AGSA-Net is a hyperspectral image classification framework that incorporates spectral unmixing priors through an abundance-guided self‑attention network. It first estimates physically meaningful subpixel abundance maps with non‑negativity and sum‑to‑one constraints, then uses these abundances to build an affinity prior that directs a spectral transformer to focus on class‑discriminative interactions. The transformer features are fused with compact abundance descriptors for final classification, and experiments on Indian Pines, Augsburg, and Berlin datasets show improved performance, especially in heterogeneous urban scenes.
By Nafisa Anjum, Satavisa Dey Borno, Ananna Saha, Mir Faiyaz Hossain, Sifat Momen, Nabeel Mohammed, Shafin Rahman
MemCorr-DP is a diffusion policy that uses counterfactual correspondence conditioning to align a frozen RoMa v2 reference trajectory with the current scene in explicit 3D relations. By assigning opposite behaviors the same physical state and noisy action while retaining reference-specific denoising targets, and then fine‑tuning with mixed‑condition data, the policy learns to adapt from ground‑truth geometry to measured correspondence errors. In a door‑opening task with large spatial and viewpoint shifts, MemCorr‑DP achieves 96.67% closed‑loop success versus 88.00% for a visual Transformer, demonstrating that explicit 3D reference relations provide robust conditioning when spatial and viewpoint changes are compounded.
By Tan Su, Haoxiang Yang, Ruxin Wang, Binghui Xie
The paper introduces visual adaptations of counterfactual tests—vCT and vCCT—to evaluate whether chain-of-thought explanations in vision‑language models faithfully reflect the visual evidence driving predictions. Using these tests, the authors benchmark eight open‑source VLMs on two datasets and find that CoTs often fail to track visual evidence, sometimes omitting removed objects or mentioning them inconsistently. They also release two new datasets, Counter‑SNLI‑VE and Counter‑A‑OKVQA, consisting of image pairs that differ by a single object to facilitate further research.
By Bayar Menzat, Maximilian S\"uss, Ruizhi Wang, Benno Steinegger, Thomas Lukasiewicz, Oana-Maria Camburu
The study introduces the RSM framework, which includes the ARNAI artifact removal network, to enhance automated measurement of spinopelvic parameters on postoperative lumbar spine radiographs containing implants. Adding ARNAI to the Transformer-based FCBFormer model raised the mean Dice similarity coefficient from 0.814 to 0.870 and significantly reduced the mean L4–L5 segmental Cobb angle error from about 15.8° to 4.7°, a 70% improvement. The framework also improved intraclass correlation coefficients for key parameters, surpassing 0.70 for pelvic tilt, lumbar lordosis, and sacral slope.
By Sang-Jin Park, Jinyoung Choi, Seokwon Kim, Seungeon Song, Insu Park, Dougho Park, Taeyeon Kim, Youjin Lee, Donghoon Yang, Jaeman Cho, Joongwon Yang, Mansu Kim, Heumdai Kwon, Hong Gyu Baek, Dae Chul Cho, Injung Kim
The paper introduces the Flow Moment, a reasoning pattern marked by sustained, process‑confirming verbalizations, contrasting with the revision‑oriented Aha Moment. It proposes Flow‑CoT, a rewritten version of reasoning traces that preserves content while highlighting Flow Markers, and uses it as auxiliary supervision in on‑policy self‑distillation (OPSD). The authors further present Aha‑Flow Distillation (AFD), a dual‑mode extension of OPSD that pairs concise solution‑based supervision (Aha branch) with rewritten Flow‑CoT under a confident reasoning instruction (Flow branch). Experiments on AIME25 and HMMT25 with Qwen3‑8B and Qwen3‑4B models show consistent performance gains, and controlled ablations confirm that the dual‑mode training structure contributes to the improvement.
By Xiaodong Wang, Peixi Peng
The paper introduces Adversarial Scenario Attack (ASA), a query‑based black‑box method that discovers natural transformation vulnerabilities in vision models by exploring background, weather, and material/color edits via a multimodal language model and a text‑guided generative editor. ASA outperforms previous query‑based generative attacks on ImageNet classifiers, achieving higher success rates with fewer queries while maintaining perceptual quality. The approach also shows image‑level and prompt‑level transferability, indicating reusable vulnerabilities across models and images.
By Dongsu Song, DaeYun GO, Jay Hoon Jung
The paper introduces AgentLeak, a black‑box attack that clones the task‑solving capabilities of a strong LLM agent onto a weaker one by exploiting differences between successful and failed executions. Unlike prior skill‑stealing methods that only recover explicit skill artifacts, AgentLeak identifies and incorporates missing procedural behaviors, boosting task pass rates by over 40% and closing more than 80% of the capability gap across 20 scenarios. The study demonstrates that observable execution behavior can leak proprietary procedural knowledge, posing a confidentiality risk for LLM agents.
By Xiaoting Lyu, Yuhong Wu, Yufei Han, Shichang Liu, Liang Zhang, Bin Wang, Bin Wang, Xiaobo Ma, Wei Wang
The paper introduces FACTS, a structured Fisher Approximation for compressing Vision Transformers (ViTs) using Fisher-weighted SVD, which enforces token‑local aggregation while preserving within‑token activation‑gradient dependence. It also presents Constrained Rank Search (CoRS) to optimize layer‑wise rank allocation under a fixed FLOP budget. Experiments on ViTs and hybrid architectures show that FACTS improves accuracy‑efficiency trade‑offs, outperforming the strongest SVD baseline by up to +5.8 percentage points on Swin‑B without requiring finetuning.
By Moritz Thoma, Maximilian Groezinger, Maximilian Forstenh\"ausler, Emad Aghajanzadeh, Ryan Pegoud, Manoj Rohit Vemparala, Pierpaolo Mori, Alexander Frickenstein, Daniel Mueller-Gritschneder, Ulf Schlichtmann
MV-STRIDE is a Multi‑View hierarchical Spatial Reasoning dataset that models dependencies among perception, scene understanding, and contextual reasoning to support 3D spatial cognition. It introduces a QA generation pipeline that enforces cross‑view constraints, producing multi‑level reasoning tasks with chain‑of‑thought supervision. Experiments show that training on MV‑STRIDE yields state‑of‑the‑art performance on multi‑view spatial benchmarks, enabling MLLMs to reason robustly across diverse viewpoints.
By Jin Xu, Xiaojian Huang, Zhuodong Luo, Zhihong Zhang, Xin Liu, Jiansheng Wei, Xinzhi Wang, Jie Zhao, Xuejin Chen
The paper introduces a perturbation-based framework to evaluate quality metrics for large language model (LLM) generated Asset Administration Shells (AAS). By systematically degrading AAS outputs across multiple dimensions, the study identifies that exact property name matching and value-based recall, along with name-based F1 score, best reflect quality changes. Experiments on 6,400 AAS instances from 200 products across GPT‑4o‑mini, Qwen3, and DeepSeek‑R1 reveal how different perturbations affect metrics and highlight variations among model families and product segments.
By Janek Gro{\ss}, Elena Zentgraf, Jens Heidrich
BlueprintAgent (BPA) is a multimodal agent that converts scanned reinforced‑concrete building blueprints into simulation‑ready frame models by treating a large language model (MLLM) as the primary reader and decision maker, while OCR and computer vision provide localized evidence. BPA implements engineering constraints as callable validators that trigger targeted MLLM revisits over specific regions, enabling precise beam‑column support, span count, and 3D continuity checks. In evaluations on 300 real scanned sheets from 20 projects, BPA achieved a macro‑averaged Beam F1 of 0.994, vastly outperforming single‑MLLM zero‑shot (0.301) and fixed‑pipeline (0.820) baselines.
By Zhouyuan Xu, Chen Yang, Linhao Wang, Jiansheng Fan, Chen Wang
The paper proposes a new method for credit scoring research articles that distinguishes between a paper’s original contribution and the prior work it builds upon. It introduces a hierarchical ‘contribution tree’ framework that conserves importance across a document’s structure and separates original from citation-derived credit. Large language models are employed as noisy comparative estimators to scale the analysis, and the approach is extended to collections of articles via weighted citation graphs to produce corpus-level contributions and normalized influence scores.
By Sana Ebrahimi, Suraj Shetiya, Abolfazl Asudeh
The study shows that instruction‑tuned language models, when asked to generate responses with varying emotional intensity, consistently undershoot the requested affect. By conditioning a model on continuous Valence‑Arousal targets and measuring output with a frozen regressor, the authors find that the gain for valence is only 0.26 and for arousal 0.13 on Llama‑3.1‑8B, far below the ideal value of 1. They trace this undershoot to the preference‑learning pipeline: training data such as EmoBank are neutral‑heavy and the candidate pool rarely contains extreme affect, so Direct Preference Optimization lacks examples to learn from. Expanding the target space uniformly and sampling a hotter, larger candidate pool raises valence gain to 0.40 and improves extrapolation with minimal in‑distribution cost, a result that also holds for Qwen3‑8B. Arousal remains more variable because the base model rarely generates highly aroused candidates.
By Hyunwoo Kim, Usama Khalid
The paper introduces Feedback‑Enriched Environments (FEEs) as a new approach to training large language models as autonomous agents for long‑horizon tasks. By shifting from action guidance to observation enrichment during later stages of exploration, FEEs improve performance across SciWorld and BFCL benchmarks with various Qwen3 model scales and RL algorithms. The study shows that FEEs stabilize training, promote proactive exploration, embed environmental guidance into policy weights, and highlight intra‑group feedback consistency as key for stable optimization.
By Hongbang Yuan, Zhuoran Jin, Yixin Cao
CausalChapter is a framework designed to improve chaptering of long instructional videos by using interventional dependency modeling. It introduces a Local Dependency Shift module that identifies boundary drops between adjacent temporal windows and a Cross-Segment Support Selection module that re‑ranks historical contexts to enhance chapter description generation. Experiments on long‑video chaptering benchmarks demonstrate that CausalChapter improves boundary localization, chapter description quality, and cross‑chapter coherence.
By Xinran Duan, Guozhang Li, Yaoyao Zhong, Mei Wang, Lizhi Wang, Hua Huang
The paper introduces AHLERT, a system that automatically extracts environment-aware hunt leads from Cyber Threat Intelligence reports. It combines a hybrid retriever—dense vector search plus multi-hop knowledge‑graph traversal seeded with MITRE ATT&CK—with ontology‑grounded retrieval‑augmented generation to constrain leads to a defender’s assets. Evaluations on public CTI reports show that AHLERT doubles mean F1 scores and achieves an effectiveness score of ~86.95% compared to off‑the‑shelf LLM models.
By Akash Prakash, Boubakr Nour, Makan Pourzandi, Chadi Assi, Mourad Debbabi
The paper introduces NPRec, a model‑agnostic framework that uses counterfactual reasoning to neutralize popularity bias in large language model–based recommender systems. By generating debiased textual guidelines that separate intrinsic user interests from popularity signals, NPRec injects these guidelines at inference time to guide the LLM’s generation without updating parameters. Experiments on three real‑world datasets show improved recommendation accuracy, explanation quality, and debiasing performance.
By Guanrong Li, Haolin Yang, Xinyu Liu, Zhen Wu, Rui Xia, Xinyu Dai
arXiv:2609.06289v1 Announce Type: cross
Abstract: As large language models (LLMs) are increasingly deployed in alignment-sensitive contexts, activation steering has emerged as a lightweight, inferenc...
By Mohammad Mahdi Abootorabi, Armin Saghafian, Ali Bazshoushtari, Hamid Rezaei, EunJeong Hwang, Vered Shwartz, Parvin Mousavi, Purang Abolmaesumi
arXiv:2609.06647v1 Announce Type: new
Abstract: Although professional workflows leverage large language models widely, the interpretation for auditing unconstrained free-text generation is usually in...
By Elvin Babayev, Artem Sinitsa, Arash Hajisharifi, Kabir Bakhshaei
arXiv:2609.05806v1 Announce Type: new
Abstract: Emotion Recognition in Conversations (ERC) aims to identify speakers' emotions in multi-turn dialogue. Accurate emotion recognition can support a wide...
By Amir Ben Khalifa, Fanny Bezancon, Amine Trabelsi, Bessam Abdulrazak