OptiSAR-Net++ introduces a new cross‑domain remote sensing visual grounding task (CD‑RSVG) and the first large‑scale benchmark dataset, OptSAR‑RSVG. The framework replaces Transformer decoding with a CLIP‑based contrastive approach, employing a patch‑level Low‑Rank Adaptation Mixture of Experts for efficient cross‑domain feature decoupling and a text‑guided dual‑gate fusion module for improved semantic‑visual alignment. Experiments show state‑of‑the‑art performance on OptSAR‑RSVG and DIOR‑RSVG, with notable gains in localization accuracy and computational efficiency.
By Xiaoyu Tang, Jun Dong, Jintao Cheng, Rui Fan
The paper introduces Pre-hoc Sparsity (PrHS), a method that selects key-value (KV) cache entries before attention scoring to avoid posterior bias in large language model inference. By bounding mutual‑information loss through the dropped attention mass, PrHS offers explicit accuracy control and implements three orthogonal selectors across time, depth, and layer. Experiments on LLaMA and Mistral models show that PrHS cuts retrieval overhead by over 90%, achieves higher sparsity than HShare, and delivers significant speedups and reduced FLOPs on NVIDIA A100 GPUs while maintaining near‑dense accuracy.
By Yifei Gao, Lei Wang, Rong-Cheng Tu, Qixin Zhang, Jun Cheng, Dacheng Tao
J-Zero introduces a unified Challenger–Solver–Judge self‑evolution framework that operates without any initial data. The Challenger and Solver co‑evolve through adversarial task generation and response improvement, while the Judge adapts using known preference pairs derived from the Solver’s outputs rather than its own scores. Experiments show J‑Zero surpasses baselines by 4.2 points on verifiable tasks and 8.0 points on unverifiable tasks, maintaining improvement over ten iterations versus baseline degradation after two.
By Gyouk Chu, Myeongho Jeon, Teresa Yeo, Eunho Yang
The paper introduces SO(3) Equivariant Neural Kalman Networks (SENK), a model that predicts vibrational spectra by cascading an equivariant transformer backbone, an Equivariant Neural Kalman bridge, and an NBO-informed electronic-prior pathway. SENK improves accuracy over DetaNet on QM9S and QMe14S datasets while maintaining full-spectrum IR and Raman fidelity from small molecules to drug-like systems. It also remains stable and selectively enhances spectrally sensitive features in biomolecular systems with complex stereoelectronic effects, enabling transferable vibrational spectroscopy across diverse molecular materials.
By Zetong Li, Zhuosong Xie, Hengyu Fan, Jiaao Yu, Qiyao Hua, Zheng Lu, Liming Xu, Juanni Wu, Honglin Li
The paper introduces the Transformer-Mamba for Flow Field (TM4FF), a physics-constrained operator learning model designed to improve partial differential equation solving in computational fluid dynamics. TM4FF incorporates a Residual Wavelet Mamba layer for feature denoising, a Transformer-based attention mechanism for enhanced feature fusion, and a physics-informed loss that enforces the Navier-Stokes equations using Fourier derivatives. Experiments on four CFD datasets demonstrate that TM4FF achieves high accuracy and robust generalization across varying flow conditions.
By Zhuo Zhang, Shun Zou, Canqun Yang, Xi Yang
The paper examines how users’ prompts to large language models evolve over time, applying speech act and politeness theory to a corpus of 2,000 English prompts from 2023 and 2025. It finds a shift toward more indirect, implicit, and fragmentary directive speech acts, with a notable 14.9‑percentage‑point drop in explicit propositional content and a decline in politeness markers. This suggests users increasingly rely on the model’s inferential abilities, treating it as a competent implicature resolver.
By Kristina \v{S}ekrst, Virna Karli\'c
The study investigates whether Jev, a typed classifier that outputs probabilities over allowed answers without generating text, can replace large language model (LLM) rubric judges. Across nine panels from seven benchmarks, Jev’s accuracy differed significantly from LLM judges in only 8 of 27 paired comparisons, performing best on binary criteria and worse only on graded ones, while most other comparisons were inconclusive. In terms of cost and speed, Jev was 29 to 325 times cheaper and 30 to 220 times faster than the flash‑tier LLM judges, and a cascade approach that defers uncertain Jev verdicts to an LLM yielded only modest gains.
whyItMatters":"The findings suggest that a lightweight classifier like Jev can serve as an efficient first‑stage evaluator, potentially reducing the reliance on expensive and slow LLM judges in automated grading pipelines."
By Delip Rao, Chris Callison-Burch
The paper presents a new approach to automatic audio description (AD) that treats the task as a constrained global optimization problem. It jointly decides what visual content is narratively important, when it can be spoken without overlapping dialogue, and how to phrase it within time limits. Using large language models for salience estimation and a mixed‑integer linear program for scheduling, the system outperforms prior methods on the REFRAMED benchmark, especially in temporal placement and narrative relevance.
By Igor Sterner, Mirella Lapata, Alex Lascarides, Frank Keller
The paper introduces Smol‑VL‑BLV, a compact vision‑language model designed for blind and low‑vision users. It employs a 500M decoder transformer with teacher‑student distillation and Group Relative Policy Optimization to add spatial detail, directional cues, and hazard detection to post‑training. After a lightweight finetuning step, the model achieves significant gains on spatial, social, OCR, and VQA benchmarks while remaining under 450 MB and running entirely offline on a mid‑range Android phone.
By Rishabh Choudhary, Shreyansh Raj, Umesh Goyal, Shubh Kashyap, Shrestha Kumar, Sushovan Jena, Komal Kumar, Hisham Cholakkal, Aditya Nigam
LOGIC is a framework that performs contextual biasing for speech large language models by integrating directly in logit space. It decouples context injection from input processing, allowing explicit control over biasing strength and reducing entity word error rates by an average of 9% relative to baseline methods. When combined with prompting, LOGIC further lowers entity word error rates by 5% relative to prompt-only approaches, with only a modest 2.8% runtime overhead.
By Peidong Wang, Jian Xue, Jinyu Li
The study examines whether audio language models encode distinctive phonetic features similarly when processing spoken versus written input. By comparing mean representations of minimal phoneme pairs across six models, seven features, and 15 languages, the authors find that only voicing in the Qwen2.5-Omni models consistently shows a shared directional representation after correcting for multiple tests. The results suggest that the model family, rather than its size, determines how phonetic features are represented across audio and text streams.
By Yuanhao Chen, Peter Chin
The paper addresses look‑ahead bias in large language models (LLMs) used for financial prediction, which arises because LLMs are trained on long time‑series data. It proposes a low‑cost solution that adjusts the logits of a base model at inference time using two smaller, specialized models—one fine‑tuned to forget certain information and another to retain it. Experiments show that this method removes both verbatim and semantic knowledge, corrects biases, and outperforms previous approaches.
By Humzah Merchant, Bradford Levy
The study evaluates OpenAI’s o1‑preview large language model on a wide range of complex reasoning tasks across domains such as computer science, mathematics, natural sciences, medicine, linguistics, and social sciences. It reports high success rates, including 83.3% on competitive programming, 100% on high‑school math reasoning, and superior performance in radiology reporting, chip design, anthropology, geology, quantitative investing, and social media analysis. While the model excels at intricate reasoning and knowledge integration, it still shows occasional errors on simpler problems and struggles with some highly specialized concepts.
By Tianyang Zhong, Zhengliang Liu, Yi Pan, Yutong Zhang, Zeyu Zhang, Yifan Zhou, Shizhe Liang, Zihao Wu, Yanjun Lyu, Peng Shu, Xiaowei Yu, Chao Cao, Hanqi Jiang, Hanxu Chen, Yiwei Li, Junhao Chen, Huawen Hu, Yiheng Liu, Huaqin Zhao, Shaochen Xu, Haixing Dai, Lin Zhao, Ruidong Zhang, Wei Zhao, Zhenyuan Yang, Jingyuan Chen, Peilong Wang, Wei Ruan, Hui Wang, Huan Zhao, Jing Zhang, Yiming Ren, Shihuan Qin, Tong Chen, Jiaxi Li, Arif Hassan Zidan, Afrar Jahin, Minheng Chen, Sichen Xia, Jason Holmes, Yan Zhuang, Jiaqi Wang, Bochen Xu, Weiran Xia, Jichao Yu, Kaibo Tang, Yaxuan Yang, Bolun Sun, Lifeng Chen, Tao Yang, Guoyu Lu, Xianqiao Wang, Lilong Chai, He Li, Jin Lu, Xin Zhang, Bao Ge, Xintao Hu, Lian Zhang, Hua Zhou, Lu Zhang, Shu Zhang, Zhen Xiang, Yudan Ren, Jun Liu, Xi Jiang, Yu Bao, Wei Zhang, Xiang Li, Gang Li, Wei Liu, Dinggang Shen, Andrea Sikora, Xiaoming Zhai, Dajiang Zhu, Tuo Zhang, Tianming Liu
SplatLabel is an automated pipeline that uses a 4D Gaussian representation to generate LiDAR segmentation and semantic occupancy grids with predictive confidence. It models dynamic scenes through an explicit temporal manifold, tracking moving actors without requiring pre‑annotated 3D bounding boxes. By integrating 360‑degree LiDAR depth maps and distilling soft probabilities from 2D models, it resolves semantic ambiguities over time and space, and evaluates pseudo‑labels via a selective classification framework that balances precision and recall.
By Nitya Nanvani, Andras Palffy, Holger Caesar
The Epydemix Agent Framework adds an AI‑friendly layer to the Epydemix Python library for stochastic epidemic modeling. It provides four key features: model and parameter discovery, declarative scenario validation, execution via tested code, and result inspectability, enabling an AI agent to manage the entire modeling workflow from natural‑language input to quantitative outputs. The authors demonstrate the framework with a vaccination strategy case study and evaluate it over 50 agent sessions, showing reductions in turns, output tokens, and cost compared to direct Python use.
By Nicol\`o Gozzi, Ciro Cattuto, Alessandro Vespignani
CounterRoute is an online reinforcement‑learning framework that jointly learns how to route a language model’s reasoning between a ‘think’ and a ‘direct answer’ mode, using a single shared policy derived from a dual‑mode checkpoint. It employs counterfactual rollouts to credit routing decisions and a curriculum that starts with forced dual‑mode rollouts before shifting to self‑routed updates, achieving better accuracy‑efficiency trade‑offs across nine benchmarks. The method reduces generated tokens by up to 51% on Qwen3‑8B while improving macro‑average accuracy, and its routing strategy generalizes to unseen coding, science, and commonsense tasks.
By Ruochen Jiao, Besnik Fetahu, Zhenyu Shi, Priyanka Nigam
The paper "Learning to Ideate for Scientific Impact" explores using delayed signals of scientific uptake—specifically citation-normalized impact—as feedback to steer large language models toward generating high‑impact research ideas. The authors build a dataset of over 100,000 computer science papers, train a reward model to predict citation impact from goal‑idea pairs, and align an idea generator via supervised fine‑tuning and reinforcement learning. Evaluation with a reference‑grounded protocol shows that the RL‑tuned model consistently produces ideas with higher estimated impact than baseline models.
By Shubham Kale, Aniketh Garikaparthi, Manasi Patwardhan
The study investigates whether large language models (LLMs) can identify code they have generated, potentially leading to self‑favoring or collusive behavior. Experiments across 15 model‑benchmark pairs show that models can attribute authorship with balanced accuracy between 49% and 58%, but this ability largely stems from superficial cues such as solution length. Removing surface features like docstrings, comments, and type hints reduces attribution accuracy to chance, indicating that surface cues drive the effect.
By Ehsan Barkhordar, Surendrabikram Thapa
The paper examines how large language models (LLMs) encode script knowledge across their layers using logistic regression probing and logit‑lens analysis. Findings show that both the input and instructed output scripts are represented in the earliest layers, but the model commits to the actual output script only in the final layers, with intermediate representations defaulting to Latin. This two‑stage process is consistent across methods and is more pronounced in larger models, suggesting a link between model depth and script commitment.
By David Kletz, Sandra Mitrovi\'c, Itay Sabato, Ljiljana Dolami\'c, Fabio Rinaldi
The paper investigates whether large language models (LLMs) assess politeness in ways that match human judgments. Using two English datasets—one with continuous ratings and another with three‑way categorical labels—the authors compare seven LLMs to human annotations. They find that models agree more with each other than with humans, show systematic neutral bias in categorical predictions, and that alignment varies with explicit linguistic cues and rapport‑building strategies.
By Rong Wang, Kun Sun, Yadong Guo