Peer Revenue Fully funded

Papers

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

All of arXiv

Monday 28 September

Machine Learning

Routing Without Embeddings: Fast And Interpretable Routing With Regular Expressions

Yifan Lu, Qiyue Zhang, Haotian Shan et al. (+2 more)

Large Language Model (LLM) routers commonly rely on neural query embeddings, with larger encoders expected to better capture query intent and difficulty. Yet scaling Qwen2.5 encoders from 0.5B to 72B parameters brings little improvement in routing accuracy (Figure 1b), suggesting that small encoders may already capture the query properties needed for routing. We therefore investigate which properties matter and whether they can be extracted directly from text without a neural encoder. We introduce REGEXROUTE, a pipeline that uses sparse autoencoders (SAEs) to discover interpretable regular-expression (regex) features. Using unlabeled text, an LLM turns descriptions of grouped SAE latents into regex extractors and refines them to match latent activation patterns. These extractors supply numerical features to a lightweight routing head, eliminating neural encoding at inference (Figure 1a). Across four benchmarks, one fixed set of 128 features achieves 76.43% average routing accuracy, comparable to 76.41% for the strongest neural text encoder baseline, with much smaller latency and strong robustness. These findings establish explicit, interpretable text features as a practical basis for designing and understanding LLM routers.
Read the full abstract

All authors Yifan Lu, Qiyue Zhang, Haotian Shan, Hanjie Chen, Jiarong Xing

Computer Vision

DORA: Dynamic Online Reinforcement Agent for Token Pruning in Vision Transformers

Kaixuan He, Song Chen and Yi Kang

Vision Transformers (ViTs) incur quadratic self-attention cost in the number of tokens. Most token-reduction methods adapt token identities within a prescribed layer-wise compression schedule, or search a static mask offline, and thus limit online adaptation of when and how much to prune. We propose DORA (Dynamic Online Reinforcement Agent), which learns an input-adaptive pruning policy itself for frozen ViTs. At each eligible block, a hierarchical actor decides whether to prune, how many tokens to remove, and which tokens to remove from each image's evolving representation. Because early deletions change the states observed by later decisions, DORA formulates pruning as a finite-horizon Markov decision process. Complete-prefix shadow evaluations convert final-prediction fidelity into localized per-step credit, while closed-loop accuracy feedback adjusts the fidelity penalty toward a shared accuracy-drop target. A privileged critic and all shadow computations are training-only. Deployment retains the frozen backbone and a lightweight actor that applies hard deletion and packed variable-length FlashAttention, converting token reduction into measured speedups. On ImageNet-1K with DeiT-Base, DORA reduces FLOPs by 38.4% relative to the uncompressed backbone within one percentage point of accuracy loss. Averaged across four ViT-type backbones at matched accuracy, DORA uses 13.2% fewer FLOPs and achieves 32.4% higher throughput than the corresponding per-backbone baseline means. Under zero-shot transfer to ImageNet-A, these gains widen to 20.3% and 45.6%, respectively.
Read the full abstract

General Relativity and Quantum Cosmology

Relic Neutrinos Probing Small Scale Primordial Non-Gaussianity

Suraj Gavhale, Shin'ichi Nojiri, Sergei D. Odintsov et al. (+1 more)

Relic neutrinos retain a record of electromagnetic entropy production after neutrino decoupling, probing primordial modes erased from the primary CMB. We derive the cubic photon number source from acoustic damping and its leading folded sensitive bispectrum window, with neutrino decoupling, thermalization, and acoustic coherence combined in a single response kernel. For excited initial states, this kernel filters the physical folded profile and yields a sharpness dependent transmission condition, directly connecting small scale primordial non-Gaussianity to relic neutrino capture rates.
Read the full abstract

All authors Suraj Gavhale, Shin'ichi Nojiri, Sergei D. Odintsov, Oem Trivedi

High Energy Physics (Theory)

Three-loop anomalous dimensions of leading-twist operators in the Gross-Neveu-Yukawa model

Mrigankamauli Chakraborty and Leonid A. Shumilov

We compute the anomalous dimensions of the leading-twist operators at three-loop level in the $\overline{\text{MS}}$ scheme in the Gross-Neveu-Yukawa model. Calculations are done for a number of low-spin operators using computer algebra systems, and the analytical form is reconstructed by solving a set of Diophantine equations. Calculations are optimized using the emergent supersymmetry of the corresponding generalized theory. For the obtained results, we perform a comprehensive analysis at the critical point, including a comparison with the $1/N$ expansion, a check of the generalized Gribov-Lipatov reciprocity, and an analysis of the behavior next to the rightmost singularity. The latter allows us to extract the values of Regge intercepts for all the leading-twist trajectories in the form of an $\epsilon$ expansion.
Read the full abstract

Artificial Intelligence

Test-Time Scaling via Budgeted Multi-Attribute Verification

Bo Xue, Ji Cheng, Shen-Huan Lyu et al. (+2 more)

Verifying LLM-generated answers under a shared computational budget requires jointly deciding which candidates to inspect and which verification attributes to evaluate. We formulate this problem as multi-attribute good-arm identification under a global budget: each candidate is an arm evaluated along several costly attributes, and the goal is to certify as many candidates as possible whose mean scores exceed the prescribed thresholds on all attributes. We propose BMA-GAI, an algorithm that combines cost-aware arm selection with adaptive sampling of attributes. Every observation serves both to guide adaptive allocation and to support anytime-valid certification, which removes the need for a separate confirmation stage. We establish an asymptotic coverage guarantee for BMA-GAI and derive a matching information-theoretic converse that characterizes the intrinsic complexity of the problem, thereby proving that BMA-GAI is first-order optimal away from critical budget levels. Experiments on synthetic benchmarks and an LLM answer-verification task show that BMA-GAI allocates the verification budget more efficiently and certifies more high-quality candidates than competing methods.
Read the full abstract

All authors Bo Xue, Ji Cheng, Shen-Huan Lyu, Yuanyu Wan, Shuang Qiu

Machine Learning

One Rollout Is All You Get: Fully Test-Time Adaptation for GUI Agents

Ziqiang Wang, Li Gu, Zhixiang Chi et al. (+6 more)

GUI agents are deployed with frozen weights and discard everything they experience on the job. Existing ways to update an agent's weights assume something deployment withholds: ground truth, rollouts beyond the single attempt (retries, samples, practice runs), or a learning phase other than deployment. Because GUI actions can be irreversible, a deployed agent gets one attempt per task occurrence, in arrival order, and every attempt counts. No ground truth is available at any point. We define fully test-time adaptation for GUI agents by these constraints and pair it with a minimal weight-space method, SOLO. Auxiliary models read each episode: a judge selects the episodes it deems successful, and a proposer-verifier pair relabels a failed episode's prefix with the subtask that prefix completed. Admitted episodes enter a short sliding window, and each admission updates a small adapter by top-K self-distillation on the agent's own predictions, provided the window holds a judged success. On recurring task streams built from WebArena, VisualWebArena and MobileWorld, SOLO improves on the frozen agent with both UI-TARS-7B and Qwen3-VL-8B, by three to six points of success rate, and exceeds two in-setting memory methods on the web streams.
Read the full abstract

All authors Ziqiang Wang, Li Gu, Zhixiang Chi, Linlian Jiang, Zihuan Jiang, Linqiang Guo, Siobhan Reid, Zhi Liu, Yang Wang

Computation and Language

Certified Selective Automation of LLM Agent Evaluation

Chengguang Gan, Yunhao Liang, Qinghao Zhang et al. (+1 more)

Evaluating LLM agents still ends with a human reading trajectories, because automatic judges carry no guarantee on how often they are wrong. We ask the operational question: what fraction of agent evaluation can a judge take over, with a certificate that the error rate among auto-decided trajectories stays below a budget alpha? Agent corpora resist the standard answer: many agents attempt the same tasks, so trajectories arrive in correlated clusters, and the i.i.d. certificates of existing selective-judging methods can overstate what is safe: a naive certificate can claim 98% automation while its realized error exceeds the budget in 17.5% of task resamples. We introduce a task-level bootstrap certificate that is valid in every regime we test while matching the naive certificate's coverage; finite-sample cluster-valid alternatives certify nothing at realistic task counts. Under this certificate, a 4B logprob judge trained with SFT and reject-weighted GRPO certifies 0.30-0.59 of evaluation on tool-use and web corpora at alpha=0.1, the only judge, among strongly elicited frontier models, certifying on both headline corpora. Certified coverage is predictable before training from base rate and discrimination alone (leave-one-corpus-out R^2=0.96). Finally, the certificate doubles as a self-training filter: pseudo-labels harvested inside certified regions have contamination bounded by alpha by construction (realized 0.000-0.041 across six harvests), letting a judge enter an unseen domain at in-domain strength with zero target training labels.
Read the full abstract

All authors Chengguang Gan, Yunhao Liang, Qinghao Zhang, Shiwen Ni

Computer Vision

Text-Vision Synergistic Token Caching: A Training-Free Framework for Efficient Vision-Language-Action Inference

Qianer Li, Chengjie Zhang, Jingwen Chen et al. (+3 more)

Vision-Language-Action (VLA) models enable generalizable robotic control but remain computationally expensive. Token caching provides a training-free, plug-and-play acceleration alternative. However, existing VLA caching does not fully exploit a key inductive bias of VLA models: text-vision synergy, wherein textual semantics guide the precise visual grounding of task-relevant regions. In particular, existing designs insufficiently account for head-wise reliability in attention aggregation and layer-wise stability in cache reuse. To address this, we propose Text-Vision Synergistic Token Caching (TVCache), a training-free framework for efficient VLA inference. TVCache filters attention heads based on text-vision information focus to improve task-relevant and physically consistent visual grounding. Concurrently, we introduce a reuse-layer selection mechanism guided by text-vision entropy differences to avoid caching unstable representations and improve cache resource allocation. Extensive experiments across four representative VLA models, two simulation benchmarks, and real-world robotic tasks demonstrate the effectiveness and generality of TVCache. At matched token-retention ratios, TVCache consistently improves task success over existing VLA caching with comparable computational cost. On OpenVLA-OFT, it improves average success by up to 14.5 percentage points over VLA-Cache at 12.5% retention while reducing FLOPs by 2.45x relative to full-token inference.
Read the full abstract

All authors Qianer Li, Chengjie Zhang, Jingwen Chen, Zanjia Tong, Jiyuan Zhang, Hong Zhang

Distributed Computing

HyDra: Demystifying and Taming Dynamic Context Parallelism at Production Scale

Zihao Fan, Yunzhuo Liu, Bo Jiang et al. (+4 more)

Long-context training runs on sequences whose lengths span orders of magnitude, and dynamic context parallelism (DCP) gives each sequence its own CP degree. Existing DCP systems either do not scale or perform poorly on mainstream models, leaving Megatron-Core (Mcore) DCP as the only option at production scale. Mcore DCP, however, sizes each degree to fit memory, which grows linearly with length while attention grows quadratically, so comparable token counts hide unequal computation. In our production 256K-context training job on more than 11K GPUs under Mcore DCP, per-rank microbatch times differ by up to 5x at similar token counts. The skew leads to a 46% pipeline bubble and a 13% data-parallel bubble. We present HyDra, a scalable load-driven DCP system. Its scheduler balances computation by pulling every rank toward one load target, and balance in turn makes that target solvable in closed form. It places sequences with lazy heaps rather than whole-pool scans. That balance asks for CP degrees larger than memory requires, so its CP engine nests an inner Ulysses group in a shallow outer ring, letting a higher degree lower computation and communication together. Evaluation at both scales shows consistent gains. On a 512-GPU testbed, HyDra raises throughput over Mcore DCP by 1.18x on average at 32K context and 2.48x at 256K. On a 2,048-GPU production job, it shrinks the pipeline bubble from 36% to 14%, cuts scheduling time by 2.3x, and raises throughput by 1.10-1.43x (avg. 1.25x) over Mcore DCP and 1.33-1.90x (avg. 1.59x) over static CP.
Read the full abstract

All authors Zihao Fan, Yunzhuo Liu, Bo Jiang, Changgang Zheng, Lin Zheng, Ray Ying, Key Zhang

Software Engineering

The Construction of an Empirical Dataset of Incomplete Software Changes from Open Source Projects

Savira Ramadhanty, Profir-Petru Pârţachi, Yoshiya Ishida et al. (+1 more)

During software development, a modification to a software component may propagate across the system, requiring precise identification and correct revision of all affected components. This is a complex task, and developers often (45.7%) miss related changes. To address this, several methods have been developed to extract co-change rules from files that are frequently changed together in the revision history. However, previous research evaluated the methods using artificially created incomplete changes, which may not be representative of real-world data. To solve this problem, we construct a dataset by mining incomplete changes from a collection of open-source software, using information about induced bugs and their respective fixes from an issue tracking platform. We also analyze the characteristics of incomplete changes using this constructed dataset and found that 89.4% of missed changes involved five or fewer files. Finally, we re-evaluate LCExtractor, an existing co-change rule extraction method, on our constructed dataset, and we identify the optimal sorting criterion and the impact of the number of used commits.
Read the full abstract

All authors Savira Ramadhanty, Profir-Petru Pârţachi, Yoshiya Ishida, Takashi Kobayashi

Artificial Intelligence

Dynamical Parameters: An Interpretability Framework for Time-Series Foundation Models

Kang Yang, Gaofeng Dong, Liying Han et al. (+1 more)

This work studies a central gap in interpreting time-series foundation models (TSFMs): a dynamical property may be accessible in a hidden state even when the forecast fails to respond correctly as that property changes. We formalize these properties as Dynamical Parameters, including trend slope, oscillation frequency, and autoregressive dependence. We compare their representation accessibility, measured by recovery from hidden states, with their forecast response, measured by agreement with the expected forecast change. Across nine frozen TSFMs and thirteen laws, 42 of 63 model-parameter cells achieve accessibility above 0.95, whereas their median reference-aligned response relative to the conditional reference is only 0.46. To explain this gap, causal geometry compares the hidden-state change required to produce the reference response with the change induced by the parameter intervention. Directly modifying the hidden state recovers the reference response, but the parameter intervention often moves the state in a different direction. These results show that accessible parameter information need not be expressed in forecasts when input changes miss the required hidden-state direction.
Read the full abstract

All authors Kang Yang, Gaofeng Dong, Liying Han, Mani Srivastava

Statistics Theory

Augmented James–Stein estimation for leading eigenvectors and eigenspaces in high dimensions

Giheon Seong, Seungki Hong and Sungkyu Jung

Building on the James–Stein approach to leading eigenvector estimation (Goldberg and Kercheval, Proc. Natl. Acad. Sci. USA 120, e2207046120, 2023), we develop a data-adaptive augmented James–Stein shrinkage framework for estimating leading eigenvectors and eigenspaces under a generalized spiked population model, in the high-dimensional regime where the dimension $p$ and sample size $n$ grow proportionally. For each spiked eigenvector, we construct an augmented target subspace that combines auxiliary information, either from domain knowledge or prior information, with the remaining sample spiked eigenvectors. This augmentation allows information shared across the sample spiked components to be exploited while retaining a fully data-driven shrinkage rule. We show that the resulting eigenvector estimator strictly improves upon standard PCA whenever the target subspace contains nonvanishing information about the population eigenvector, while asymptotically reverting to PCA when the target is uninformative. The individual estimators further yield a nested sequence of estimators for all leading spiked eigenspaces, with analogous dominance properties. The proposed estimator also strictly improves upon the existing HDLSS-motivated shrinkage estimator in the proportional high-dimensional regime. Simulation studies demonstrate substantial finite-sample gains and robustness to misspecification of the number of spikes.
Read the full abstract

Computer Vision

PlaylistEval: Can Video-Language Judges Be Trusted at Day Scale and Beyond?

Shayekh Bin Islam and Hwanjun Song

Video-language models are increasingly used as judges of video understanding, both for evaluating model outputs and for training reward models. Whether their judgments remain reliable when the evidence is buried in day-long videos has yet to be established. Existing benchmarks cannot answer this. Their videos are typically only a few minutes long, many answer pairs can be separated from the transcript alone, and collecting human judgments does not scale to ultra-long videos. We introduce PlaylistEval, an agentic framework that builds video-language judge benchmarks over 100-hour playlist collection without human annotation. It automatically generates questions with paired answers whose differences are controlled by causal degradation, so that every pair demands retrieval across the collection. The resulting benchmark contains 630 pairs across seven domains spanning both static and dynamic knowledge, and on a stratified subset of 152 pairs it agrees with human judgments 93.0% of the time (IAA 0.781). Evaluating 17 omnimodal and multimodal models from eight families reveals that frontier judges reach only 75.4% pairwise accuracy, while open-source judge models perform far behind. We further show that both retrieval and final judgment depend on using multiple modalities, and that judge accuracy degrades as the playlist set grows. We release our pipeline, benchmark, and evaluation code at https://playlisteval.github.io.
Read the full abstract

Artificial Intelligence

ControlScope: Workflow Revision and Reliability in LLM Agents

Jingjie Ning, Xueqi Li, Yibo Kong et al. (+1 more)

How much of a running workflow should a language model agent revise? ControlScope compares continuing generated code, editing the next tool call's data arguments, and replacing the unfinished workflow from the same public execution state. The nested permissions separate available repairs from the actions an agent selects. We evaluate one-time and repeated reviews across filesystem tasks, ALFWorld, and AppWorld. Across two source programs per task and three reasoning-reviewer draws on 20 filesystem tasks, FULL completes 15-16 tasks versus 13 for KEEP; across four fast draws it completes 10-13 versus 13. Fresh student-record confirmation reproduces a batch-read repair. ALFWorld fast panels yield KEEP/ARG/FULL scores of 85/86/87 on 87 tasks across 52 scenes and 134/134/127 on 134 tasks across four scenes; reasoning on the 87-task cohort also yields 85/86/87 with substantial review cost. An AppWorld V1 official-test panel of 585 task instances from 195 scenario templates shows small net differences. Frozen replays expose viable agent-written replacements interrupted by later revision in two failed file-organization runs. An offline source-trajectory midpoint comparison shows later reviews completing an insufficient repair. Five-call protection saves 19.4% of logged model output and loses one success across 20 fresh source runs. An argument-only shortcut shows that the broader sampled policy can overlook a cheaper successful edit available in both operation sets. These outcomes tie repair access to actual choices and subsequent execution.
Read the full abstract

All authors Jingjie Ning, Xueqi Li, Yibo Kong, Dongting Li

Algebraic Geometry

Gauss–Manin connections extend to holonomic coadmissible D-cap-modules

Andreas Bode and Finn Wiersig

We prove that the pushforwards of Gauss–Manin connections on smooth rigid analytic varieties along Zariski-open immersions are coadmissible and holonomic over Ardakov–Wadsley's sheaf D-cap of infinite order differential operators. This may be viewed as a rigid analytic variant of the classical theorem of Griffiths, Deligne, and Katz that Gauss-Manin systems have regular singularities after compactification. The proof combines p-adic de Rham comparison and Diao–Lan–Liu–Zhu's logarithmic Riemann–Hilbert correspondence with extendability results for log-connections with nilpotent residues. In the algebraic snc case, we also establish weak holonomicity via a p-adic Bernstein-Sato criterion of Bitoun and the first author.
Read the full abstract

Classical Analysis and ODEs

Weak Endpoint Estimate for Commutators of Rough Maximal Singular Integral operators

Shenglong Lin and Qingying Xue

Let $d\geq 2$ and $T^*_{\Omega,b}$ be the commutator of the rough maximal singular integral on $\mathbb{R}^d$ defined by $$T^*_{\Omega,b}f(x)=\sup_{\varepsilon>0}\left|\int_{|x-y|>\varepsilon}\bigl(b(x)-b(y)\bigr)\frac{\Omega(x-y)}{|x-y|^d}f(y)\,dy\right|.$$ Assume that $\Omega\in L(\log L)^2(\mathbb{S}^{d-1})$ has mean zero and that $b\in\operatorname{BMO}(\mathbb{R}^d)$. We prove that, for every $\lambda>0$, \[\bigl|\{x\in\mathbb{R}^d:T^*_{\Omega,b}f(x)>\lambda\}\bigr|\lesssim_{\Omega,b}\int_{\mathbb{R}^d}\frac{|f(x)|}{\lambda}\log\!\left(e+\frac{|f(x)|}{\lambda}\right)\,dx.\] The argument starts from a Calderón-Zygmund decomposition of $f$ and a dyadic linearization of the maximal truncation. A decomposition of $\Omega$ by size, followed by regularization of the spatial cutoffs and a microlocal decomposition, reduces the problem to a family of estimates with quantitative decay. The required decay follows by interpolating localized $L^1$ and $L^3$ bounds and using an analytic-family argument for the commutators.
Read the full abstract

Machine Learning

Riemannian Difference-of-Convex Optimization for K-Means Clustering

Meng Xu, Bo Jiang, Hanfu Zhang et al. (+2 more)

K-means is a widely adopted clustering approach in signal processing and machine learning. In this paper, we study K-means clustering through a cardinality-constrained formulation on a compact embedded submanifold. We replace the cardinality constraint with a difference-of-convex (DC) penalty and establish a global error bound to prove that the penalized and constrained formulations share the same global minimizers whenever the penalty parameter exceeds a finite threshold. To solve the resulting nonsmooth Riemannian DC problem, we reformulate it as a minimax problem and propose RADA-DC, a Riemannian alternating descent ascent method combining dual regularization with DC linearization. Under standard assumptions and suitable parameter choices, RADA-DC finds an $\epsilon$-Riemannian critical point within $O(\epsilon^{-3})$ iterations. We conduct experiments on synthetic and real-world datasets to demonstrate that the proposed method outperforms the tested baselines, including K-means++, in solution quality at competitive computational cost when the number of clusters is large.
Read the full abstract

All authors Meng Xu, Bo Jiang, Hanfu Zhang, Ya-Feng Liu, Anthony Man-Cho So

Computer Vision

MaLiang-Harness: A Programmable Path to Image and Video Generation

Haoyu Zhao, Zihao Zhang, Xudong Wang et al. (+4 more)

Executable programs offer explicit control over how images and videos are constructed, but generating runnable code is only the beginning of visual creation. A program can execute correctly while violating the requested composition, appearance, or motion. We define this discrepancy as the Program-to-Visual (P2V) gap and introduce MaLiang-Harness, a unified framework for organizing MLLM-driven visual generation into a persistent process of construction, inspection, and revision. Its central design is to make the evolving visual program, its construction history, and its verification share a common revision reference. We define the Persistent Executable Generation (PEG) state as preserving programs and task context. Traceable Generation Process (TGP) connects edits to rendered evidence, and Revision-aware Editing and Verification (REV) supports restoration and checks the current revision before completion. Together, these mechanisms coordinate planning, execution, and visual feedback across rendering backends. We evaluate 11 powerful closed-source MLLMs on MaLiang-IBench and four on MaLiang-VBench, measuring generation success, visual quality, and computational cost. GPT-6-Astra achieves 100% generation success on both benchmarks, with 96.0% of image tasks and 76.9% of video tasks meeting all quality thresholds. The comparison also reveals a mismatch between general capability scores and visual generation performance, with similarly scored models differing substantially in their ability to satisfy visual requirements. MaLiang-Harness provides a systematic basis for studying how MLLMs translate executable code into visual outcomes, exposing both the potential of programmable generation and the limitations of general benchmarks as predictors of this ability. The project is available at https://github.com/gulucaptain/MaLiang-Harness.
Read the full abstract

All authors Haoyu Zhao, Zihao Zhang, Xudong Wang, Jiaxi Gu, Zuxuan Wu, Yu-Gang Jiang, Shuicheng Yan

Functional Analysis

$\mathcal F$-Transitivity of Backward Shifts on Lattice Graphs

Xiang Chen, Cui Wang and Ze-hua Zhou

We study $\mathcal F$-transitivity of backward shifts on finite- and infinite-width directed lattice graphs. For the unilateral and bilateral finite-width lattices on weighted $\ell^p$- or $c_0$-spaces, we obtain exact weight characterizations of $\mathcal F$- and $\widetilde{\mathcal F}$-transitivity for an arbitrary Furstenberg family $\mathcal F$. If $\mathcal F$ is finitely invariant, these conditions also characterize topological $\mathcal F$-recurrence. For the infinite quadrant lattice with radial weights on weighted $\ell^2$-spaces, a finite-dimensional reduction and sharp smallest-singular-value estimates for Pascal-type transfer matrices yield equivalent characterizations of $\mathcal F$-transitivity, hypercyclicity and mixing. This provides a partial answer, in the radial $\ell^2$-setting, to the open problem posed by Baranov, Lishanskii and Papathanasiou concerning hypercyclicity on the infinite lattice graph. Examples at the critical exponential growth rate illustrate the finer dynamical information provided by our criteria.
Read the full abstract

Representation Theory

On Representations of $\mathrm{GL}_n(\mathrm{D})$ admitting a generalized linear period

Prem Dagar and Hariom Sharma

Let $\mathrm{D}$ be a quaternion division algebra over a non-Archimedean local field $\mathrm{F}$ of characteristic zero, and let $\mathrm{G}_n=\mathrm{GL}_n(\mathrm{D})$. Let $\mathrm{H}_{1,n-1}=\{\mathrm{diag}(g_1,g_2)\in\mathrm{G}_1,\ g_2\in\mathrm{G}_{n-1}\}$ and, for $s \in \mathbb{R}$, define $\chi_s(\mathrm{diag}(g_1,g_2))=\nu(g_1)^{2s}\nu(g_2)^{-2s}$. We classify, for $n=3,4$, the irreducible smooth representations of $\mathrm{G}_n$ admitting a generalized linear period with respect to $(\mathrm{H}_{1,n-1},\chi_s)$. Motivated by these results, we conjecture a complete classification for all $n>2$. Assuming this conjecture, we characterize such representations in terms of Langlands parameters: an irreducible smooth representation $\pi$ of $\mathrm{G}_n$ admits such a period if and only if $\mathfrak{L}(\pi)$ contains a Weil-Deligne subrepresentation isomorphic to $\mathfrak{L}(\nu^{-2s})$, the $(2n-4)$-dimensional parameter of $\nu^{-2s}$ on $\mathrm{G}_{n-2}$, and the four-dimensional quotient is the Langlands parameter of either the trivial representation of $\mathrm{G}_2$, with $s=\pm\frac{n-2}{2}$, or an irreducible infinite-dimensional $\mathrm{H}_{1,1}$-distinguished representation of $\mathrm{G}_2$. We also verify the Lapid-Prasad conjecture in this setting: the $L$-packet of an irreducible representation of $\mathrm{G}_n$ admitting a linear period with respect to $\mathrm{H}_{1,n-1}$ is invariant under $\rho\mapsto\widetilde{\rho}^{\theta}$.
Read the full abstract

Information Retrieval

SPRINT: Single-Step Generative Recommendation via Average Probability Velocity

Zhuo Cai, Shoujin Wang, Peilin Zhou et al. (+3 more)

Semantic ID (SID) based generative recommendation represents each item as a sequence of discrete tokens, and recommends by generating the SID of the item a user would like to interact with. Both dominant paradigms in this domain generally pay for generation token by token: autoregressive models decode the tokens left-to-right, while non-autoregressive models decode in parallel yet still need multiple rounds of refinement to stay competitive. Therefore, both generally spend multiple forward passes per item, a cost that is prohibitive in latency-sensitive recommender systems. We ask whether an item can be generated in a single forward pass, and answer it through a new perspective which we call average probability velocity. We view SID generation as a flow of token generation probabilities and characterize it by its average velocity over the whole generation process. We prove that this average velocity is fully determined by the average generation probability of each token. Therefore, we directly parameterize and learn the probabilities of all tokens in a single forward pass with a bidirectional Transformer. As these probabilities are generated independently across positions and the coherence among tokens is lost, we further design a dual-level flow contrastive objective to restore the coherence among an item's tokens. It contrasts the target SID against negative SIDs at both the token and SID levels. The token level ranks the generation probabilities of the target tokens above those of negative SIDs, while the SID level scores the tokens of each SID as a whole item for capturing token coherence of each item. Extensive experiments show that our model not only generates recommendations far more efficiently ($8.39-10.04\times$ speedup over the second-fastest AR/NAR method) but also attains superior recommendation accuracy ($7.77\%$ average improvement over the second-best.
Read the full abstract

All authors Zhuo Cai, Shoujin Wang, Peilin Zhou, Min Xu, Julian McAuley, Fang Chen

Astrophysics of Galaxies

Improving Lens Modelling from Ground-Based Imaging with Deconvolution

Eric Paic, Kenneth C. Wong, Anupreeta More et al. (+1 more)

Accurate and precise mass modelling of strong lensing systems is essential for extracting insights into cosmology, galaxy environments, and the nature of dark matter. The Hyper Suprime-Cam Subaru Strategic Program (HSC-SSP), which has mapped $\sim$1000 square degrees of the sky, discovered over 1000 definite or probable galaxy-scale gravitational lens candidates. As the resolution of ground-based imaging poses challenges for accurate lens modelling, high-resolution imaging is traditionally sought after for detailed analyses of these systems. In this work, we instead propose applying the STARRED algorithm, which leverages starlet wavelet regularisation, as a preprocessing step before lens modelling to maximise the scientific output of the large HSC-SSP sample. We test this approach on seven HSC-like mock lensing systems with different lensing configurations. Our results show that the resolution boost from $\sim$0.6" to $\sim$0.15" enables improved accuracy and sub-3% precision measurements of key lensing parameters such as the Einstein radius and the mass axis ratio for ground-based images. These results pave the way for accurate measurements of key strong lensing quantities from a large sample of ground-based observations.
Read the full abstract

All authors Eric Paic, Kenneth C. Wong, Anupreeta More, Anton T. Jaelani

Instrumentation and Detectors

EXO-200 Public Data Release for AI/ML Applications

S. Al Kharusi, G. Anton, I. Badhrees et al. (+103 more)

We present a public release of a subset of calibration data from the EXO-200 experiment, a liquid xenon time projection chamber designed to search for neutrinoless double-beta decay in $^{136}$Xe. The released dataset consists of approximately one million events collected during $^{228}$Th calibration campaigns performed near the end of Phase-II operations between February 26 and March 2, 2018. For each event, the dataset includes raw detector waveforms from the U-wire, V-wire, and avalanche photodiode (APD) readout channels together with reconstructed event quantities including charge energy, light energy, rotated energy, and charge-cluster information. The dataset is distributed in HDF5 format and is intended to support data preservation, educational activities, and the development of modern machine-learning techniques for rare-event physics. This paper describes the detector, dataset contents, file structure, and public access mechanism.
Read the full abstract

All authors S. Al Kharusi, G. Anton, I. Badhrees, P. S. Barbeau, V. Belov, T. Bhatta, M. Breidenbach, T. Brunner, G. F. Cao, W. R. Cen, C. Chambers, B. Cleveland, M. Coon, A. Craycraft, T. Daniels, L. Darroch, S. J. Daugherty, J. Davis, S. Delaquis, A. Der Mesrobian-Kabakian, R. DeVoe, A. Dolgolenko, M. J. Dolinski, J. Echevers, B. Eckert, W. Fairbank Jr., D. Fairbank, J. Farine, S. Feyzbakhsh, P. Fierlinger, Y. S. Fu, D. Fudenberg, P. Gautam, R. Gornea, G. Gratta, C. Hall, E. V. Hansen, J. Hoessl, P. Hufschmidt, M. Hughes, A. Iverson, A. Jamil, C. Jessiman, M. J. Jewell, A. Johnson, A. Karelin, L. J. Kaufman, T. Koffas, R. Krücken, A. Kuchenkov, K. S. Kumar, Y. Lan, A. Larson, B. G. Lenardo, D. S. Leonard, G. S. Li, S. Li, Z. Li, C. Licciardi, Y. H. Lin, R. MacLellan, T. McElroy, T. Michel, B. Mong, D. C. Moore, K. Murray, O. Njoya, O. Nusair, A. Odian, I. Ostrovskiy, H. Peltz Smalley, A. Perna, A. Piepke, A. Pocar, F. Retière, A. L. Robinson, P. C. Rowson, S. Schmidt, D. Sinclair, K. Skarpaas, A. K. Soma, V. Stekhanov, M. Tarka, S. Thibado, J. Todd, T. Tolba, T. I. Totev, R. Tsang, B. Veenstra, V. Veeraraghavan, P. Vogel, J.-L. Vuilleumier, M. Wagenpfeil, J. Watkins, M. Weber, L. J. Wen, U. Wichoski, G. Wrede, S. X. Wu, Q. Xia, H. Xu, D. R. Yahne, L. Yang, Y.-R. Yen, O. Ya. Zeldovich, T. Ziegler

Optimization and Control

Asymptotic Lower Bounds for Continuous Optimization

Oliver Hinder and Yuntian Jiang

Nonasymptotic convergence rates for optimization problems have been extensively studied across a wide range of settings, with carefully constructed worst-case instances establishing matching lower bounds for many algorithm classes. Recent work has shown, however, that these rates can often be improved in the asymptotic regime, while the corresponding asymptotic lower bounds remain largely unknown. We propose a general technique for converting existing nonasymptotic lower bound constructions into asymptotic lower bounds on Hilbert spaces. This allows us to show that several known asymptotic convergence upper bounds in Hilbert spaces are tight. In particular, we recover tightness of the $o(n^{-2})$ suboptimality guarantee for accelerated gradient descent on smooth convex functions \cite{attouch2016rate}, as well as the $o(n^{-1/2})$ suboptimality guarantee for gradient descent on smooth nonconvex functions \cite{gratton2025refining}. For stochastic convex optimization, we also develop a method that achieves an $o(n^{-1/2})$ suboptimality guarantee in finite dimensions, and prove a matching one-dimensional lower bound.
Read the full abstract

Computer Vision

CAT-Free: Multi-View Pedestrian Localization without Calibration, Annotations, or Target-Scene Training via Adaptive Geometric Filtering

Taigo Sakai, Hiroki Kouno, Naoki Kato et al. (+1 more)

Multi-camera pedestrian localization is useful for wide-area monitoring in public and commercial spaces. However, deploying these systems often requires considerable setup for each new environment. Existing methods typically require camera calibration, position annotations, or target-scene training. CAT-Free removes all three requirements. It uses synchronized RGB video as its only scene-specific input. Camera configuration is estimated directly from the video. Pedestrian locations are then estimated by combining observations from multiple cameras. Automatic camera estimation is not always accurate. This can produce unreliable pedestrian locations. CAT-Free therefore introduces two adaptive geometric filters. They remove unreliable position estimates. Their thresholds are estimated from each input sequence. CAT-Free achieves 82.5, 84.5, and 65.7 MODA on WildTrack, MultiviewX, and GMVD. It uses no supplied calibration, position annotations, or target-scene training. Published methods using such scene-specific information report 88.2–95.0 MODA on WildTrack and 83.9–96.5 on MultiviewX under their respective protocols. CAT-Free also transfers without retuning. It reaches 74.9 MODA on four additional sequences and 78.6 on an unseen 8-camera installation. Finally, localization uncertainty predicts MODA with $r=-0.98$. This provides a label-free estimate of localization reliability.
Read the full abstract

All authors Taigo Sakai, Hiroki Kouno, Naoki Kato, Kazuhiro Hotta

Machine Learning

One Sequence, Many Decodings: CAGenMol-2 Recasts Drug Design as Masked Molecular Inference

Yanting Li, Enyan Dai, Lei Wang et al. (+2 more)

Drug design couples property evaluation, conditional generation, structure-based design, and local optimization, yet machine learning systems typically address these capabilities with separate task-specific models. We introduce CAGenMol-2, a masked diffusion molecular language model that represents molecules, continuous scalar properties, and 3D protein pockets within a single wrapped sequence. Within this pretrained interface, downstream operations are selected by which sequence regions are observed or masked at inference, allowing one checkpoint to perform property prediction, property- and pocket-conditioned generation, and partial-constraint design without task-specific architectures or backbone fine-tuning. We further propose Adaptive Fragment Optimization (AdaFO), a gradient-free mask-and-refill search that turns the masked decoder into an iterative local molecular optimizer. On CrossDocked2020, AdaFO increases Success Rate from 30.2% to 70.8%, the best reported under this protocol, while largely preserving drug-likeness and diversity. Finally, scaffold-preserving directional editing and CRBN/VHL case studies demonstrate its use in compound design workflows spanning local molecular editing, structure-based prioritization, and downstream simulation-based screening.
Read the full abstract

All authors Yanting Li, Enyan Dai, Lei Wang, Wen-Cai Ye, Li Liu

Robotics

When World Models Lie: Adaptive Safety Analysis Under Wrong Imaginations

John Cao and Somil Bansal

World models offer a powerful substrate for safety reasoning in high-dimensional robotic systems, but they are also fallible: their predictions can be biased, miscalibrated, or confidently wrong. This creates a central challenge for latent-space safety filters, which often learn Hamilton-Jacobi safety value functions on the dynamics of a world model. If the world model is incorrect, the resulting value function can inherit its errors and produce overconfident safety estimates. Existing latent safety filters often rely on auxiliary signals such as ensemble disagreement or value-target consistency residuals for adaptation, but these signals can remain small even when the world model's predictions deviate from observations. We propose an adaptive latent safety filter that calibrates safety reasoning using directly observed world-model error. Our method uses Adaptive Conformal Inference to construct online uncertainty sets from discrepancies between predicted and observation-inferred latent states, then evaluates safety pessimistically by minimizing the learned value function over these sets. This allows the filter to remain minimally conservative when the world model is accurate, while becoming more cautious when observations reveal model mismatch. We provide a finite-time coverage guarantee for the adaptive uncertainty radius. Through simulation and hardware experiments, we show that our method significantly reduces failures relative to state-of-the-art latent safety filters while preserving task completion.
Read the full abstract

Computer Vision

PSM: Dataset Distillation Based on Precise Statistical Matching by Difficulty

Hongxu Ma, Guang Li, Shijie Wang et al. (+6 more)

Dataset distillation (DD) condenses a large original dataset into a small distilled dataset with high training utility. Decoupled statistical matching methods substantially reduce distillation time and memory overhead while achieving strong performance. However, they typically supervise all distilled samples using running statistics estimated from the entire original dataset. These statistics mainly capture the average feature distribution while overlooking differences in sample difficulty, limiting their ability to characterize the difficulty structure of the original data. To address this issue, we propose Precise Statistical Matching (PSM) by difficulty. After pretraining, PSM uses the Global Precision Score (GPS) to estimate image difficulty, ranks the samples within each class, and partitions each class into IPC (images per class) difficulty groups. During distillation, Statistics Updated Again (SUA) updates the teacher's batch normalization (BN) running statistics through forward passes on original samples from each group, providing difficulty-specific supervision for the corresponding distilled batch. Meanwhile, Initial Sample Screening (ISS) initializes distilled samples using original images from the corresponding difficulty group, providing an effective starting point for precise matching. Experiments across multiple datasets and model architectures demonstrate that PSM broadens the difficulty range of distilled samples and improves downstream performance in most evaluated settings. Code will be released.
Read the full abstract

All authors Hongxu Ma, Guang Li, Shijie Wang, Dongzhan Zhou, Suorong Yang, Baoli Sun, Takahiro Ogawa, Miki Haseyama, Zhihui Wang

Nuclear Theory

Nuclear scattering phase shifts with factorized geometric-time RODEO on a quantum processor

Myeong-Hwan Mun, Jubin Park, Myung-Ki Cheoun et al. (+1 more)

We reconstruct elastic $s$-wave neutron-proton ($np$) phase shifts from trapped spectra measured on IBM Aachen using a compressed RODEO circuit. For a schematic square-well interaction represented by classically constructed four-dimensional effective Hamiltonians, four positive-energy levels at five trap strengths supply twenty inputs to classical modified effective range expansion (MERE) extrapolation to free space. The six-cycle factorized geometric-time RODEO implementation (FG-R6) combines ancilla reuse, exact query-phase separation, post-compilation binding, and six numerically optimized geometric evolution times. Numerical cycle-count tests support this choice within the adopted local spectral tolerances. At the same three-qubit width and equal shot budgets, FG-R6 reduces median compiled depth and two-qubit-gate count by 33.1% and 37.2% relative to ten-cycle dynamic direct RODEO (direct R10), increases the median fitted amplitude, and lowers median finite-shot energy uncertainty from 0.610 to 0.534 keV. For this dataset's primary fit on the 0.1-30.0 MeV grid, the maximum central phase-shift deviation from exact-energy MERE decreases from $0.842^\circ$ to $0.163^\circ$, despite a larger root-mean-square (RMS) trapped-energy deviation. The full-grid gain arises mainly in the low-energy extrapolation region; direct R10 has slightly smaller residuals relative to the same reference on the 10.0-30.0 MeV common energy-interpolation subset. The FG-R6 central curve differs from the analytical square-well solution by at most $0.156^\circ$. Fit-form sensitivity remains appreciable and is assessed separately from finite-shot uncertainty. This reduced-space benchmark connects RODEO circuit compression to nuclear continuum observables and shows why circuit performance must be assessed through the scattering reconstruction and its energy range, not spectral RMS errors alone.
Read the full abstract

All authors Myeong-Hwan Mun, Jubin Park, Myung-Ki Cheoun, Eunja Ha

Robotics

TLC-DiT: Task-Aligned Local Visual Conditioning for Robust Multitask Robot Manipulation

Xianbo Cai, Hideyuki Ichiwara, Zihang Wang et al. (+2 more)

Language-conditioned robot policies have made clear progress in multitask manipulation, but task-relevant local visual evidence usually stays hidden inside a visual backbone or attention layers. This leaves the policy difficult to inspect and fragile under visual change, two symptoms of a missing explicit, task-aligned local visual channel. We present TLC-DiT, a plug-in extension of the Multitask Diffusion Transformer (DiT) policy that adds explicit task-guided local visual feature maps without changing the diffusion objective or the action-generation process. For each camera view, frozen DINOv2 patch features are modulated by the CLIP task embedding through FiLM and refined by a lightweight CoordConv CNN adapter into smooth spatial maps, which are concatenated with the original global image, language, joint-state, and timestep conditions. On LIBERO, TLC-DiT reaches a 93.5% average success rate, compared with 86.5% for Multitask DiT and 79.25% for SmolVLA. On LIBERO-plus, the total success rate improves from 54.07% to 57.24%, with larger gains under camera, background, and sensor-noise changes. In real-world bimanual tasks, TLC-DiT raises Teabag Putting completion from 44% to 89% while maintaining comparable Match Box Opening performance. Feature-map visualizations confirm that the model attends to task-relevant regions across views and perturbations, providing a direct way to inspect the visual evidence.
Read the full abstract

All authors Xianbo Cai, Hideyuki Ichiwara, Zihang Wang, Yijun Lu, Tetsuya Ogata