The paper investigates how asking vision‑language models to report what a director sees can alter their subsequent action choices. In a Director Task, prompting the model to identify visible blocks before selecting a block increases the likelihood of choosing a hidden block by up to 88 percentage points across several GPT models and 17 points for Qwen3.8‑27B, despite accurate reports of visible blocks. The authors argue that the question shifts the model’s perspective from the director’s to its own, and they propose a baseline paired‑prompt design for future perspective‑taking benchmarks.
By Jiaxuan Guo, Xingyu Ren, Kejia Zhang, Jingxin Yang, Youran Sun, Haizhao Yang
The paper investigates how model size and quantization affect social bias and abstention in question answering. Using the BBQ benchmark and models from Qwen3, Gemma-3, and SmolLM2, it finds that larger models improve accuracy but bias does not scale simply, while quantization effects differ across settings. Importantly, many configurations with fewer biased responses also show increased abstention, suggesting that lower bias may partly result from more frequent abstention.
By Kuniko Paxton, Ifeoluwa Wuraola, Dhavalkumar Thakker
UniRAP is a unified model that infers part-level physical affordances and translates language instructions into actionable geometric representations. It treats the task as a conditional multimodal generation problem, integrating visual inputs, textual instructions, and optional prompts into a shared spatiotemporal representation via a unified interface token mechanism. The model employs a Unified Affordance Decoder to jointly perform object detection, part-level affordance segmentation, and 4‑DoF interaction pose estimation, and uses a curriculum-based transfer training strategy to improve data efficiency with limited textual annotations.
By Linfei Li, Ruining Hu, Lin Zhang, Zhong Wang, Fengyi Zhang, Ying Shen, Binqiang Wang, Xin Zhang
ThreadRAG is a retrieval‑augmented generation framework that preserves contextual consistency by structuring retrieved documents into Semantic Nodes and building a Semantic Topology Graph (STG). It uses Energy‑Driven Pathfinding (EDP) to identify query‑relevant subgraphs and linearize them into coherent textual evidence paths. Experiments on HotpotQA, 2WikiMultihop, and another benchmark show ThreadRAG outperforms strong baselines, achieving a 14.4% EM gain on HotpotQA and a 5.8% EM lead on 2WikiMultihop while cutting token cost to 23.3% of the strongest baseline.
By Yufeng Li, Zaijing Li, Fuchen Zhang, Zongzong Wu, Ming Zhao, Fengxiao Tang
The paper investigates whether fine‑tuning large language models (LLMs) on value survey responses can shift their value‑conditioned outputs and downstream behavior. Researchers first estimate baseline value profiles for several open‑source LLMs using the ten Schwartz values, then fine‑tune the models on survey‑style supervision that selectively alters target value ratings while preserving non‑target values. The resulting shifts are evaluated on held‑out survey questions and on out‑of‑domain behavioral proxies, such as a contextual moral‑judgment benchmark from Reddit AITA posts and a text‑based interactive decision‑making benchmark. The study finds that survey‑based supervision can change both survey responses and downstream behavior, but the transfer is strongly model‑ and value‑dependent and influenced by how value descriptions are framed.
By Shangrui Nie, Florian Mai, David Kacz\'er, Charles Welch, Zhixue Zhao, Lucie Flek
The paper introduces GENIE, a fine‑grained metric designed to assess the novelty of responses generated by large language models on a task‑specific basis. Unlike holistic metrics, GENIE evaluates novelty across multiple features, revealing which aspects of model outputs are truly novel. The authors also apply GENIE to evaluate mitigation methods aimed at enhancing creativity, showing where such methods succeed or fall short.
By Ramya Namuduri, Manya Wadhwa, Anshun Asher Zheng, Greg Durrett, Junyi Jessy Li
The paper evaluates whether a 9‑billion parameter decision foundation model, Clef‑Flash, is necessary for symptom‑to‑disease classification. Fine‑tuning the model on the 24‑class Symptom2Disease benchmark yields 95.42% accuracy, but simple TF‑IDF linear classifiers achieve higher accuracy (98.75%) with far faster inference. Ablation studies show that predictions can hinge on single descriptive clauses, indicating that the benchmark may not fully capture the model’s capabilities.
By S. M. Mahbub Uddin Ahmed
The paper introduces Dynalize, an online penalty system that mitigates reward hacking in preference reward models during policy training. By having a language model periodically analyze rollouts and generate penalties for behaviors the reward model inadvertently rewards, Dynalize prevents the policy from inflating scores without improving quality. Experiments in clinical counseling and creative writing show that Dynalize sustains longer training and yields better downstream performance, suppressing undesirable patterns even in multi‑turn settings and after retraining the reward model.
By Afonso Baldo, Ant\'onio Farinhas, Marcos Treviso, Ricardo Rei, Nuno M. Guerreiro
PRISM is a training‑free framework for financial information retrieval that combines prompt engineering, in‑context learning, and a candidate refinement module, with optional multi‑agent coordination. On FiQA‑2018 it improves Recall@100 from 0.7783 to 0.8326, and on FinAgentBench it achieves a Cohere‑refined top‑50 pool that raises NDCG@5 while cutting per‑query cost by about 35%. The authors also release an expert‑annotated evaluation set, a label‑robustness sweep, and a Pareto frontier of quality‑cost‑latency trade‑offs, and PRISM ranks third on the private FinAgentBench leaderboard without fine‑tuning.
By Chun Chet Ng, Jia Yu Lim, Zhen Hao Chu, Yixi Zhou, Wei Zeng Low
DGCM-SSFA is a compact multi‑domain attention framework that synthesizes high‑fidelity 7T MRI from routine 3T scans using only 11 million parameters. It combines a Dual‑Granularity Cross‑Modulated Feature Modulator (DGCM) and a Sparse Strip Frequency Attention (SSFA) module within a U‑Net Transformer backbone to enhance local edge details and capture global spatial context efficiently. Experiments on UNC and BNU 3T‑to‑7T datasets show competitive structural fidelity, achieving SSIM scores of 0.8431 (BNU T2w) and 0.8100 (BNU T1w) while remaining far smaller than Transformer‑based alternatives.
By Zhichun Fan, Zhiqiang Chen, Lanling Liu, Lingjun Liu, Zhonghua Xie
The paper proposes a training‑free acceleration technique for transformer‑based blind compressed image restoration models. By profiling computation hotspots and channel‑wise activation distributions, the authors identify recurrent projection operators that can be replaced with low‑rank approximations. They introduce Activation‑Calibrated Singular Value Decomposition (ASVD), which derives low‑rank factors offline from pretrained weights and a few calibration images, achieving significant parameter and computation reductions while improving PSNR‑B over a TSVD baseline.
By Suhang Yao, Xin Li, Xijun Wang, Yufei Wang, Yeying Jin, Zhibo Chen
The paper introduces Iterative Gaussian Synopsis, a top‑down unfolding framework that generates compact, progressive levels of detail for 3D Gaussian Splatting (3DGS). Starting from a full‑resolution 3DGS model, it iteratively prunes Gaussian primitives using a learnable mask, building a multi‑level hierarchy that preserves visual quality while reducing storage. The method combines hierarchical spatial grids with a shared Anchor Codebook to capture global structure and local details, enabling efficient, level‑specific rendering with minimal data overhead.
By Yuqin Lu, Yihua Dai, Yang Zhou, Guiqing Li, Shengfeng He
SpineContextResUNet is a lightweight 3D Residual U‑Net that uses a Context Block with parallel multi‑dilated convolutions to efficiently capture long‑range anatomical dependencies for spine CT segmentation. It achieves Dice scores of 88.17% on VerSe2020 and 88.13% on CTSpine1K while maintaining a small ~1.7 M parameter footprint. Compared to resource‑heavy Transformers and baselines that fail on commodity hardware, SpineContextResUNet delivers robust inference on edge devices such as the Nvidia Jetson Orin Nano.
By K S Nithurshen, Saurabh J. Shigwan
SDPose leverages pre‑trained diffusion models, specifically the Stable Diffusion U‑Net, to extract robust latent features for human pose estimation. By identifying the most effective upsampling levels, fusing these features with a lightweight aggregation module, and jointly training with keypoint heatmap supervision and a latent reconstruction loss, SDPose achieves strong cross‑domain generalization. The authors introduce the COCO‑OOD benchmark, comprising style‑transferred and corrupted subsets, to evaluate robustness, and demonstrate that SDPose matches Sapiens on COCO while outperforming competitors in zero‑shot cross‑domain scenarios.
By Shuang Liang, Jing He, Lejun Liao, Chuanmeizhi Wang, Guo Zhang, Yingcong Chen, Yuan Yuan
The paper introduces TOUCKANs, a variant of Kolmogorov–Arnold networks that use finite‑degree polynomial outer functions to create a decomposable sum–product structure. This design allows expectations, variances, and higher moments to be computed via deterministic recursions over one‑dimensional integrals, eliminating the need for costly Monte Carlo sampling even in high dimensions. Experiments on uniform and Gaussian benchmarks up to 50 dimensions show deterministic mean and variance errors below those of fixed‑budget Monte Carlo, and the method successfully enforces conservation laws and normalization constraints in physical simulations such as Cahn–Hilliard mass conservation and Schrödinger eigenpair computation.
By Juan Diego Toscano, Sascha Ranft, Zhiwei Gao, Li-Lian Wang, George Em Karniadakis
VoxProg introduces voxel programming for 3D spatial reasoning, enabling language models to translate task interpretations into executable programs that operate on local voxel views of scene surfaces. By indexing surface geometry on regular grids, the approach isolates task-relevant regions and boundaries, allowing downstream pose programs to compute target SE(3) poses. The method is evaluated on the 71-task VoxProg-Bench benchmark, showing consistent superiority over relation-based, keypoint-based, and mesh-based baselines.
By Zhichao Zhai, Menglei Zhao, Guikun Chen, Zihao Yan, Kanzhi Wu, Wenguan Wang, Jun Xiao
BIRD‑RL is a new agentic reinforcement learning framework designed to scale training of small language models (SLMs) in data‑centric, stateful environments such as SQL debugging. It introduces a four‑stage curriculum that alternates between improving agent reasoning and trajectory‑adaptive exploration, and relies on a Trajectory‑scoped Persistent Execution Infrastructure to safely reuse database replicas while maintaining trajectory isolation. This approach yields a 12.8× improvement in spatial efficiency and enables 7B and 14B models to achieve competitive success rates on the BIRD‑CRITIC‑SQLite benchmark, matching proprietary agents like Claude‑Opus‑4.6.
By Ge Qu, Jinyang Li, Xiaohan Xu, Xiaolong Li, Bowen Qin, Nan Huo, Ziwei Tang, Chenhao Ma, Reynold Cheng
The paper introduces hyperspherical simplex representations derived from softmax outputs and logits to enable backward-compatible retrieval across independently trained models. By exploiting the simplex geometry induced by cross‑entropy training, the authors propose a deterministic linear transformation that realigns class prototypes when new classes are added, eliminating the need for compatibility‑specific training or architectural changes. Theoretical analysis under von Mises–Fisher assumptions shows that these aligned regular‑simplex representations satisfy the formal backward‑compatibility conditions, and empirical results demonstrate competitive performance on standard benchmarks without extra losses or mappings.
By Simone Ricci, Niccol\`o Biondi, Federico Pernici, Alberto Del Bimbo
The paper introduces the Velocity Ambiguity Profile (VAP), a time‑local Bayes‑risk floor for velocity prediction in flow matching, and decomposes the loss into an irreducible ambiguity term and a reducible excess term. It provides statistical benchmarks for intrinsic local‑regression and Gaussian‑mixture scenarios, and validates the diagnostic with controlled experiments and CIFAR‑100 DINOv2 feature flow, showing that guided time sampling can reduce velocity loss and Fréchet distance.
By Yunpeng Mei, Xiaowen Zhu, Chenyu Wang, Chenbo Xin, Hongjie Cao, Jiamin Wang, Jie Chen, Gang Wang
The paper introduces the InterModel Vigorish (IMV), a metric that quantifies differences in probabilistic fit as an expected betting advantage over a reference model. It presents three applications—SHAP-IMV for feature attribution, Multi-IMV for class discrimination, and Ablate-IMV for assessing architectural components—tested across diverse data types and multiple random seeds. The authors provide an open-source library (imvpy) and notebooks to replicate and extend their work.
By Jiani Yan, Pengyun Wang, Klint Kanopka, Ben Domingue, Charles Rahal