIC suggests that a two-link robotic arm setup is easier than a single-link setup - which contradicts the robotic control and empirical RL perspective whereby the two-link setup is inherently more complex. Likewise, for the same setup, POIC estimates that tasks with sparse rewards are easier than those with dense rewards. Thus, we show that both PIC and POIC contradict typical understanding and empirical results from RL. These findings highlight the need to move beyond RWG-based metrics towards better metrics that can more reliably capture task complexity in non-tabular RL with our task framework as a starting point.

---

### Open-Vocabulary Domain Generalization in Urban-Scene Segmentation

Authors: Dong Zhao, Qi Zang, Nan Pu, Wenjing Li, Nicu Sebe, Zhun Zhong
Published: 2026-02-21
Categories: cs.CV
arXiv: https://arxiv.org/abs/2602.18853v2

Domain Generalization in Semantic Segmentation (DG-SS) aims to enable segmentation models to perform robustly in unseen environments. However, conventional DG-SS methods are restricted to a fixed set of known categories, limiting their applicability in open-world scenarios. Recent progress in Vision-Language Models (VLMs) has advanced Open-Vocabulary Semantic Segmentation (OV-SS) by enabling models to recognize a broader range of concepts. Yet, these models remain sensitive to domain shifts and struggle to maintain robustness when deployed in unseen environments, a challenge that is particularly severe in urban-driving scenarios. To bridge this gap, we introduce Open-Vocabulary Domain Generalization in Semantic Segmentation (OVDG-SS), a new setting that jointly addresses unseen domains and unseen categories. We introduce the first benchmark for OVDG-SS in autonomous driving, addressing a previously unexplored problem and covering both synthetic-to-real and real-to-real generalization across diverse unseen domains and unseen categories. In OVDG-SS, we observe that domain shifts often distort text-image correlations in pre-trained VLMs, which hinders the performance of OV-SS models. To tackle this challenge, we propose S2-Corr, a state-space-driven text-image correlation refinement mechanism that mitigates domain-induced distortions and produces more consistent text-image correlations under distribution changes. Extensive experiments on our constructed benchmark demonstrate that the proposed method achieves superior cross-domain performance and efficiency compared to existing OV-SS approaches.

---

### Rank-Aware Spectral Bounds on Attention Logits for Stable Low-Precision Training

Authors: Seyed Morteza Emadi
Published: 2026-02-21
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2602.18851v1

Attention scores in transformers are bilinear forms $S_{ij} = x_i^\top M x_j / \sqrt{d_h}$ whose maximum magnitude governs overflow risk in low-precision training. We derive a \emph{rank-aware concentration inequality}: when the interaction matrix $M = W^Q W^{K\top}$ has rank $r \ll d$, tail probabilities for $\max_{i,j}|S_{ij}|$ decay as $\exp(-d^{2}α^{2}/(γr))$ rather than $\exp(-dα^{2})$, where $γ> 1$ is a typicality parameter. For transformer attention where $r = d_h$, this yields $8$--$28\times$ tighter concentration than rank-agnostic bounds in modern architectures. We apply this result to FP8 training, deriving \emph{geometry-aware scale factors} that provide principled overflow guarantees without observing activations. The method computes per-layer scales from the spectral norm $\|W^Q W^{K\top}\|_2$ via implicit power iteration, includes a grouped query attention formulation that avoids key expansion, and remains compatible with fused attention kernels. Across GPT-2 XL to Llama-2-70B, geometry-aware scaling eliminates overflows in transient scenarios where delayed scaling fails, while achieving comparable downstream MMLU accuracy.

---

### When the Inference Meets the Explicitness or Why Multimodality Can Make Us Forget About the Perfect Predictor

Authors: J. E. Domínguez-Vidal, Alberto Sanfeliu
Published: 2026-02-21
Categories: cs.RO, cs.AI
arXiv: https://arxiv.org/abs/2602.18850v1

Although in the literature it is common to find predictors and inference systems that try to predict human intentions, the uncertainty of these models due to the randomness of human behavior has led some authors to start advocating the use of communication systems that explicitly elicit human intention. In this work, it is analyzed the use of four different communication systems with a human-robot collaborative object transportation task as experimental testbed: two intention predictors (one based on force prediction and another with an enhanced velocity prediction algorithm) and two explicit communication methods (a button interface and a voice-command recognition system). These systems were integrated into IVO, a custom mobile social robot equipped with force sensor to detect the force exchange between both agents and LiDAR to detect the environment. The collaborative task required transporting an object over a 5-7 meter distance with obstacles in the middle, demanding rapid decisions and precise physical coordination. 75 volunteers perform a total of 255 executions divided into three groups, testing inference systems in the first round, communication systems in the second, and the combined strategies in the third. The results show that, 1) once sufficient performance is achieved, the human no longer notices and positively assesses technical improvements; 2) the human prefers systems that are more natural to them even though they have higher failure rates; and 3) the preferred option is the right combination of both systems.

---

### Exact Attention Sensitivity and the Geometry of Transformer Stability

Authors: Seyed Morteza Emadi
Published: 2026-02-21
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2602.18849v2

We develop a sensitivity analysis for transformer attention in a geometry aligned with tokenwise computation. Our main result is the exact identity $\|J_τ(u)\|_{\infty\to1}=θ(p)/τ$ for the Jacobian $J_τ(u)$ of the tempered softmax $u\mapsto\mathrm{softmax}(u/τ)$, where $θ(p)=4\max_{S\subseteq(L)}p(S)(1-p(S))$ measures how evenly the attention distribution can be bisected rather than how concentrated it is. We combine this identity with a block-$\infty$/RMS norm under which row-stochastic attention mixing is nonexpansive. This yields a distribution-aware local Jacobian bound for multi-head attention and a sequence-length-independent Lipschitz bound on bounded input sets, with explicit dependence on width, input magnitude, temperature, and projection norms. We also identify a structural distinction between normalization placements: a pre-LN residual-sublayer Jacobian contains an additive identity term, whereas a post-LN residual-sublayer Jacobian does not. A LayerNorm projection lemma gives a sufficient condition under which the LayerNorm-only term in the post-LN expansion contracts geometrically; the condition is not tested by our experiments. Across three Pre-LN early-training runs of $774$M-parameter models, attention becomes substantially more concentrated while the median lower-bound certificate for $θ(p)$ remains near one at every sampled layer and checkpoint. This certifies near-maximal exact sensitivity for at least half of the sampled rows within each layer. A minority of rows enters a dominant-atom regime with lower exact sensitivity, consistent with the deterministic relationship between $p_{\max}$ and $θ(p)$.

---

### Exact Attention Sensitivity and the Geometry of Transformer Stability

Authors: Seyed Morteza Emadi
Published: 2026-02-21
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2602.18849v1

Despite powering modern AI, transformers remain mysteriously brittle to train. We develop a stability theory that explains why pre-LayerNorm works, why DeepNorm uses $N^{-1/4}$ scaling, and why warmup is necessary, all from first principles. Our framework has two pillars: (1) We derive the \emph{exact} operator norm of the softmax Jacobian, $\|J_{softmax}(u/τ)\|_{\infty\to 1} = θ(p)/τ$, where the balanced-mass factor $θ(p)\in(0,1)$ quantifies attention sensitivity. (2) We introduce a block-$\infty$/RMS geometry aligned with tokenwise computation, yielding Lipschitz bounds independent of sequence length. Using this framework, we prove that pre-LN preserves identity gradient paths while post-LN compounds LayerNorm Jacobians exponentially with depth, and we show that DeepNorm's $N^{-1/4}$ emerges from the quartic structure of attention's four projection matrices. We validate our theory on 774M-parameter models and find that, contrary to the intuition that attention sharpens during training to reduce sensitivity, $θ(p) \approx 1$ persists throughout. Transformer stability arises entirely from architectural gradient flow, not from attention dynamics. This finding changes how we reason about training: the architecture itself must handle sensitivity, not learned attention patterns.

---

### DUET-VLM: Dual stage Unified Efficient Token reduction for VLM Training and Inference

Authors: Aditya Kumar Singh, Hitesh Kandala, Pratik Prabhanjan Brahma, Zicheng Liu, Emad Barsoum
Published: 2026-02-21
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2602.18846v2

Vision-language models (VLMs) have achieved remarkable multimodal understanding and reasoning capabilities, yet remain computationally expensive due to dense visual tokenization. Existing efficiency approaches either merge redundant visual tokens or drop them progressively in language backbone, often trading accuracy for speed. In this work, we propose DUET-VLM, a versatile plug-and-play dual compression framework that consists of (a) vision-only redundancy aware compression of vision encoder's output into information-preserving tokens, followed by (b) layer-wise, salient text-guided dropping of visual tokens within the language backbone to progressively prune less informative tokens. This coordinated token management enables aggressive compression while retaining critical semantics. On LLaVA-1.5-7B, our approach maintains over 99% of baseline accuracy with 67% fewer tokens, and still retains >97% even at 89% reduction. With this dual-stage compression during training, it achieves 99.7% accuracy at 67% and 97.6% at 89%, surpassing prior SoTA visual token reduction methods across multiple benchmarks. When integrated into Video-LLaVA-7B, it even surpasses the baseline -- achieving >100% accuracy with a substantial 53.1% token reduction and retaining 97.6% accuracy under an extreme 93.4% setting. These results highlight end-to-end training with DUET-VLM, enabling robust adaptation to reduced visual (image/video) input without sacrificing accuracy, producing compact yet semantically rich representations within the same computational budget. Our code is available at https://github.com/AMD-AGI/DUET-VLM.

---

### Echoes of ownership: Adversarial-guided dual injection for copyright protection in MLLMs

Authors: Chengwei Xia, Fan Ma, Ruijie Quan, Yunqiu Xu, Kun Zhan, Yi Yang
Published: 2026-02-21
Categories: cs.CV
arXiv: https://arxiv.org/abs/2602.18845v2

With the rapid deployment of multimodal large language models (MLLMs), disputes regarding model ownership have become increasingly frequent, raising significant concerns about intellectual property protection. In this paper, we propose a framework for generating copyright triggers for MLLMs, enabling model publishers to embed verifiable ownership information into the model. The goal is to construct trigger images that elicit ownership-related textual responses exclusively in fine-tuned derivatives, while remaining inert in other non-derivative models. Our method constructs a tracking trigger image by treating the image as a learnable tensor, performing adversarial optimization with dual-injection of ownership-relevant semantic information. The first injection is achieved by enforcing textual consistency between the output of an auxiliary MLLM and a predefined ownership-relevant target text; the consistency loss is backpropagated to inject this ownership-related information into the image. The second injection is performed at the semantic-level by minimizing the distance between the CLIP features of the image and those of the target text. Furthermore, we introduce an additional adversarial training stage involving the auxiliary model. It is specifically trained to resist generating ownership-relevant target text, thereby enhancing robustness in heavily fine-tuned derivative models. Extensive experiments demonstrate the effectiveness of our dual-injection approach in tracking model lineage under various fine-tuning and domain-shift scenarios. Code is at https://github.com/kunzhan/AGDI

---

### When Agda met Vampire

Authors: Artjoms Šinkarovs, Michael Rawson
Published: 2026-02-21
Categories: cs.LO, cs.AI
arXiv: https://arxiv.org/abs/2602.18844v1

Dependently-typed proof assistants furnish expressive foundations for mechanised mathematics and verified software. However, automation for these systems has been either modest in scope or complex in implementation. We aim to improve the situation by integrating proof assistants with automated theorem provers (ATPs) in a simple way, while preserving the correctness guarantees of the former. A central difficulty arises from the fact that most ATPs operate in classical first-order logic, whereas these proof assistants are grounded in constructive dependent type theory. We identify an expressive fragment of both languages -- essentially equational Horn -- that admits sound, straightforward translations in both directions. The approach produces a prototype system for Agda forwarding proof obligations to the ATP Vampire, then transforming the resulting classical proof into a constructive proof term that Agda can type-check. The prototype automatically derives proofs concerning the properties of a complex field equipped with roots of unity, which took professional Agda developers two full days to complete. The required engineering effort is modest, and we anticipate that the methodology will extend readily to other ATPs and proof assistants.

---

### ABD: Default Exception Abduction in Finite First Order Worlds

Authors: Serafim Batzoglou
Published: 2026-02-21
Categories: cs.AI, cs.SC
arXiv: https://arxiv.org/abs/2602.18843v3

We introduce ABD, a benchmark for default-exception abduction over finite first-order worlds. Given a background theory with an abnormality predicate and a set of relational structures, a model must output a first-order formula that defines exceptions, restoring satisfiability while keeping exceptions sparse. We formalize three observation regimes (closed-world, existential completion, universal completion) with exact SMT verification. Evaluating ten frontier LLMs on 600 instances, the best models achieve high validity but parsimony gaps remain, and holdout evaluation reveals distinct generalization failure modes across regimes.

---

### Detecting AI-Generated Forgeries via Iterative Manifold Deviation Amplification

Authors: Jiangling Zhang, Shuxuan Gao, Bofan Liu, Siqiang Feng, Jirui Huang, Yaxiong Chen, Ziyu Chen
Published: 2026-02-21
Categories: cs.CV
arXiv: https://arxiv.org/abs/2602.18842v1

The proliferation of highly realistic AI-generated images poses critical challenges for digital forensics, demanding precise pixel-level localization of manipulated regions. Existing methods predominantly learn discriminative patterns of specific forgeries and often struggle with novel manipulations as editing techniques continue to evolve. We propose the Iterative Forgery Amplifier Network (IFA-Net), which shifts from learning "what is fake" to modeling "what is real". Grounded in the principle that all manipulations deviate from the natural image manifold, IFA-Net leverages a frozen Masked Autoencoder (MAE) pretrained on real images as a universal realness prior. Our framework operates through a two-stage closed-loop process: an initial Dual-Stream Segmentation Network (DSSN) fuses the original image with MAE reconstruction residuals for coarse localization, followed by a Task-Adaptive Prior Injection (TAPI) module that converts this coarse prediction into guiding prompts to steer the MAE decoder and amplify reconstruction failures in suspicious regions for precise refinement. Extensive experiments on four diffusion-based inpainting benchmarks show that IFA-Net achieves an average improvement of 6.5% in IoU and 8.1% in F1-score over the second-best method, while demonstrating strong generalization to traditional manipulation types.

---

### L2G-Net: Local to Global Spectral Graph Neural Networks via Cauchy Factorizations

Authors: Samuel Fernández-Menduiña, Eduardo Pavez, Antonio Ortega
Published: 2026-02-21
Categories: cs.LG
arXiv: https://arxiv.org/abs/2602.18837v2

Despite their theoretical advantages, spectral methods based on the graph Fourier transform (GFT) are seldom used in graph neural networks (GNNs) due to the cost of computing the eigenbasis and the lack of vertex-domain locality in the resulting representations. As a result, most GNNs rely on local approximations such as polynomial Laplacian filters or message passing, which limit their ability to model long-range dependencies. In this paper, we introduce an exact factorization of the GFT into operators acting on subgraphs, which are then combined via a sequence of Cauchy matrices. Building on this factorization, we propose a new class of spectral GNNs, termed L2G-Net (Local to Global Net). Unlike existing spectral methods, which are either fully global (when using the GFT) or local (when using polynomial filters), L2G-Net operates by processing the spectral representations of subgraphs and then combining them via structured matrices. Our algorithm avoids full eigendecompositions, exploiting graph topology to construct the factorization with quadratic complexity in the number of nodes, scaled by the maximum cut size between subgraphs. Experiments stressing long-range dependencies on large graphs show that L2G-Net scales to regimes out of reach for the standard GFT, and is competitive with state-of-the-art methods with orders of magnitude fewer learnable parameters.

---

### CLAP Convolutional Lightweight Autoencoder for Plant Disease Classification

Authors: Asish Bera, Subhajit Roy, Sudiptendu Banerjee
Published: 2026-02-21
Categories: cs.CV
arXiv: https://arxiv.org/abs/2602.18833v1

Convolutional neural networks have remarkably progressed the performance of distinguishing plant diseases, severity grading, and nutrition deficiency prediction using leaf images. However, these tasks become more challenging in a realistic in-situ field condition. Often, a traditional machine learning model may fail to capture and interpret discriminative characteristics of plant health, growth and diseases due to subtle variations within leaf subcategories. A few deep learning methods have used additional preprocessing stages or network modules to address the problem, whereas several other methods have utilized pre-trained backbone CNNs, most of which are computationally intensive. Therefore, to address the challenge, we propose a lightweight autoencoder using separable convolutional layers in its encoder decoder blocks. A sigmoid gating is applied for refining the prowess of the encoders feature discriminability, which is improved further by the decoder. Finally, the feature maps of the encoder decoder are combined for rich feature representation before classification. The proposed Convolutional Lightweight Autoencoder for Plant disease classification, called CLAP, has been experimented on three public plant datasets consisting of cassava, tomato, maize, groundnut, grapes, etc. for determining plant health conditions. The CLAP has attained improved or competitive accuracies on the Integrated Plant Disease, Groundnut, and CCMT datasets balancing a tradeoff between the performance, and little computational cost requiring 5 million parameters. The training time is 20 milliseconds and inference time is 1 ms per image.

---