The paper investigates how to predict the performance of decoding schedules for generative models that blend diffusion and autoregressive (AR) approaches. By modeling diffusion, AR, and hybrid models as paths on a corruption lattice, the authors define a schedule cost based on discarded parallel steps and show that the minimal zero‑cost schedule is determined by the data’s geometry, such as the treedepth of a Markov graph. They further demonstrate that schedules with fewer steps than the treedepth incur a positive cost, which can be ranked in advance using a kernel derived from pretrained weights, and validate these predictions across text, image, and video generation tasks.
arXiv:2610.08862v1 Announce Type: cross
Abstract: Household robots must accommodate new user instructions while executing ongoing tasks. Existing agents often regenerate or extensively revise the rem...
By Ci Zhang, Enfu Nan, Arman Akbari, Lin Zhao, Li Wang, Chen Wang, Weiwei Chen, Yanzhi Wang, Geng Yuan
arXiv:2506.11481v2 Announce Type: replace
Abstract: Scene Change Detection (SCD) evaluates changes using predefined query-reference (i.e., present-past) image pairs. However, this formulation overloo...
By Kyusik Cho, Suhan Woo, Hongje Seong, Euntai Kim
BEACON-SP is an ontology‑grounded Graph Retrieval‑Augmented Generation framework designed to aid clinicians in suicide risk assessment by integrating diverse clinical, behavioral, social, and temporal data. It constructs patient knowledge graphs guided by a unified suicide prevention ontology that merges the Three‑Step Theory, the Integrated Motivational‑Volitional Model, and the Suicide Social Determinants of Health Ontology. In a benchmark of 1,500 queries across 15 clinical categories, BEACON‑SP outperformed a vector‑based RAG baseline in completeness, clinical relevance, and evidence grounding, with 76.4% of paired comparisons favoring the GraphRAG approach.
By Kemal Davaslioglu, Nathan Conger, Sastry Kompella, Yalin E. Sagduyu, Nathaniel D. Bastian
MORA is a drift‑robust time‑series anomaly detection framework that addresses the ambiguity between distribution drift and true anomalies by defining a temporal change disambiguation problem. It reconstructs a local target from paired short‑ and long‑term views, using the reconstruction gap as a measure of contextual support to adjust the primary anomaly score. The method requires no drift annotations or online adaptation and demonstrates strong robustness to non‑stationarity while maintaining sensitivity to genuine anomalies on four TSAD benchmarks.
By Xudong Mou, Tiejun Wang, Rui Wang, Hui Wang, Pin Liu, Tianyu Wo, Xudong Liu, Renyu Yang
arXiv:2610.09853v1 Announce Type: new
Abstract: Pose-free feed-forward 3D Gaussian Splatting (3DGS) reconstructs a scene from sparse, unposed images in a single network pass, removing the need for ca...
By Chanung Park, Seunghyeon Song, Joo Chan Lee, Eunbyung Park, Jong Hwan Ko
arXiv:2610.10513v1 Announce Type: cross
Abstract: Language-model agents are increasingly asked to carry out open-ended scientific research, yet their results are usually graded against a known answer...
By Yinling Zhang, Langchen Liu, Dongbin Xiu, Xueyan Zou, Xu Kuang, Mengdi Wang, Shilong Liu
The paper introduces a unified evaluation framework for face de-identification (FDeID), addressing inconsistencies in current protocols. It presents UtilFace, a demographically balanced benchmark assembled from four large-scale datasets, and HiFD, a hierarchical metric that simultaneously measures identity suppression, multi-level utility preservation, and image quality. Using this framework, the authors compare adversarial, GAN-based, and diffusion-based methods, revealing trade-offs and failure modes that were previously hidden, and they release the benchmark and toolkit for reproducible research.
By Hui Wei, Hao Yu, Hui Kuurila-Zhang, Guoying Zhao
The study investigates whether large language models (LLMs) that allocate extra computation during inference—termed reasoning models—reduce classic decision biases compared to their non‑reasoning counterparts. Using 30 vignettes covering six cognitive biases and varying token budgets up to 8,192 tokens, the authors find that reasoning models are not less biased, and increased deliberation does not reliably diminish bias magnitude. Only anchoring showed a bias in the human direction, while other biases either remained unchanged or moved further from human patterns, suggesting that test‑time reasoning does not guarantee rationality.
By Obada Kraishan
arXiv:2610.09255v1 Announce Type: new
Abstract: Neural PDE surrogates are trained on numerical solver outputs that contain both physical evolution and solver-specific discretization errors. Because s...
By Ridham Patel
arXiv:2610.10076v1 Announce Type: cross
Abstract: Calibration is an essential requirement for probabilistic predictions to be useful for decision making. While state-of-the-art prediction methods oft...
By Jakob Benjamin Wessel, Sam Allen
arXiv:2610.10168v1 Announce Type: cross
Abstract: Split conformal prediction uses prediction errors on held-out (calibration) data to determine how wide the prediction intervals should be. It guarant...
By Ayush Baran Sen, Arkajyoti Saha
arXiv:2610.09446v1 Announce Type: new
Abstract: Large language models (LLMs) should abstain from scientific multiple-choice questions when no option is valid, but frequent abstention alone does not d...
By Benjamin Wilcox, Dawei Gao, Pradeeban Kathiravelu, Douglas Causey, Kewei Sha, Yunhe Feng
arXiv:2610.09464v1 Announce Type: new
Abstract: Political discourse often uses rhetorical and persuasive language to frame narratives, influence public opinion, and mobilize audiences. While Bangla n...
By Rohit Kumar Sen, Anik Chowdhury
arXiv:2610.09660v1 Announce Type: new
Abstract: Automatic Short Answer Scoring (ASAS) requires models that can score student responses against question-specific criteria while remaining efficient and...
By Zhifan Sun, Sebastian Gombert, Fabian Zehner, Leon Camus, Longwei Cong, Hendrik Drachsler
arXiv:2610.09665v1 Announce Type: new
Abstract: On-policy post-training can improve large language models by learning from their own trajectories, but requires costly rollout generation. We ask wheth...
By Tianle Wang, Jiayu Liu, Ruizhi Zhao, Ning Miao
The paper introduces TriviaRoomQA, a multilingual benchmark of 3,300 parallel multiple-choice questions across 288 topics, plus an additional 5,340 French-only questions. It evaluates 30 open-weight large language models (7–70 B parameters) from various regions, finding strong performance on knowledge-intensive subjects like history and geography but weaker results on everyday popular‑culture topics such as celebrities, music, movies, and news. The study also shows that model performance varies across languages for the same questions, indicating that factual knowledge access is not always language‑independent.
By Anna Mosolova, Djam\'e Seddah
WAPR is a zero‑shot wide‑angle pose refinement model that can correct candidate 6D poses with rotational errors up to 90°, achieving fast inference (≤1 s per frame) and high throughput (≈25 detections per second). It leverages rotational symmetry priors to canonicalize pose targets and introduces the SA6D dataset, which augments 944 GSO scans into ~50 k object instances and ~2 M RGB‑D images. Experiments on seven BOP core datasets demonstrate that WAPR sets new state‑of‑the‑art performance for unseen‑object pose estimation in both fast and unconstrained settings.
By Yulin Wang, Mengting Hu, Hongli Li, Jianghao Zhou, Chen Luo
MeshCarve introduces a flow‑matching approach that generates artisan meshes entirely within compact latent spaces, separating vertex positions and edge connections to avoid joint encoding issues. It employs a hierarchical sparse transformer backbone with VertexVAE and EdgeVAE, which anchor on discrete vertices to drastically shorten token sequences and use spatial‑aware compression without sacrificing reconstruction quality. The method outperforms state‑of‑the‑art autoregressive and flow‑matching techniques on Objaverse and generalizes to Toys4K, representing one of the first fully compressed latent artisan mesh generators.
By Xiyu Wang, Ruocheng Wu, Yufei Wang, Zhihao Li, Lanqing Guo, Bihan Wen
The paper introduces a domain-informed adaptive sampling method for physics-informed neural networks (PINNs) applied to metal additive manufacturing. By combining a conditional Flow Matching model with a domain-informed base distribution, the approach generates adaptive collocation points that better capture process-dependent heat dissipation regions. Experiments on numerical benchmarks show a 62.1% reduction in relative L2 error compared to state‑of‑the‑art PINN baselines.
By Hyeonsu Lee, Jihoon Jeong