The paper introduces a finite‑width geometric framework that explains how learned feature geometries are organized, transported, and selectively aligned in deep neural networks. It quantifies incompatibilities among weight‑generated covariance, gates, and backward sensitivities using three families of commutators, and provides exact layerwise identities that decompose these commutators into sources such as downstream transport, adjacent‑layer imbalance, and nonlinear gate‑covariance interactions. The study demonstrates that spectral alignment is a layer‑ and scale‑dependent compatibility phenomenon governed by transport, interaction, cancellation, and possible damping, rather than a universal consequence of training.
By Kaj Nystr\"om
The paper surveys how Large Foundation Models (LFMs) can be integrated into Human‑AI Collaboration (HAI) to enhance problem‑solving and decision‑making. It outlines four key areas—human‑guided model development, collaborative design principles, ethical and governance frameworks, and high‑stakes applications—while emphasizing that effective HAI systems arise from careful, human‑centered design rather than merely stronger models. The survey also identifies open challenges related to safety, fairness, and control, aiming to guide future research toward reliable, trustworthy, and beneficial LFM‑based partnerships.
By Vanshika Vats, Marzia Binta Nizam, Minghao Liu, Ziyuan Wang, Richard Ho, Mohnish Sai Prasad, Vincent Titterton, Sai Venkat Malreddy, Riya Aggarwal, Yanwen Xu, Lei Ding, Jay Mehta, Nathan Grinnell, Li Liu, Sijia Zhong, Devanathan Nallur Gandamani, Xinyi Tang, Rohan Ghosalkar, Celeste Shen, Rachel Shen, Nafisa Hussain, Kesav Ravichandran, James Davis
The paper presents a robust multi‑agent reinforcement learning framework for small unmanned aircraft systems (sUAS) to maintain separation assurance when GPS data is degraded or spoofed. By modeling state observation corruption as a zero‑sum game, the authors derive a closed‑form adversarial perturbation that eliminates iterative inner optimization and can be evaluated in linear time. Integrating this perturbation into a policy‑gradient MARL algorithm yields a counter‑policy that achieves near‑zero collision rates in high‑density simulations even with up to 35% observation corruption, outperforming non‑adversarial baselines.
By Alex Zongo, Filippos Fotiadis, Ufuk Topcu, Peng Wei
The paper introduces ClusterMixer, a token mixer based on hierarchical clustering in hyperbolic space, designed to be transparent and interpretable. It forms the core of a new vision backbone called HCFormer, which incorporates multiple clustering strategies to maintain strong performance. Experiments show HCFormer surpasses existing backbones on tasks such as image classification, object detection, instance segmentation, and semantic segmentation.
By Jianan Wei, Guikun Chen, Zhiyuan Weng, Chunchao Guo, Yujia Wang, Wenguan Wang
arXiv:2608.23175v1 Announce Type: new
Abstract: In light-field (LF) imaging systems, dense spatial sampling from a camera array enables powerful post-capture capabilities such as refocusing and depth...
By Sakshi Goel, Ayush Goyal, K S Venkatesh, Koteswar Rao Jerripothula
The paper introduces BanglaSafe, a benchmark of 879 Bengali prompts that covers 17 culturally grounded harm categories and five prompting conditions. Evaluation of 18 frontier LLMs shows that 53.6% of responses are unsafe or partially unsafe, with 14.7% containing strictly harmful content. The study finds that the writing style within Bengali has a stronger impact on safety than the language switch itself, and that current safety classifiers struggle to reliably evaluate Bengali content.
By Naymul Islam, Nusrat Jahan Lia, Shubhashis Roy Dipta, Sabik Bin Sultan, Abdullah Khan Zehady
The paper introduces a new class of adversarial examples that are markedly different from original inputs yet produce the same model output. It presents algorithms such as NI-FGSM, NI-FGM, and their momentum variants (NMI-FGSM, NMI-FGM) to generate these examples. The authors demonstrate that these adversarial examples are not confined to the vicinity of training data but are spread throughout the sample space.
By Xingyang Nie, Caoliang Zhang, Su Pan, Biao Wang, Huilin Ge, Tao Fang
arXiv:2608.23375v1 Announce Type: cross
Abstract: Gumbel-based inference verification bounds LLM weight exfiltration by only forgiving token choices that plausibly arise from honest GPU nondeterminis...
By Nikita Kezins
ONOTE is a unified framework that treats music as a scientifically structured domain of measurable cross-representation correspondences, focusing on omnimodal notation processing centered on sheet music. It introduces a test-only benchmark drawing from diverse musical sources—including staff, Jianpu, and tablature—across varied genres, instruments, and structural conditions, with aligned multimodal derivatives. The framework supports four complementary tasks—score understanding, notation conversion, audio transcription, and symbolic generation—while constructing a provenance-bearing proposition hypergraph from external music-theory materials for evidence retrieval and deterministic validity checks.
By Menghe Ma, Siqing Wei, Yuecheng Xing, Ziyue Zhu, Zhenghong Lin, Yaheng Wang, Fanhong Meng, Peijun Han, Luu Anh Tuan, Haoran Luo
The paper investigates how medical large language models (LLMs) may exhibit narrative anchoring bias when presented with the same clinical case in different patient voices. Using the NarrativeShield SDoH MedQA dataset, the authors evaluate three Qwen2.5 instruction‑tuned LLMs (1.5B, 3B, 7B) on 300 clinical cases, reporting metrics such as persona‑level accuracy, counterfactual consistency, correct consistency, and narrative sensitivity error. The 7B model achieves the highest accuracy (56.33 %) and correct consistency (40.33 %), yet narrative sensitivity errors remain substantial (31.67 %).
By Ahnaf Atef Choudhury, Ramkrishna Saha
DIAG is a Diagnostic Iterative Alignment and Generation framework designed to improve data efficiency in aligning large language models for mathematical reasoning. It adaptively reshapes the practice distribution by first diagnosing valid preference-pair yield to calibrate exploration and exploitation, then generating targeted practice from the model’s failure traces. The approach is theoretically framed as a teacher‑mediated approximation to KL‑regularized reweighting, and experiments show that DIAG increases preference-pair yield and reasoning performance under the same training budget.
By Guhan Chen, Songtao Tian, Bohan Li, Hejin Wang, YeXin Xie, Zixiong Yu
The paper proposes Artifact-Complementary Expert Fusion (ACEF), a two‑stage framework that enhances AI‑generated image detection by combining two types of reconstruction artifacts—VAE/DDIM and SRGAN—into aligned synthetic negatives. ACEF first builds artifact‑specific experts using LoRA adaptation on a frozen backbone, then fuses their multi‑layer evidence with Layer‑wise Artifact‑Complementary Fusion (LACF) to mitigate conflicts between artifact manifolds. Experiments on 13 benchmarks show that this approach improves generalizability over existing state‑of‑the‑art methods.
By Yiheng Li, Yang Yang, Wenhao Wang, Zichang Tan, Zecheng Lin, Li Gao, Zhen Lei
arXiv:2608.23100v1 Announce Type: cross
Abstract: Robot co-design via bi-level optimization couples within-lifetime controller learning for fitness evaluation with cross-generational morphological ev...
By Junru Song, Yang Yang, Yaqing Xu, Ying Wen, Wei Peng, Guozhen Li, Wei'en Zhou, Wen Yao
arXiv:2608.21450v1 Announce Type: new
Abstract: Knowledge-Based Visual Question Answering (KB-VQA) relies on retrieving external information to answer queries involving long-tail entities. However, e...
By Hangrui Xu, Zhengxian Wu, Yunyao Yu, Zhuohong Chen, Rui Cong, Xiangwen Deng, Zhifang Liu, Peng Jiao, Haoqian Wang
The paper introduces the Communication Map, a method that charts every potential communication channel in a transformer model using only its weights. It generalizes previous coupling metrics into a single coefficient covering all 18 connection classes, revealing that 70‑89% of head pairs are non‑randomly oriented and identifying strong or avoiding couplings. The authors demonstrate the map’s utility by recovering known induction circuits and uncovering a two‑dimensional stream subspace whose removal eliminates induction capabilities across several models.
By Richard Zhe Wang
TEE-X is a TEE‑aware acceleration framework designed to run large vision models, such as Vision Transformers, entirely within Trusted Execution Environments. It introduces a sensitivity‑aware modularization technique and vectorization to overcome memory constraints and latency challenges on edge devices. The framework is validated on OP‑TEE for Arm TrustZone and optimized for the NVIDIA Jetson AGX Xavier, achieving GPU‑level inference latency with minimal accuracy‑latency trade‑offs.
By Kurt M Wilson, Mohaiminul Al Nahian, Abeer Matar A. Almalky, Sadat Shahriyar, Souvik Kundu, Zhishan Guo, Abdullah Al Arafat, Adnan Siraj Rakin
arXiv:2608.22960v1 Announce Type: new
Abstract: Coding agents are increasingly evaluated not only by whether they solve a task, but also by how they execute it. However, existing process-level evalua...
By Jiawei He, Mengyu Shi, Jie jia, Xikai Yang, Dong Sun
arXiv:2608.22367v1 Announce Type: new
Abstract: Diffusion multimodal large language models (dMLLMs) frequently produce long-form outputs marred by semantic drift and repetition, with quality generall...
By Yikai Zhao, Qiyan Zhao, Jiaquan Zhang, Xiaofeng Zhang, Xiaosong Yuan, Pengzhou Cheng
arXiv:2608.23149v1 Announce Type: cross
Abstract: The alignment of Large Language Models heavily relies on English-centric high-quality preference data, which often leads to suboptimal performance in...
By Seungyoon Lee, Minhyuk Kim, Jungseob Lee, Heuiseok Lim
arXiv:2608.21836v1 Announce Type: new
Abstract: Large language models have become increasingly capable agents for low-level code and kernel optimization, but isolated kernel benchmarks provide only a...
By Hui Zeng, Pengfei Yang, Yanxin Chen, Fusong Ju, Xinran Wei