LeanGRPO：消除扩散强化学习中的冗余重计算

LeanGRPO重排数据并行并引入两种免重算调度，消除扩散强化学习冗余重算，实现最高1.83倍加速且不损原目标。

增强金融问答：银行财务报表的新型基准数据集

面向银行财报问答构建基准，近千问覆盖多机构年报，分块增强与检索优化使检索及答案准确率大幅提升。

Interface-Induced Trajectory Censoring

FiMI银行：面向印度零售银行业的自主模型

偏好优化安全拒答升至80%，可验证奖励RL优化工具任务并减29%词元，两种方法互补。

MetaStructAtlas：面向全身PET/CT功能与结构推理的定位式3D视觉语言数据集及基准

提出MetaStructAtlas和MetaStructVQA，含490套PET/CT、5万掩膜、10万问答对，支持全身功能与结构推理。

From Nowcasting to Forecasting: Adapting a Reanalysis-Trained

Extracting Forgotten Prompts from Targeted Unlearned Models

WeatherNext 3: Increasing resolution and performance of global weather models with raw observations

The Dually Flat Geometry of Planning as Inference

Instruction Duplication as an Inference-Time Control Primitive

Epistemic Warrant for LLM Recommendations: Characterizing the Basis for Reliance When Ground Truth Is Unavailable

Resolution-Aware Experimental Design under Partial Identifiability

Out-of-Distribution Generalisation with Sequence Models in Offline Multi-Agent Reinforcement Learning

Federated Causal Discovery via Regression-Directed Cumulants

LLM4CKD: Large Language Models for Early Stage Chronic Kidney Disease Screening

InSituMeasure: Probing Situated Measurement Grounding in Industrial Scenes with Multimodal Large Language Models

FLY-EVAL++: An Evidence-Driven Evaluation Protocol for Safety-Constrained Flight Prediction with Large Language Models

LLaDA-Image: Building Strong Image Generators with Fully Open Training Recipes

Unfold The World: Factorize 4D Properties in Reinforcing Spatial Reasoning

RealCADBench: Benchmarking Parametric CAD Modeling from Industrial Design Intents

Sequential Beats Joint: On the Interplay between On-Policy Distillation and RLVR

CROCODIL: Cross-Model Code Editing with LLMs

Two-Stage Reinforcement Learning for Sound and Adversarial Test Generation in Code LLMs

Projected Riemannian Gradient Descent for the Bures-Wasserstein Barycenter: Dimension-Independent Linear Convergence at Unit Step Size

On the Interaction Between Model Compression and Test-Time Adaptation

Neural-Network Maxent: a general extension with learned nonlinearity, applied to time-series for Desert Locust distribution modelling

IRWOZ 2.0: A Large Language Model-driven Dialogue Dataset for Industrial Robot Conversations

DRACO: Fine-Grained Credit Assignment with Dynamic Rubrics for Long-Horizon Agent Training

Why Gated DeltaNet Survives 4-Bit Quantization: NVFP4 W4A4 for the Recurrent Half of a Hybrid 27B LLM

Spurious Advantage Hidden in GRPO

Urban Boundaries, Social Barriers: A Benchmark and Vision-Centric Framework for Mapping Gated Communities and Equity Implications

VisCAD: A Foundation Model Suite with Multimodal Industrial CAD Intelligence

Fixed Suffix Dependency Ratio: Quantifying the Dual-Track Mechanism of Gender Assignment in Latvian Loanwords

SPARK: Input-Conditioned Sparse Activation Modulation for Frozen DiT-based Super-Resolution

RuleMem: Active Rule Memory for Long-Term Conversational Agents

Beyond Majority Vote: Multi-Perspective Adjudication for Medical Hallucination Detection

Local Updates, Global Learning (LUGL): Playing Games with non-incremental Learners

KnowVis: Knowledge-Centric Visual Summarization for Video Lectures

A Reverse Sign Language Dictionary: Open-Vocabulary Sign Recognition from Continuous Signing via Video Captioning and Description Retrieval

Alignment-Free Text-Audiobox for Voice Dubbing and Full-Duplex Dialogue Synthesis

Investigating the Ability of Large Language Models to Analyze Recipes for Diabetes

Fill My Mirror: Geometry-Constrained Mirror Inpainting

Representational alignment yields generalizable safety in language models

ENEAS: Embedding-guided Neural Ensemble for Adaptive Segmentation

Translation as a Decision Space: A Multi-Agent Perspective on Low-Resource Dialect Generation

面向移动网络站点能效低效识别的对等相对表示学习框架

提出无监督对等相对方法，用能量感知MDE嵌入识别高能耗站点，优于传统异常检测，助力大规模网络节能优化。

可溯源TTS：迈向强溯源性的无水印语音合成

提出联合训练法实现无水印TTS强追溯，在保持甚至提升音频质量的同时增强泛化性。

重新思考大语言模型的在策略蒸馏（二）：单个训练样本

单查询在策略蒸馏即达全数据大部分收益，源于状态覆盖高而吸收慢；瓶颈在算法。

用于多尘农业条件的传感器测试环境概念

提出室内外试验场概念，含封闭粉尘循环试验台与真实农业场景，用于评估粉尘对传感器及算法的影响。

从欺骗性输出到欺骗性机制：语言模型欺骗研究的因果框架

新框架区分欺骗行为与机制；实验表明欺骗性行为可无对应机制，接收者信息能因果影响偏好，但非模型能动性。