Peer Revenue Returns subject to peer review

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

All of arXiv

Sunday 11 October

Sound

BridgeCodec: Mamba Enhanced Neural Audio Codec with Schrödinger Bridge at Low Bitrate

Zijian Lin, Jing Yang, Jinghao Luo et al. (+3 more)

Neural audio codecs achieve remarkable performance but operate as closed ecosystems with rigidly co-trained encoderdecoder pairs, severely hindering interoperability. To resolve this, we propose BridgeCodec, a versatile framework that decouples these pairs by translating latent representations between mismatched, frozen endpoints. We formulate this cross-codec translation as a Schrödinger Bridge optimal transport problem, enabling robust mapping between disparate distributions without predefined noise priors. To efficiently capture longrange speech temporal dependencies, we integrate a Mambaenhanced U-Net architecture. We validate BridgeCodec on an extreme deployment scenario: mapping a lightweight 8 kHz source encoder directly to a 48 kHz target decoder at an ultralow bitrate of 1 kbps. Experimental results demonstrate that BridgeCodec achieves superior wideband reconstruction and high perceptual quality, successfully bridging heavily bottlenecked, heterogeneous codecs.
Read the full abstract

All authors Zijian Lin, Jing Yang, Jinghao Luo, Zhuo Wang, Fan Fan, Zhiyong Wu

Computer Vision

EmoTrace-Bench: Diagnosing Emotional Expression in Text-to-Video Generation through Controlled Cues and Intensity

Yeheng Wang, Liudan Shi, Yi Zheng et al. (+5 more)

Text-to-video (T2V) models are increasingly used to generate videos with targeted emotional expressions, yet existing evaluation methods mainly rely on holistic video-text alignment and provide limited insight into whether failures arise from target-emotion expression, intensity control, or specific affective cues. We introduce EmoTrace-Bench, a fine-grained diagnostic benchmark that controls emotion intensity and selected affective cues in prompts and evaluates generated videos at multiple levels. EmoTrace-Bench contains 448 controlled prompts covering eight emotions, two intensity levels, and diverse cue conditions. Using 12 T2V models, we generate 5,376 videos and evaluate target-emotion presence, perceived intensity, and ten audiovisual dimensions of emotional expression with an automated evaluator calibrated and validated against human annotations on a subset of the data. Results reveal substantial differences across models and prompt designs. Prompts that explicitly describe a character's emotional reaction are followed more reliably than prompts that describe only events or scenes, whereas adding more affective cues or using more explicit cinematic audiovisual language does not consistently improve emotional generation. These results illustrate how aggregate alignment scores may overlook localized differences in target-emotion expression, intensity control, and cue realization. EmoTrace-Bench provides a traceable and interpretable framework for diagnosing emotional video generation without modifying existing generation pipelines, and we release the prompts and the automated evaluation rules.
Read the full abstract

All authors Yeheng Wang, Liudan Shi, Yi Zheng, Jingke Li, Shiyuan Liu, Xin Tao, Yuanxing Zhang, Guoyu Sun

High Energy Physics (Experiment)

Measurement of the cross sections for $e^{+}e^{-}\to p\bar{n}\pi^{-}$ at center of mass energies from 3.510 to 4.946 GeV

BESIII Collaboration: M. Ablikim, M. N. Achasov, P. Adlarson et al. (+784 more)

Using a data sample corresponding to an integrated luminosity of 43.8~fb$^{-1}$ collected with the \mbox{BESIII} detector at 44 center-of-mass energy points between 3.510 and 4.946~GeV, the Born cross sections for the process $e^+e^- \to p \bar{n} \pi^-$ and its charge-conjugated process are measured. The energy dependence of the dressed cross sections is investigated to search for possible contributions from vector charmonium and charmonium-like states. No evidence for contributions from the $\psi(3770)$, $Y(4230)$, $Y(4360)$, $\psi(4415)$, $Y(4500)$, or $Y(4710)$ resonances is observed. The upper limits at the 90% confidence level are determined for the products of the electronic partial widths and the branching fractions for the decays of these resonances into the $p \bar{n} \pi^- + \mathrm{c.c.}$ final state.
Read the full abstract

All authors BESIII Collaboration: M. Ablikim, M. N. Achasov, P. Adlarson, X. C. Ai, C. S. Akondi, R. Aliberti, A. Amoroso, Q. An, Y. H. An, M. S. Anderson, Y. Bai, O. Bakina, H. R. Bao, X. L. Bao, M. Barbagiovanni, V. Batozskaya, K. Begzsuren, N. Berger, M. Berlowski, M. B. Bertani, D. Bettoni, F. Bianchi, E. Bianco, A. Bortone, I. Boyko, R. A. Briere, A. Brueggemann, D. Cabiati, H. Cai, M. H. Cai, X. Cai, A. Calcaterra, G. F. Cao, N. Cao, S. A. Cetin, X. Y. Chai, J. F. Chang, T. T. Chang, G. R. Che, Y. Z. Che, C. H. Chen, Chao Chen, G. Chen, H. S. Chen, H. Y. Chen, M. L. Chen, S. J. Chen, S. M. Chen, T. Chen, W. Chen, X. R. Chen, X. T. Chen, X. Y. Chen, Y. B. Chen, Y. Q. Chen, Z. K. Chen, J. Cheng, L. N. Cheng, S. K. Choi, X. Chu, G. Cibinetto, F. Cossio, J. Cottee-Meldrum, H. L. Dai, J. P. Dai, X. C. Dai, A. Dbeyssi, R. E. de Boer, D. Dedovich, C. Q. Deng, Z. Y. Deng, A. Denig, I. Denisenko, M. Destefanis, F. De Mori, E. Di Fiore, X. X. Ding, Y. Ding, Y. X. Ding, J. Dong, L. Y. Dong, M. Y. Dong, X. Dong, Z. J. Dong, M. C. Du, S. X. Du, Shaoxu Du, X. L. Du, Y. Q. Du, Y. Y. Duan, Z. H. Duan, P. Egorov, G. F. Fan, J. J. Fan, K. X. Fan, Y. H. Fan, J. Fang, Jin Fang, S. S. Fang, W. X. Fang, Y. Q. Fang, L. Fava, F. Feldbauer, G. Felici, C. Q. Feng, J. H. Feng, Q. X. Feng, Y. T. Feng, M. Fritsch, C. D. Fu, J. L. Fu, Y. W. Fu, H. Gao, Xu Gao, Y. Gao, Y. N. Gao, Y. Y. Gao, Yunong Gao, Z. Gao, S. Garbolino, I. Garzia, L. Ge, P. T. Ge, Z. W. Ge, C. Geng, A. Gilman, K. Goetzen, J. Gollub, J. B. Gong, J. D. Gong, L. Gong, W. X. Gong, W. Gradl, M. Greco, M. D. Gu, M. H. Gu, C. Y. Guan, A. Q. Guo, H. Guo, J. N. Guo, L. B. Guo, M. J. Guo, R. P. Guo, X. Guo, Y. P. Guo, Z. Guo, A. Guskov, J. Gutierrez, J. Y. Han, T. T. Han, X. Han, F. Hanisch, K. D. Hao, X. Q. Hao, F. A. Harris, C. Z. He, K. K. He, K. L. He, F. H. Heinsius, C. H. Heinz, Y. K. Heng, C. Herold, P. C. Hong, G. Y. Hou, X. T. Hou, Y. R. Hou, Z. L. Hou, H. M. Hu, J. F. Hu, Q. P. Hu, S. L. Hu, T. Hu, Y. Hu, Y. X. Hu, Z. M. Hu, G. S. Huang, K. X. Huang, L. Q. Huang, P. Huang, X. T. Huang, Y. P. Huang, Y. S. Huang, T. Hussain, N. Hüsken, N. in der Wiesche, J. Jackson, Q. Ji, Q. P. Ji, W. Ji, X. B. Ji, X. L. Ji, Y. Y. Ji, L. K. Jia, X. Q. Jia, D. Jiang, S. J. Jiang, X. S. Jiang, Y. Jiang, J. B. Jiao, J. K. Jiao, Z. Jiao, L. C. L. Jin, S. Jin, Y. Jin, M. Q. Jing, X. M. Jing, T. Johansson, S. Kabana, X. L. Kang, X. S. Kang, B. C. Ke, V. Khachatryan, A. Khoukaz, O. B. Kolcu, B. Kopf, L. Kröger, L. Krümmel, Y. Y. Kuang, M. Kuessner, X. Kui, N. Kumar, A. Kupsc, W. Kühn, Q. Lan, W. N. Lan, T. T. Lei, M. Lellmann, T. Lenz, C. Li, C. H. Li, C. K. Li, Chunkai Li, Cong Li, D. M. Li, F. Li, G. Li, H. B. Li, H. J. Li, H. L. Li, H. N. Li, H. P. Li, Hui Li, J. N. Li, J. S. Li, J. W. Li, K. Li, K. L. Li, L. J. Li, L. K. Li, Lei Li, M. H. Li, M. R. Li, M. T. Li, P. L. Li, P. R. Li, Q. M. Li, Q. X. Li, R. Li, S. Li, S. X. Li, S. Y. Li, Shanshan Li, T. Li, T. Y. Li, W. D. Li, W. G. Li, X. Li, X. H. Li, X. K. Li, X. L. Li, X. Y. Li, X. Z. Li, Y. Li, Y. H. Li, Y. B. Li, Y. C. Li, Y. G. Li, Y. P. Li, Z. H. Li, Z. J. Li, Z. L. Li, Z. X. Li, Z. Y. Li, Zaiyi Li, C. Liang, H. Liang, Y. F. Liang, Y. T. Liang, Z. Z. Liang, G. R. Liao, L. B. Liao, M. H. Liao, Y. P. Liao, J. Libby, A. Limphirat, C. C. Lin, C. X. Lin, D. X. Lin, T. Lin, B. J. Liu, B. X. Liu, C. Liu, C. X. Liu, F. Liu, F. H. Liu, Feng Liu, G. M. Liu, H. Liu, H. B. Liu, H. M. Liu, Huihui Liu, J. B. Liu, J. J. Liu, K. Liu, K. Y. Liu, Ke Liu, Kun Liu, L. Liu, L. C. Liu, Lu Liu, M. H. Liu, P. L. Liu, Q. Liu, S. B. Liu, T. Liu, W. M. Liu, W. T. Liu, X. Liu, X. K. Liu, X. L. Liu, X. P. Liu, X. T. Liu, X. Y. Liu, Y. Liu, Y. B. Liu, Yi Liu, Z. A. Liu, Z. D. Liu, Z. Q. Liu, Z. X. Liu, Z. Y. Liu, Ziqiang Liu, X. C. Lou, H. J. Lu, J. G. Lu, X. L. Lu, Y. Lu, Y. H. Lu, Y. P. Lu, Z. H. Lu, C. L. Luo, J. R. Luo, J. S. Luo, M. X. Luo, T. Luo, X. L. Luo, Z. Y. Lv, X. R. Lyu, Y. F. Lyu, Y. H. Lyu, C. L. Ma, F. C. Ma, H. L. Ma, Heng Ma, J. L. Ma, L. L. Ma, L. R. Ma, Q. M. Ma, R. Q. Ma, R. Y. Ma, T. Ma, X. T. Ma, X. Y. Ma, F. E. Maas, I. MacKay, M. Maggiora, S. Maity, S. Malde, Q. A. Malik, L. M. Mansur, Y. J. Mao, Z. P. Mao, S. Marcello, A. Marshall, F. M. Melendi, Y. H. Meng, Z. X. Meng, G. Mezzadri, H. Miao, T. J. Min, T. Mineeva, R. E. Mitchell, X. H. Mo, A. F. Mohammad, B. Moses, N. Yu. Muchnoi, J. Muskalla, Y. Nefedov, F. Nerling, H. Neuwirth, Z. Ning, S. Nisar, Q. L. Niu, W. D. Niu, Y. Niu, C. Normand, S. L. Olsen, Q. Ouyang, I. V. Ovtin, S. Pacetti, Y. Pan, C. Y. Pang, A. Pathak, Y. P. Pei, M. Pelizaeus, G. L. Peng, H. P. Peng, X. J. Peng, Y. Y. Peng, K. Peters, K. Petridis, J. L. Ping, R. G. Ping, S. Plura, V. Prasad, L. Pöpping, F. Z. Qi, H. R. Qi, L. Y. Qian, S. Qian, W. B. Qian, C. F. Qiao, J. H. Qiao, J. J. Qin, J. L. Qin, L. Q. Qin, L. Y. Qin, P. B. Qin, X. P. Qin, X. S. Qin, Z. H. Qin, J. F. Qiu, Z. H. Qu, J. Rademacker, K. Ravindran, C. F. Redmer, A. Rivetti, M. Rolo, G. Rong, S. S. Rong, F. Rosini, Ch. Rosner, M. Q. Ruan, W. R. Ruangyoo, N. Salone, A. Sarantsev, Y. Schelhaas, M. Schernau, K. Schoenning, M. Scodeggio, W. Shan, X. Y. Shan, Z. J. Shang, J. F. Shangguan, L. G. Shao, M. Shao, C. P. Shen, H. F. Shen, W. H. Shen, X. Y. Shen, B. A. Shi, Ch. Y. Shi, H. Shi, J. L. Shi, J. Y. Shi, M. H. Shi, S. Shi, S. Y. Shi, X. Shi, H. L. Song, J. J. Song, M. H. Song, T. Z. Song, W. M. Song, Y. X. Song, Zirong Song, S. Sosio, S. Spataro, S. Stansilaus, F. Stieler, M. Stolte, S. S Su, G. B. Sun, G. X. Sun, H. Sun, H. K. Sun, J. F. Sun, K. Sun, L. Sun, R. Sun, S. S. Sun, T. Sun, W. Y. Sun, Y. C. Sun, Y. H. Sun, Y. J. Sun, Y. Z. Sun, Z. Q. Sun, Z. T. Sun, H. Tabaharizato, N. T. Tagsinsit, C. J. Tang, G. Y. Tang, J. Tang, J. J. Tang, L. F. Tang, Y. A. Tang, Z. H. Tang, L. Y. Tao, M. Tat, J. X. Teng, J. Y. Tian, W. H. Tian, Y. Tian, Z. F. Tian, K. Yu. Todyshev, I. Uman, E. van der Smagt, B. Wang, Bin Wang, Bo Wang, B. Q. Wang, C. Wang, Chao Wang, Cong Wang, D. Y. Wang, F. K. Wang, H. J. Wang, H. R. Wang, J. Wang, J. H. Wang, J. J. Wang, J. P. Wang, K. Wang, L. L. Wang, L. W. Wang, M. Wang, Mi Wang, N. Y. Wang, P. Wang, S. Wang, Shun Wang, T. Wang, W. Wang, W. P. Wang, X. F. Wang, X. L. Wang, X. N. Wang, Xin Wang, Y. Wang, Y. D. Wang, Y. F. Wang, Y. H. Wang, Y. J. Wang, Y. L. Wang, Y. N. Wang, Yanning Wang, Yaqian Wang, Yi Wang, Yuan Wang, Z. Wang, Z. L. Wang, Z. Q. Wang, Z. Y. Wang, Zhi Wang, Ziyi Wang, D. Wei, D. H. Wei, D. J. Wei, H. R. Wei, F. Weidner, H. R. Wen, S. P. Wen, U. Wiedner, G. Wilkinson, J. F. Wu, L. H. Wu, L. J. Wu, Lianjie Wu, S. G. Wu, S. M. Wu, X. W. Wu, Y. C. Wu, Y. J. Wu, Z. Wu, H. L. Xia, L. Xia, B. H. Xiang, D. Xiao, G. Y. Xiao, H. Xiao, Y. L. Xiao, Z. J. Xiao, C. Xie, K. J. Xie, Y. Xie, Y. G. Xie, Y. H. Xie, Z. P. Xie, T. Y. Xing, D. B. Xiong, G. F. Xu, H. Y. Xu, Q. J. Xu, Q. N. Xu, T. D. Xu, X. P. Xu, Y. Xu, Y. C. Xu, Z. S. Xu, F. Yan, L. Yan, W. B. Yan, W. C. Yan, W. H. Yan, W. P. Yan, X. Q. Yan, Y. Y. Yan, H. J. Yang, H. L. Yang, H. X. Yang, J. H. Yang, L. Y. Yang, R. J. Yang, X. Y. Yang, Y. Yang, Y. G. Yang, Y. H. Yang, Y. M. Yang, Y. Q. Yang, Y. Z. Yang, Youhua Yang, Z. Y. Yang, W. J. Yao, Z. P. Yao, M. Ye, M. H. Ye, Z. J. Ye, K. Yi, Junhao Yin, Qiqin Yin, Z. Y. You, B. X. Yu, C. X. Yu, G. Yu, J. S. Yu, L. W. Yu, T. Yu, X. D. Yu, Y. C. Yu, Yongchao Yu, C. Z. Yuan, H. Yuan, J. Yuan, Jie Yuan, L. Yuan, M. K. Yuan, S. H. Yuan, Y. Yuan, C. X. Yue, Ying Yue, A. A. Zafar, F. R. Zeng, S. H. Zeng, X. Zeng, Y. J. Zeng, Yujie Zeng, Y. C. Zhai, Y. H. Zhan, B. L. Zhang, B. X. Zhang, D. H. Zhang, G. Y. Zhang, Gengyuan Zhang, H. Zhang, H. C. Zhang, H. H. Zhang, H. L. Zhang, H. Q. Zhang, H. R. Zhang, H. Y. Zhang, Han Zhang, J. Zhang, J. J. Zhang, J. L. Zhang, J. Q. Zhang, J. S. Zhang, J. W. Zhang, J. X. Zhang, J. Y. Zhang, J. Z. Zhang, Jianyu Zhang, Jin Zhang, Jiyuan Zhang, L. M. Zhang, Lei Zhang, N. Zhang, P. Zhang, Q. Zhang, Q. Y. Zhang, Q. Z. Zhang, R. Y. Zhang, S. H. Zhang, S. N. Zhang, Shulei Zhang, X. M. Zhang, X. Y. Zhang, Y. T. Zhang, Y. H. Zhang, Y. P. Zhang, Yao Zhang, Yu Zhang, Yu Zhang, Z. Zhang, Z. D. Zhang, Z. H. Zhang, Z. L. Zhang, Z. X. Zhang, Z. Y. Zhang, Z. Z. Zhang, Zh. Zh. Zhang, Zhaoke Zhang, Zhilong Zhang, Ziyang Zhang, Ziyu Zhang, G. Zhao, J.-P. Zhao, J. Y. Zhao, J. Z. Zhao, L. Zhao, Lei Zhao, M. G. Zhao, R. P. Zhao, S. J. Zhao, Y. B. Zhao, Y. L. Zhao, Y. P. Zhao, Y. X. Zhao, Z. G. Zhao, A. Zhemchugov, B. Zheng, B. M. Zheng, J. P. Zheng, W. J. Zheng, W. Q. Zheng, X. R. Zheng, Y. H. Zheng, B. Zhong, C. Zhong, X. Zhong, H. Zhou, J. Q. Zhou, S. Zhou, X. Zhou, X. K. Zhou, X. R. Zhou, X. Y. Zhou, Y. X. Zhou, Y. Z. Zhou, A. N. Zhu, J. Zhu, K. Zhu, K. J. Zhu, K. S. Zhu, L. X. Zhu, Lin Zhu, S. H. Zhu, T. J. Zhu, W. D. Zhu, W. J. Zhu, W. Z. Zhu, Y. C. Zhu, Z. A. Zhu, X. Y. Zhuang, M. Zhuge, J. H. Zou, J. Zu

Neural and Evolutionary Computing

CoFlu-Brain: An fMRI Connectivity Foundation Model with Cofluctuation-Based Pretraining

Ziteng Sui, Junfeng Xia, Wenhao Ye et al. (+2 more)

Functional magnetic resonance imaging (fMRI) commonly characterizes large-scale brain organization through functional connectivity (FC). However, an observed FC matrix is not a fixed brain graph but a statistical estimate determined by the finite BOLD frames used to compute it. Conventional FC-based approaches typically represent each scan with a single Pearson-correlation matrix, leaving frame-resolved variations in regional interactions unexploited. To address this limitation, we introduce CoFlu-Brain, which organizes framewise cofluctuation observations into multiple retained and dropped views, each summarized as an FC matrix. The model learns connectivity structure reproducible across retained views and predicts the representations of the corresponding dropped views, while adopting SIGReg from LeJEPA to prevent representation collapse. We pretrained CoFlu-Brain on 13,444 subjects from eight dataset groups and evaluated its representations via frozen-encoder linear probing on seven classification and two age-prediction tasks across seven datasets. CoFlu-Brain showed strong transfer across demographic, developmental, and clinical prediction tasks, outperforming the evaluated released models on most tasks. Random frame partitioning improved aggregate transfer, while the complete objective achieved the strongest overall performance. Code is available.
Read the full abstract

All authors Ziteng Sui, Junfeng Xia, Wenhao Ye, Mo Wang, Quanying Liu

Artificial Intelligence

Hidden in Plain Sight: Advancing Harmful Content Understanding via Perception-Aware Adversarial Self-Play

Yisong Xiao, Benlei Cui, Haiwen Hong et al. (+8 more)

Despite significant advances in language understanding, Large Language Models (LLMs) can still miss meaning hidden in plain sight, failing to identify harmful intent in expressions readily understood by humans. Malicious users exploit this gap by strategically disguising harmful content to evade automated moderation, posing a serious threat to online platform safety. The vulnerability reflects a fundamental cognitive asymmetry between human readers, who integrate contextual, phonetic, and visual cues to decipher the intent behind deliberate obfuscation, and text-based LLMs, which rely primarily on token sequences and have limited exposure to evolving evasions. To bridge this asymmetry, we propose PASP (Perception-Aware Adversarial Self-Play), which combines multi-perceptual views with adversarial self-play to improve understanding of harmful intent as evasions evolve. Specifically, PASP constructs textual, phonetic, and visual views of each input, preserving its original context while exposing pronunciation and glyph cues obscured in token space. During adversarial self-play, an attacker uses defender feedback to generate intent-preserving evasions, while the defender is trained using an intent-invariance reward to understand harmful intent across changing surface forms. Extensive experiments show that PASP reduces adaptive attack success rates by up to 73.71% and 74.49% on ChineseHarm-Bench and EvoHarmBench, respectively, preserves model capabilities with minimal degradation, and effectively interprets real-world evasive expressions in e-commerce.
Read the full abstract

All authors Yisong Xiao, Benlei Cui, Haiwen Hong, Shiji Zhao, Zonghao Ying, Yaoyuan Zhang, Longtao Huang, Aishan Liu, Hui Xue, Ting Ma, Xianglong Liu

Quantum Algebra

The First Modular Zhu Algebra of $L(\frac12,0)_F$ in Characteristic Seven

Haihua Huang

Let $F$ be an algebraically closed field of characteristic $7$, and let $V=L(\frac12,0)_F$. For the degree-four singular vector $q=(L(-2)^2-2L(-4))\mathbf1$ in the universal vacuum vertex operator algebra $\bar V$, set $V_q=\bar V/I(q)$. We prove that the natural quotient map induces an isomorphism $A_1(V_q)\cong A_1(V)\cong F[T]/(T(T-4)(T-5))\cong F^3$. Explicit $O_1(V_q)$-relations and four families of triangular PBW straightening relations give the upper bound, while three one-dimensional homogeneous subspaces give the matching lower bound.
Read the full abstract

Combinatorics

Turán Numbers of Linear Stars and Expanded Trees

Xinmin Hou and Anyao Wang

For an $r$-uniform hypergraph $\mathcal H$, let $\ex_r(n,\mathcal H)$ denote the maximum number of edges in an $n$-vertex $r$-uniform hypergraph containing no copy of $\mathcal H$. The $r$-uniform expansion $T^{(r)}$ of a tree $T$ is obtained by adding $r-2$ new vertices to each edge, with the added sets pairwise disjoint and disjoint from $V(T)$. In particular, the expansion of $K_{1,k}$ is the linear star $\mathcal L_k^r$, whose $k$ edges intersect pairwise in a single common vertex. Chung and Frankl (1987) determined $\ex_3(n,\mathcal L_k^3)$ and all extremal hypergraphs for every $k\ge3$ and sufficiently large $n$. We extend the odd-$k$ part of their result to every $r\ge5$, determining the exact Turán number and the unique extremal hypergraph up to isomorphism for sufficiently large $n$. More generally, we extend this result to expansions of trees for which the minimum degree among the vertices in the smaller color class is odd and at least three. For every $r\ge5$ and sufficiently large $n$, we establish lower and upper bounds on their Turán numbers. When no two vertices attaining this minimum degree in that color class have a common neighbor, we determine the exact Turán number and the unique extremal hypergraph up to isomorphism. The proofs combine the delta-system method with stability arguments and a detailed analysis of extremal constructions.
Read the full abstract

Signal Processing

A Multi-View Low-Latency Direction-Sensing for Low-Altitude Network via a Multi-Armed Array

Jinbing Jiang, Feng Shu and Jiangzhou Wang

Due to the directional antenna pattern, the received signal power decreases when the target deviates from the array boresight, degrading direction-sensing accuracy. Although rotatable arrays can alleviate this problem, repeated array rotations will induce an incremental long sensing latency. To break the dilemma of harvesting the directive gain from element and making a low-latency sensing, a multi-armed array comprising multiple subarrays with different orientations is proposed to provide wide spatial coverage encompassing low-altitude and terrestrial regions while enabling multi-view direction sensing. However, the different views lead to different direction-sensing accuracies. To fully exploit the complementary information from multiple views, two fusion methods, called SNR-weighted fusion and logistic-weighted fusion, are developed. The former directly assigns fusion weights according to the estimated received SNRs, whereas the latter employs an exponential SNR mapping to further emphasize high-SNR subarrays. Furthermore, the Cramer-Rao lower bound (CRLB) is derived to provide a theoretical benchmark for the proposed multi-view direction-sensing system. Simulation results show that logistic-weighted fusion achieves better accuracy in the very low-SNR regime, while both proposed methods approach the CRLB at medium and high SNRs and maintain stable sensing performance.
Read the full abstract

Computer Vision

Rethinking Supervision in Document Unwarping: A Self-consistent Flow-free Approach

Shaokai Liu, Hao Feng and Wengang Zhou

In recent years, the proliferation of smartphones has led to an upsurge in the digitization of document files via these portable devices. However, images captured by smartphones often suffer from distortions, thereby negatively affecting digital preservation and downstream applications. To address this issue, we introduce DRNet, a novel deep network for document image rectification. Our approach is based on three key designs. Firstly, we exploit the intrinsic geometric consistency inherent in document images to guide the learning process of distortion rectification. Secondly, we design a coarse-to-fine rectification network to leverage the representations derived from the distorted document image, thereby enhancing the rectification result. Thirdly, we propose a unique perspective for supervising the learning of rectification networks, where undistorted document images are employed for supervision, which is free of warping mesh as ground truth in existing methods. Technically, both low-level pixel alignment and high-level semantic alignment jointly contribute to the learning of the mapping relationship between deformed document images and distortion-free ones. We evaluate our method on the challenging DocUNet Benchmark dataset, where it sets a series of state-of-the-art records, demonstrating the superiority of our approach compared to existing learning-based solutions. Additionally, we conduct a comprehensive series of ablation experiments to further validate the effectiveness and merits of our method.
Read the full abstract

Commutative Algebra

Gorenstein Prüfer domains with non-Prüfer integral closure

Hwankoo Kim

There exists a one-dimensional local Gorenstein Prüfer domain whose integral closure is not Prüfer. We prove that the recently constructed coherent counterexample to the normalization problem has this stronger Gorenstein property. The argument establishes permanence of divisoriality of finitely generated ideals under ultraproducts and quotients by divided primes. It also supplies uniform matrix factorizations: in an ultraproduct of plane-curve domains, every nonzero ideal generated by $r$ elements is a cycle of a two-periodic totally acyclic complex with free terms of rank $r$. The divided prime defining the one-dimensional example is flat, so these complexes descend through the quotient and localization. A specified two-generated ideal remains noninvertible in every integral overring. Over the normalization, its associated Laurent overring has nonzero first Tor with every injective cogenerator, although both rings are Gorenstein flat over the original domain. Examples distinguish this failure from a nonnormal Gorenstein curve with Prüfer normalization and show that the periodicity conclusion is stronger than the Gorenstein Prüfer property alone.
Read the full abstract

Game Theory

Randomized Algorithm for MMS Allocation of Chores with Online Agent Arrivals

Haolong Li, Fangxiao Wang and Xiaowei Wu

We study the fair allocation of $m$ indivisible chores among $n$ agents in the agent-arrival model against an oblivious adversary. The chore set and the number of agents are known in advance. When agent $i$ arrives, her nonnegative additive cost function $c_i$ is revealed, and the algorithm must immediately and irrevocably assign her a subset of the unallocated chores. We focus on the notion of maximin share (MMS) and measure the fairness of the realized allocation $X$ by the ex-post MMS approximation ratio $\rho(X)$. We measure an algorithm's performance by its expected ratio, and also the guarantees that hold with high probability. For binary additive costs, we achieve an expected ex-post MMS ratio of $3/2+o(1)$, and show that every randomized online algorithm has an expected ratio of at least $3/2-o(1)$. For every $n$, the expected ratio is at most $1.755$. A variant of our algorithm guarantees an approximation ratio of $2$ with high probability, which is optimal. % For general additive costs, we present a weighted-rounding algorithm that achieves an expected ratio of $2+o(1)$, which also holds with high probability. The expected ratio is at most $2.594$ for every $n$. We also show that similar algorithms extend to additive goods, achieving a constant expected ex-post MMS ratio and guaranteeing every agent an $x$-fraction of her MMS with probability at least $1-85x^2$ for every $x\le1/15$.
Read the full abstract

Commutative Algebra

On modular vector invariant fields

Ying Han and Runxuan Zhang

Let $\mathbb{F}_q$ be a finite field of any characteristic and $\rm GL(n,\mathbb{F}_q)$ the general linear group over $\mathbb{F}_q$. Suppose $W$ denotes the standard representation of $\rm GL(n,\mathbb{F}_q)$ and $\rm GL(n,\mathbb{F}_q)$ acts diagonally on the direct sum of $W$ and its dual space $W^*$. Let $G$ be any subgroup of $\rm GL(n,\mathbb{F}_q)$. Suppose the invariant field $\mathbb{F}_q(W)^G=\mathbb{F}_q(f_1,f_2,\dots,f_k)$, where $f_1,f_2,\dots,f_k\in\mathbb{F}_q[W]^G$ are homogeneous invariant polynomials. We prove that there exist homogeneous polynomials $\ell_1,\ell_2,\dots,\ell_{n}$ in the invariant ring $\mathbb{F}_q[W\oplus W^*]^{G}$ such that the invariant field $\mathbb{F}_q(W\oplus W^*)^{G}$ is generated by $\{f_1,f_2,\dots,f_k,\ell_1,\ell_2,\dots,\ell_{n}\}$ over $\mathbb{F}_q$.
Read the full abstract

Statistical Mechanics

Extended-range percolation on lattices with cubic and non-compact neighborhoods in three dimensions

Zhe Zhang, Yuxuan Fan, Yang Yang et al. (+2 more)

Using MC simulations based on an efficient single-cluster growth algorithm, we investigate two types of three-dimensional extended-range percolation models, namely cubic and complex (non-compact) neighborhoods. We obtain high-precision percolation thresholds $p_c$ for approximately 100 distinct structures and further explore the relationship between these thresholds and the topology of the lattices, which includes the total coordination number $z$, the number of sites $z_i$ and these sites' distance to the central site $r_i$ of the given neighborhood in the $i$th coordination zone. Both bond and site percolation are analyzed for cubic neighborhoods on the SC lattice. For bond percolation, the asymptotic value of $zp_c$ tends to Bethe-lattice behavior ($zp_c\sim1$), and finite-$z$ correction is found to be consistent with $zp_{c}-1\sim a_{1}z^{-x}$, where $x=(d-1)/d$. For site percolation, the value of $zp_c$ varies non-monotonically, exhibiting an initial increase followed by a decrease with the coordination number $z$. Linear fitting leads to $zp_c\sim2.5987(1)$ for large $z$, predicting a value of continuum percolation threshold of aligned cubes $\eta_c=0.324838(12)$ using $zp_c=2^{d}\eta_c$. We then examine site percolation for non-compact neighborhoods extending to the fifth coordination zone on SC, FCC, and BCC lattices. Through a comparison of the three indices of $z$, $\xi=\sum_iz_ir_i^2/i$, and $\zeta=\sum_iz_ir_i$, we find that $\zeta$ is effective in removing the degeneracy. The results obtained show power-law dependence $p_c\propto(\zeta/K)^{-\gamma}$ with $\gamma_{\mathrm{SC}}^{\zeta}=0.703(12)$, $\gamma_{\mathrm{FCC}}^{\zeta}=0.728(10)$, and $\gamma_{\mathrm{BCC}}^{\zeta}=0.699(12)$ for SC, FCC, and BCC lattices, and $p_c\propto\zeta^{-\gamma}$ with $\gamma^{\zeta}=0.719(9)$ independent of the underlying lattice. The number $K$ indicates lattice connectivity.
Read the full abstract

All authors Zhe Zhang, Yuxuan Fan, Yang Yang, Yusong Tu, Zhipeng Xun

Software Engineering

Comparing the Pre-trained Models of Source Code by Re-pre-training under a Unified Setup

Changan Niu, Chuanyi Li, Vincent Ng et al. (+1 more)

Recent years have seen the successful application of large pre-trained models of source code (CodePTMs) to code representation learning, which have taken the field of software engineering (SE) from task-specific solutions to task-agnostic generic models. By the remarkable results, CodePTMs are seen as a promising direction in both academia and industry. While a number of CodePTMs have been proposed, they are often not directly comparable because they differ in experimental setups such as pre-training dataset, model size, evaluation tasks and datasets, etc. In this paper, we first review the experimental setup used in previous work and propose a standardized setup to facilitate fair comparisons among CodePTMs to explore the impacts of their pre-training tasks. Then, under the standardized setup, we re-pre-train CodePTMs using the same model architecture, input modalities, and pretraining tasks, etc., as they declared and fine-tune each model on each evaluation SE task for evaluating. Finally, we present the experimental results and make a comprehensive discussion on the relative strength and weakness of different pre-training tasks w.r.t. each SE task. We hope our view can inspire and advance the future study of more powerful CodePTMs.
Read the full abstract

All authors Changan Niu, Chuanyi Li, Vincent Ng, Bin Luo

Human-Computer Interaction

AuthScribe: Supporting Agency and Authorship in LLM-Assisted Reflective Journaling

Lanping Hu, Yinsi Zhou, Zhenchang Xing et al. (+2 more)

Large language models are increasingly used in reflective journaling, where articulating an experience is part of interpreting it. This raises questions about how writers retain authorship, exercise agency, and assess personalized assistance. We present AuthScribe, an LLM-assisted reflective journaling system that combines contextual anchors, writer-steered AI support, Autobiographical Citations, and read-only Provisional Drafts that require re-expression before AI-generated suggestions enter the journal. Retrieval and LLM inference run on-device. In a 14-day field deployment with 14 participants, grounding shifted from situating current experiences to reconnecting them with prior self-authored interpretations. Perceived agency remained stable through Week 1 and shifted during Week 2. Participants described authorship through transforming AI-generated suggestions into personally meaningful expression, and trusted personalized prompts when they could trace them to prior writing. We discuss design implications for adoption friction, longitudinal agency, and traceable personalization in AI systems for personal reflection.
Read the full abstract

All authors Lanping Hu, Yinsi Zhou, Zhenchang Xing, Xiaoyu Sun, Shidong Pan

Data Structures and Algorithms

Optimal High Probability Bounds for Random Order Online Sorting

Chenglin Fan and Mingyu Kang

In random order online sorting, a fixed multiset of $n$ real numbers arrives in a uniformly random order. Each value must be placed immediately and irrevocably into an empty position of an array with exactly $n$ positions. The objective is to minimize the sum of the absolute differences between consecutive entries. If the entire input were known in advance, sorting the values would be optimal and would have cost equal to the range of the input. We prove that the optimal worst case competitive ratio in this model is $\Theta(\log n)$ with high probability. In particular, we give a single online algorithm that achieves an $O(\log n)$ competitive ratio with high probability for every fixed input, and we show that every randomized online algorithm has some fixed input on which an $\Omega(\log n)$ ratio is unavoidable with high probability. The same upper bound also holds for independent samples from an arbitrary unknown distribution on the real line. The algorithm uses the random arrival order to learn the relative positions of future values from the values seen so far. It maintains a hierarchy of small regions of the input order and allows an arriving value to use nearby space when its preferred region is full. This flexibility avoids the need to reserve large amounts of space for unlikely local fluctuations. The analysis shows that large displacements become rapidly less likely as their scale grows, leading to the $O(\log n)$ high probability bound. A matching lower bound shows that no randomized online algorithm can improve the logarithmic dependence in the worst case.
Read the full abstract

Game Theory

Polynomial Algorithms for Simultaneous Share-Based and Envy-Based Guarantees

Hannaneh Akrami, Arash Ashuri and Kurt Mehlhorn

We study the simultaneous achievement of share-based and envy-based fairness in the allocation of indivisible goods with nonnegative additive valuations. The AnyPrice share (APS) is a strengthening of the maximin share (MMS). A polynomial-time algorithm is known to guarantee a $2n/(3n-1)$ approximation of APS, but without an accompanying envy-based guarantee. Separately, allocations simultaneously satisfying minimum EFX share (MXS) and envy-freeness up to one less-preferred good (EFL), and subsequently the stronger requirement of residual maximin share (RMMS) together with EFL, have been shown to exist. However, the known constructions do not provide pseudopolynomial-time guarantees. We develop algorithms that combine these share-based and envy-based fairness requirements. First, for every $\varepsilon \in (0,1)$, we give an algorithm that computes an allocation satisfying exact EFL, $(1-\varepsilon)$-MXS, and $2n/(3n-1)$-APS. Its running time is polynomial in the binary input length and $1/\varepsilon$. Thus, we match the previously established polynomial-time APS approximation while simultaneously guaranteeing exact EFL and an arbitrarily close approximation to MXS. Second, we provide a pseudopolynomial-time algorithm that simultaneously guarantees exact MXS, exact EFL, and $2n/(3n-1)$-APS. Third, for every fixed number of agents, we obtain a pseudopolynomial-time algorithm that simultaneously achieves exact RMMS and exact EFL. Finally, we establish that computing a numerical threshold between MXS and RMMS is weakly NP-hard, even for additive valuations. This identifies a computational barrier to extending our polynomial-time approach to exact MXS through independently computed residual-valid share thresholds.
Read the full abstract

High Energy Physics (Phenomenology)

A toy model for hadronic bonds in exotic hadrons

Hua-Xing Chen

Three hadronic-bond mechanisms were proposed in our previous studies to describe near-threshold exotic hadrons in terms of hadronic molecular configurations: the covalent bond arises from the sharing and exchange of valence light quarks between hadronic clusters, the creation bond originates from the activation of coherent sea $q\bar{q}$ degrees of freedom followed by the formation of effective covalent correlations, and the annihilation bond arises from valence $q\bar{q}$ annihilation and the associated mixing with a nearby compact state. A simple toy model was previously constructed only for the covalent bond, and in the present study we extend this model to incorporate the creation and annihilation bonds within a unified framework. The resulting framework organizes covalent-bond configurations as tending to lie below threshold and creation-bond configurations as tending to appear above threshold, while the annihilation bond can shift a pre-existing molecular configuration either downward or upward through mixing with a nearby compact state. An effective activation energy $E_{\rm{act}}$ is introduced for the coherent sea four-quark component involved in the creation bond, with the $Z_c(3900)$ providing the benchmark value $E_{\rm{act}}\sim17~{\rm{MeV}}$. Intriguingly, this value is numerically close to the reported mass of the $X_{17}$ particle, which we previously interpreted as a low-energy bare four-quark structure. This numerical proximity may therefore reflect common low-energy (four-quark) dynamics, although the microscopic connection remains to be established. For the isoscalar $D^*\bar{D}^*B^*\bar{B}^*$ configuration involving the $Z_c(4020)$ and $Z_b(10650)$ creation-bond subsystems linked by two additional covalent bonds, we estimate a binding energy of approximately $36~{\rm{MeV}}$ below its four-meson threshold and propose searches in the $4\mu$ and $4\mu2\pi$ final states.
Read the full abstract

Software Engineering

PassSum: Leveraging Paths of Abstract Syntax Trees and Self-Supervision for Code Summarization

Changan Niu, Chuanyi Li, Vincent Ng et al. (+3 more)

Code summarization is to provide a high-level comment for a code snippet that typically describes the function and intent of the given code. Recent years have seen the successful application of data-driven code summarization. To improve the performance of the model, numerous approaches use Abstract Syntax Trees (ASTs) to represent the structural information of the code, which is considered by most researchers to be the main factor that distinguishes code from natural language. Then, such data-driven methods are trained on large-scale labeled datasets to obtain a model with strong generalization capabilities that can be applied to new examples. Nevertheless, we argue that state-of-the-art approaches suffer from two key weaknesses: (1) inefficient encoding of ASTs; (2) reliance on a large labeled corpus for model training. As a result, such drawbacks lead to (1) oversized model, slow training, information loss and instability; (2) inability to be applied to programming languages with only a small amount of labeled data. In light of these weaknesses, we propose PassSum, a code summarization approach that addresses the aforementioned weaknesses via (1) a novel input representation which contains an efficient AST encoding method; (2) introducing three pre-training objectives and pre-training our model with a large amount of (easy-to-obtain) unlabeled data under the guidance of self-supervised learning. Experimental results on code summarization for Java, Python and Ruby methods demonstrate the superiority of PassSum to state-of-the-art methods. Further experiments demonstrate that the input representation we use has both temporal and spatial advantages in addition to performance leadership. In addition, pre-training is also shown to make the model more generalizable with less labeled data, and also to speed up the convergence of the model during training.
Read the full abstract

All authors Changan Niu, Chuanyi Li, Vincent Ng, Jidong Ge, Liguo Huang, Bin Luo

Analysis of PDEs

Spectral Gårding Elliptic Equations

Hao Fang and Biao Ma

We study spectral Gårding elliptic equations on compact Kähler manifolds. Smooth solvability is equivalent to the existence of a smooth representative in the admissible domain. The concavity theorem for ideal Gårding polynomials gives a fixed concave formulation to which Székelyhidi's second-order argument applies in every degree $2\le d\le n$, extending Lin's top-degree solvability theorem. Combined with the numerical criteria of Fu–Li–Zhang–Zhang and Chen–Nie–Xu, this yields a Nakai–Moishezon type criterion for solvability on compact Kähler manifolds. A solution converts the local ideal Gårding structure into a global Lorentzian structure. Its mixed derivatives give a strict nonlinear Hodge–Riemann hierarchy from the degree of the equation down to degree two, and the homogenization of the intersection polynomial is Lorentzian in nef class directions. Numerical criteria then produce explicit Hodge–Riemann classes, Lefschetz isomorphisms in bidegree $(1,1)$, and weighted Hodge-index inequalities.
Read the full abstract

Complex Variables

Oka complements in algebraic tori and their logarithmic covers

Yun-Heng Du, Bin Guo, Peng-Chao Wang et al. (+1 more)

Suppose $n\ge2$ and $K\subset\C^n$ is compact and polynomially convex. Write $q(z)=(e^{-iz_1},\ldots,e^{-iz_n})$, $X_K=(\C^*)^n\setminus(K\cap(\C^*)^n)$, and $Y_K=\C^n\setminus q^{-1}(K)$. We prove that $X_K$ and its logarithmic covering $Y_K$ are Oka. More strongly, $Y_K$ admits normalized holomorphic families of entire embeddings through prescribed holomorphic sections. The proof combines polynomial pair Hamiltonian transport in multiplicative coordinates, compatible with the coordinate hyperplanes, with a fixed Fatou–Bieberbach model. Applications include Oka slab crosses over arbitrary nonempty connected planar domains, complements of periodic equal-radius Euclidean balls centered at products of planar lattices, and complements of sufficiently small closed metric balls in products of elliptic curves.
Read the full abstract

All authors Yun-Heng Du, Bin Guo, Peng-Chao Wang, Song-Yan Xie

Representation Theory

A trace Paley-Wiener theorem for $\mathrm{GL}_{n}(F)\backslash \mathrm{GL}_{n}(E)$, with $E/F$ a quadratic extension of non-Archimedean local fields

Juliette Coutens

We extend the classical trace Paley-Wiener theorem for $p$-adic reductive groups, given by Bernstein, Deligne and Kazhdan in 1986, to a relative case. That is when $E/F$ is a quadratic extension of $p$-adic fields, the main result is a scalar Paley-Wiener theorem for relative Bessel distributions on $\mathrm{GL}_n(F)\backslash \mathrm{GL}_n(E)$. These distributions are relative character of the form $$\pi\mapsto I_\pi (f), \ f\in C_c^\infty(\mathrm{GL}_n(E)),$$ as $\pi$ ranges over $\mathrm{GL}_n(F)$-distinguished irreducible tempered representations, and are constructed from a $\mathrm{GL}_n(F)$-invariant functional and a Whittaker functional. Using the local Langlands correspondence, and the base-change from a unitary group, we show that the relative characters can be described as elements of the "generic" Bernstein center of the unitary group $U(n)$.
Read the full abstract

Computer Vision

Harnessing SPAD Noise: Test-Time Gated Attention for Indoor Depth Estimation

Yiquan Gao

Monocular depth estimation (MDE) has emerged as a cost-effective alternative to hardware-dependent 3D sensing modalities, yet predicting precise metric depth remains a persistent challenge due to intrinsic scale ambiguity, edge bleeding, and limited zero-shot generalization in unseen environments. Recent multi-modal approaches incorporate 1D/3D Single-Photon Avalanche Diode (SPAD) data to resolve these ambiguities, but they mainly treat SPAD noise as a harmful artifact required for denoising or leave the noise ignored. In this paper, we present an essentially different insight: native SPAD noise is an informative prior that implicitly encodes deterministic environmental characteristics. Driven by this paradigm shift, we introduce a novel, lightweight Test-Time Gated Attention framework by Harnessing SPAD Noise, therefore dubbed HSN. Tailored for indoor scenarios, HSN leverages the synergistic combination of 1D SPAD histograms and Gate Adapters (GAs). Through three complementary optimization phases, it generates channel-wise gated attention vectors to dynamically adjust encoder feature maps at inference time, effectively resolving ambiguous scale and boundary challenges without the heavy computational overhead or dense 3D modalities that other TTA methods suffer from. Extensive experiments on challenging real-world indoor benchmarks show that HSN achieves state-of-the-art metric depth accuracy in in-domain settings, while exhibiting even greater potential under zero-shot generalization to unfamiliar environments. Furthermore, interpretability analysis reveals that the proposed gated attention mechanism adaptively shifts attention toward deeper semantic layers under intensifying SPAD noise. Code will be made publicly available.
Read the full abstract

Quantum Physics

Genuine Information Backflow Requires Noncommuting Initial System States

Abdullah Alsharari and Francesco Buscemi

The Gorini-Kossakowski-Lindblad-Sudarshan (GKLS) equation, a cornerstone of open-system theory, describes Markovian semigroup dynamics in which information about the initial preparation cannot revive. Beyond this regime, information revivals are widely interpreted as backflow from a non-Markovian environment. However, a decrease followed by an increase in information retrievable from the system need not imply that information has returned from the environment [PRX Quantum 6, 020316 (2025)]. It may instead arise from noise correlations unrelated to the system's initial preparation, carried by inert degrees of freedom that never interact with the system and therefore cannot mediate backflow. Such revivals are called extrinsic; genuine backflow is reserved for revivals that cannot be explained this way. The framework developed in that work, however, assumes access to arbitrary initial system states, an assumption often incompatible with physical constraints or limited experimental control. This leaves open which properties of the available preparations are necessary for an information revival to certify genuine backflow. Here we extend the framework to allow the description of restricted families of initial states and show that genuine backflow is possible only if the family contains noncommuting states. If all available states commute, as in a classical stochastic process, every revival can be explained by a suitable inert extension and is therefore extrinsic. Noncommutativity of the available system states is thus necessary for information revivals to provide an operational signature of environmental memory.
Read the full abstract

Machine Learning

World Modeling through Spectral Alignment

Holger Molin, William Peng, Marco Bagatella et al. (+1 more)

How do we tell a world model what information to preserve? We study this question through the lens of spectral representation learning. We introduce SpecWM, a JEPA framework that explicitly specifies desired relationships between observations through a target similarity kernel and trains the encoder to reproduce those relationships in its embedding geometry. Different kernels provide different supervision, ranging from physical state information to a fully self-supervised temporal kernel requiring only the ordering of observations within trajectories. We characterize the global optima of the objective, establish conditions for linear recoverability and invariance to nuisance information, and connect temporal-kernel similarity to the behavioral successor measure under an idealized setting. Empirically, the choice of kernel strongly controls which physical quantities are recoverable from the representation. More importantly, temporal spectral alignment improves CEM planning success over LeWorldModel from 52% to 68% on OGBench Cube, 49% to 77% on OGBench Scene, and 29% to 51% on CALVIN. These results suggest that explicitly specifying representation geometry provides a simple way to steer what latent world models preserve for control.
Read the full abstract

All authors Holger Molin, William Peng, Marco Bagatella, Randall Balestriero

Machine Learning

Pessimistic Latent Task-aware Optimization for Robust Offline Meta-Reinforcement Learning

Chang-Hoon Jeong, Kisung Shin, Hanwool Sul et al. (+1 more)

Context-based offline meta-reinforcement learning (COMRL) has advanced almost entirely through better task representations, which separate task identity from artifacts of the behavior policy so that the latent reflects the task alone. The implicit expectation has been that a clean representation will transfer to out-of-support (parameter-extrapolation) tasks. On such a task, however, the encoder produces a latent outside the training support, where the policy has never been trained. We address this gap with Pessimistic Latent Task-aware Optimization (PLATO), which exposes the policy to off-support latents during training. PLATO assumes a mild manifold hypothesis: task latents lie on a low-dimensional manifold, and out-of-support tasks lie farther from the training centroid. It exploits this geometry by perturbing inferred latents outward from the centroid to synthesize counterfactual tasks. A learned decoder ensemble then rolls out short trajectories at each perturbed latent, and the policy is updated on the imagined transitions under a pessimism weight that down-weights steps on which the ensemble disagrees. We prove that the perturbation leaves the training support by a controllable margin and that the out-of-support generalization gap is bounded by the same perturbation reach, and we give explicit conditions, verified empirically, under which the epistemic disagreement at the perturbed latent stays bounded away from zero. On eight MuJoCo continuous-control benchmarks, PLATO consistently improves out-of-support returns over strong COMRL baselines while remaining competitive on in-distribution tasks. Our code is available at https://github.com/deephoony/plato.
Read the full abstract

All authors Chang-Hoon Jeong, Kisung Shin, Hanwool Sul, Byoung-Tak Zhang

Machine Learning

Riemannian Graph Tokenizer for Structural Knowledge Transfer

Qimin Zhou, Haibo Liu, Yujie Wang et al. (+2 more)

Foundation models are at the forefront of artificial intelligence. A tokenizer, converting the raw input into discrete representations that the model can understand, plays an important role to the success of foundation models. Unlike the text tokenizer that is well studied in large language models, graph tokenizer is still at its early stage, facing the challenges of tackling the non-Euclidean structures and capturing the structural semantics. How to design a graph tokenizer for structural knowledge transfer? To this end, we propose a Riemannian Graph Tokenizer (RGT) that bridges the structural knowledge and quantized representations to support cross-domain structural knowledge transfer. The connection is established by Riemannian geometry. Specifically, we first define the geometric vocabulary (trees, cycles and sequences), which captures fundamental structural patterns and reflects the intrinsic geometry of graph. Second, we construct a Riemannian quantizer with Riemannian Straight-Through Estimator to tokenise graph structures across multiple domains into discrete tokens. To ensure consistency and transferability across diverse geometric spaces, RGT further incorporates a geometry-aligned decoder that projects manifold-specific tokens into a unified tangent space. The theoretical analysis and geometric interpretations are provided to support the effectiveness of our proposed method. Extensive experiments across diverse datasets demonstrate that RGT significantly enhances structural knowledge transferability across graph domains.
Read the full abstract

All authors Qimin Zhou, Haibo Liu, Yujie Wang, Li Sun, Chuan Shi

Data Structures and Algorithms

Solving Restricted Versions of #SAT Faster than $2^m$: Algorithms and Lower Bounds

Sanjay Jain, Junqiang Peng, Frank Stephan et al. (+1 more)

Under the Strong Exponential Time Hypothesis (SETH), for any constant $\varepsilon > 0$, the #SAT problem (counting the number of satisfying assignments of a CNF formula) cannot be solved in $\mathcal{O}^*((2-\varepsilon)^n)$ or $\mathcal{O}^*((2-\varepsilon)^m)$ time, where $n$ is the number of variables and $m$ is the number of clauses. On the other hand, simple exhaustive search and the standard inclusion-exclusion give $\mathcal{O}^*(2^n)$-time and $\mathcal{O}^*(2^m)$-time algorithms, respectively. A natural line of research is to beat these SETH-based barriers for restricted versions, such as #$k$-SAT (where each clause contains at most $k$ literals) or #$d$-occ-SAT (where each variable appears in at most $d$ clauses). It is known that the $2^n$-barrier can be broken when $k$ or $d$ is any constant; yet, whether the $2^m$-barrier can also be broken in these cases remains unknown. We present two simple algorithms that overcome the $2^m$-barrier. First, for #$k$-SAT, we give an $\mathcal{O}^*(2^{m(1-1/O(k))})$-time algorithm, thereby breaking the $2^m$-barrier for any constant $k$. Second, for #$2$-occ-SAT (which also includes the problem of counting the number of edge covers in a graph as a special case), we provide an algorithm with runtime $\mathcal{O}^*(2^{m\omega/3})$, where $\omega < 2.3714$ is the matrix multiplication exponent. Furthermore, we demonstrate that both algorithms generalize to the weighted counting setting while preserving the same runtime bounds. On the hardness side, we present a reduction from #Max-$d$-SAT to #$d$-occ-SAT, showing that for any $d\ge 3$, #$d$-occ-SAT admits no $\mathcal{O}^*((2-\varepsilon)^m)$-time algorithm for any constant $\varepsilon>0$, even on positive formulas, under the hypothesis that #Max-3-SAT on $n$ variables admits no $\mathcal{O}^*((2-\varepsilon)^n)$-time algorithm for any such $\varepsilon$.
Read the full abstract

All authors Sanjay Jain, Junqiang Peng, Frank Stephan, Mingyu Xiao

Robotics

Exploring the Limits of Combinatorial Generalization in Embodied Manipulation

Xupeng Zhang, Yantai Yang, Chang Guo et al. (+2 more)

In embodied manipulation, Vision Language Action (VLA) models can execute atomic skills, but long-horizon manipulation requires recombining these skills without demonstrations for every skill combination. The full instruction route treats skill chaining as long-horizon imitation. In a controlled study, full instruction training on limited skill combination sequences reaches 40.0% ordered success on covered combinations but only 7.7% on recombination, suggesting memorization of sequence templates rather than robust skill recombination. We introduce Hierarchical Subtask Execution (HSE), a diagnostic framework that exposes skill recombination through a shared modular interface over the planner, trigger, and policy, decomposing long-horizon failures into separable modes. We introduce RoboCombine, a benchmark for skill recombination without combined skill training trajectories, with metrics for ordered success and transition conversion across subtask boundaries. RoboCombine reveals that atomic competence alone does not ensure ordered execution. Failures concentrate at transitions where downstream subtasks launch from handoff states from preceding subtasks rather than canonical atomic starts. We identify this mismatch as the start state gap. To address it, we introduce Pose Robust Task Initialization (PRTI), a lightweight atomic task post-training strategy that broadens start state coverage beyond canonical starts. Under oracle timed HSE on RoboCombine, PRTI reaches 18.16% ordered success, compared with 4.21% under random perturbation and 9.47% under a robot-reset oracle, without combined skill training trajectories. These results show that combinatorial generalization requires handoff robust atomic skills that convert first stage progress into reliable downstream execution, not merely decomposition into known subtasks. Code: https://github.com/AutoLab-SAI-SJTU/RoboCombine.
Read the full abstract

All authors Xupeng Zhang, Yantai Yang, Chang Guo, Zhaokai Yin, Zhipeng Zhang

Combinatorics

Every n points in the plane contain (m1 + o(1)) n points with no unit distance

Shisheng Li and Yong Su

Let $f(n)$ be the largest integer such that every set of $n$ points in the plane contains $f(n)$ points no two at distance $1$, and let $m_1$ be the supremum of the upper densities of Lebesgue measurable planar sets with no two points at distance $1$. Larman and Rogers observed that $f(n) \ge m_1 n$. We prove that $f(n)/n \to m_1$; that is, $f(n) = (m_1 + o(1))n$. This is the first half of a conjecture of Dúcz and Varga, and with the bound $m_1 \le 0.247$ of Ambrus, Csiszárik, Matolcsi, Varga and Zsámboki it gives $f(n) \le (0.247 + o(1))n$. The proof builds a random independent subset of the algebraic plane, invariant under algebraic translations and rotations, whose probability of containing the origin is at least the worst finite independence ratio, and turns it into a measurable planar set of at least that upper density. Its only non-classical input is the Haar rigidity theorem for wild character laws from the OpenAI preprint The Euclidean plane is not five-colorable, which has a machine-checked Lean proof. The whole argument has been formalized in Lean 4 on top of it.
Read the full abstract