Showing 1–50 of 1,424 results for author: Wen, J

Rethinking the Evaluation and Optimization of LLM-Based Social Simulation

Authors: Pei Wang, Xu Chen, Ji-Rong Wen

Abstract: LLM-based social simulation is a promising complement to traditional methods such as surveys and behavioral experiments. A core question is how to evaluate the fidelity of LLM-simulated human behavior and optimize LLMs toward it. Prevailing practice evaluates by accuracy, checking whether the model selects the single response observed from a human, and trains the LLM to reproduce this hard label.… ▽ More LLM-based social simulation is a promising complement to traditional methods such as surveys and behavioral experiments. A core question is how to evaluate the fidelity of LLM-simulated human behavior and optimize LLMs toward it. Prevailing practice evaluates by accuracy, checking whether the model selects the single response observed from a human, and trains the LLM to reproduce this hard label. However, human behavior is inherently subjective: the same person in the same situation may reasonably act differently, so an observed response is only one draw from an underlying response distribution, rendering accuracy-based evaluation unreliable and hard-label training misleading. To address these problems, we first introduce the subjectivity coefficient, an entropy-based quantity distinguishing objective tasks such as coding from subjective ones such as social simulation, and use it to systematically analyze how accuracy-based evaluation and hard-label training fail as subjectivity grows. Based on the subjectivity coefficient, we propose Subjectivity-Adaptive soft-Label Training (SALT): it pools observed outputs from semantically nearby inputs into soft distributional labels, with an aggregation radius adapted to the estimated subjectivity of each input; in the near-objective limit the neighborhood shrinks, so SALT naturally falls back to standard single-label training. Moreover, since existing datasets record only single observed responses and cannot support distributional evaluation, we construct SUBJSIM, a benchmark of 19,300 contexts covering 193 annotators and 100 subjective questions. Since real-world data typically provide only a single observation per input, our experiments train models from single observed outputs while evaluating them against the full response distributions, verifying feasibility in realistic settings. Results on SUBJSIM demonstrate the advantages of our method. △ Less

Submitted 20 August, 2026; originally announced August 2026.

Variational r-Adaptive Cloth Simulation

Authors: Jiahao Wen, Zhen Chen, Jernej Barbič, Danny M. Kaufman

Abstract: We present the first r-adaptive method for simulating cloth dynamics and statics with frictional contact in modern cloth pipelines. Thin cloth requires high effective spatial resolution to reproduce wrinkles, folds, buckling, and sharp contact features. However, applying existing variational r-adaptivity to piecewise-linear shells reveals two coupled failure modes. Discretized incremental-potentia… ▽ More We present the first r-adaptive method for simulating cloth dynamics and statics with frictional contact in modern cloth pipelines. Thin cloth requires high effective spatial resolution to reproduce wrinkles, folds, buckling, and sharp contact features. However, applying existing variational r-adaptivity to piecewise-linear shells reveals two coupled failure modes. Discretized incremental-potential (IP) optimization can become trapped in poor local minima, yielding suboptimal physical configurations. It can also lower IP artificially by collapsing elements, invalidating the finite-element approximation on which the objective relies. We address both problems with degeneracy-activated quality regularization. The regularizer remains inactive for well-shaped elements, preserving anisotropic adaptation and local densification, but becomes strong near degeneracy. It suppresses spurious low-energy basins, improves escape from suboptimal physical minima, and prevents element bunching, a cloth-specific failure in which elements progressively collapse as cloth slides across sharp contact features. For practical performance, we introduce a dynamic nonlinear solver that exploits within-timestep coherence through accelerated derivative evaluation and dynamic IPC tolerance updates for r-adaptive iterative trust-region (ITR) solves. This yields a 3-6x speedup over prior optimal ITR. Experiments on challenging frictional-contact scenarios show that, under equal vertex-count and time-budget constraints, our method achieves higher visual fidelity than fixed meshes. △ Less

Submitted 18 August, 2026; originally announced August 2026.

Comments: 11 pages

ClawGym II: Exploring Black-Box RL on Agent Harness

Authors: Huatong Song, Fei Bai, Ming Yang, Renyuan Li, Jia Deng, Jujie He, Zhange Zhang, Daixuan Cheng, Yan Xing, Qi Yun, Xuxing Chen, Danyang Li, Feng Chang, Chuan Hao, Ran Tao, Jian Yang, Bryan Dai, Wayne Xin Zhao, Mingjie Tang, Ji-Rong Wen

Abstract: Agent harnesses have substantially improved performance on long-horizon tasks by coordinating agent interactions with the environment. However, reinforcement learning through complex harnesses remains largely unexplored, as scaling such training to long-horizon agent tasks introduces fundamental challenges. In this work, we present a unified black-box RL framework for stable and scalable optimizat… ▽ More Agent harnesses have substantially improved performance on long-horizon tasks by coordinating agent interactions with the environment. However, reinforcement learning through complex harnesses remains largely unexplored, as scaling such training to long-horizon agent tasks introduces fundamental challenges. In this work, we present a unified black-box RL framework for stable and scalable optimization of general agents through complex harnesses. Concretely, we first build a sandbox-based execution infrastructure that isolates task environments and harnesses within temporary sandboxes for large-scale concurrent rollouts. We then decouple policy optimization from opaque harness execution and place a serving proxy at the model boundary to capture model calls. To reconstruct multi-turn trajectories and improve training efficiency, we organize the captured calls into prefix trees and further adapt both critic-based PPO and critic-free GRPO to optimize over the recovered tree structure. Meanwhile, we maintain training-inference consistency throughout the optimization process. Finally, we introduce mix-harness training, allowing a single model to be jointly optimized by heterogeneous harnesses. With Qwen3-30A3B, black-box RL improves Pass@1 on ClawGym-Bench by 9.98 and 14.81 points through OpenClaw and Claude Code, respectively, while remaining stable over 200-400 optimization steps. Moreover, the framework yields consistent gains on more challenging tasks such as JobBench and OfficeQA. Overall, our framework enables effective, stable, and scalable optimization of general agents through black-box harnesses, supporting unified training across heterogeneous execution systems. △ Less

Submitted 17 August, 2026; originally announced August 2026.

PERO: Efficient Robust Post-Training Foundation Models for Encrypted Traffic Classification

Authors: Wumei Du, Jiarong Wen, Kaiyu Zhang, Zi Yang, Yiqin Lv, Longfei Zhang, Dong Liang, Zheng Xie

Abstract: Encrypted traffic classification is vital for network security, yet real-world deployments are inherently sensitive to rare but high-loss errors such as misclassification of malicious traffic. The encrypted traffic foundation model, as a promising general-purpose technique, can achieve impressive overall performance. However, employing standard objectives such as empirical risk minimization often… ▽ More Encrypted traffic classification is vital for network security, yet real-world deployments are inherently sensitive to rare but high-loss errors such as misclassification of malicious traffic. The encrypted traffic foundation model, as a promising general-purpose technique, can achieve impressive overall performance. However, employing standard objectives such as empirical risk minimization often overlooks high-risk tail events, and commonly used performance metrics hardly reflect robustness limitations in risk-sensitive scenarios. Directly applying robust optimization objectives, such as conditional value-at-risk, to post-training is computationally prohibitive for large models, as identifying high-loss samples exhausts substantial computation. To this end, we propose Pre-Evaluation Robust Optimization (PERO), an efficient robust post-training framework for encrypted traffic foundation models. PERO employs a lightweight proxy to estimate sample-wise risk and selects a subset of high-risk samples to update the foundation model, decoupling risk estimation from expensive large-model optimization. Extensive experiments on typical encrypted traffic datasets show that PERO achieves competitive or superior robustness and average performance compared to outstanding robust post-training methods, while significantly reducing computational and memory costs. △ Less

Submitted 15 August, 2026; originally announced August 2026.

Comments: 16 pages, 6 figures, 6 tables, conference

MSC Class: 68T07 (Primary) 68T20; 68T01; 68T10 (Secondary) ACM Class: I.5.1

Structure of large $t$-intersecting families I: Stability for the Hilton--Milner--Frankl theorem

Authors: Jie Wen, Benjian Lv

Abstract: We study the structure of large $t$-intersecting families. A family of $k$-subsets of an $n$-set is $t$-intersecting if every two of its members intersect in at least $t$ elements. A $t$-intersecting family is non-trivial if no $t$-subset is contained in all its members. We prove several stability results for the seminal Hilton--Milner--Frankl theorem. First, for any fixed… ▽ More We study the structure of large $t$-intersecting families. A family of $k$-subsets of an $n$-set is $t$-intersecting if every two of its members intersect in at least $t$ elements. A $t$-intersecting family is non-trivial if no $t$-subset is contained in all its members. We prove several stability results for the seminal Hilton--Milner--Frankl theorem. First, for any fixed $η,\varepsilon,θ\in(0,1)$, we prove that if $k/t\geq1+η$ and $n=Ω(tk^{1+\varepsilon})$, then every non-trivial $t$-intersecting family of size greater than $(1+θ)|\mathcal{K}|$ is a subfamily of one of the two extremal families in the theorem, where $\mathcal{K}$ is an explicit large non-trivial $t$-intersecting family. The key ingredient in the proof is a removal lemma. We also obtain a classification of all $t$-intersecting families with size bounded below by $|\mathcal{K}|$ minus an explicit lower-order term, provided that $k\geq t+4\geq6$ and $n\geq t+6\cdot\max\{(t+2)^2, k(k-t)\}$. This strengthens results of Cao--Lv--Wang (2021) and Frankl (2025) for a broad range of $k$ and $t$ (for example, when $k-t\geq2\sqrt{t}$). As an application of this classification, we determine the largest $t$-intersecting families for each prescribed lower bound on $t$-diversity not exceeding $t(n-k)$, thereby obtaining $t$-intersection versions of results of Han and Kohayakawa (2017) and Kupavskii (2025). To establish these results, we develop techniques based on the spread approximation method and the $t$-cover method, which may be useful for other intersection problems. △ Less

Submitted 14 August, 2026; originally announced August 2026.

MSC Class: 05D05

GenFAR: A generalized representation of brain structure, derived from 49,246 multi-cohort MRIs via deep learning

Authors: Vishnu M. Bashyam, Guray Erus, Junhao Wen, Pratik Chaudhari, Randa Melhem, Sindhuja Govindarajan Tirumalai, Gareth Harman, Yong Fan, Colin L. Masters, Paul Maruff, Sterling C. Johnson, Jurgen Fripp, Duygu Tosun, John C. Morris, Daniel S. Marcus, Pamela LaMontagne, Tammie Benzinger, Susan R. Heckbert, Mark Espeland, Marilyn S. Albert, Andrew J. Saykin, Paul M. Thompson, Timothy J. Hohman, Susan M. Resnick, R. Nick Bryan , et al. (7 additional authors not shown)

Abstract: Deep learning models for neuroimaging have largely been developed for individual tasks, limiting knowledge transfer across applications. Here we introduce GenFAR, a modular deep learning framework that learns general, clinically informed features from brain MRIs. We trained this modular architecture on 49,246 individuals across 11 cohorts, using 17 diverse classification and regression tasks spann… ▽ More Deep learning models for neuroimaging have largely been developed for individual tasks, limiting knowledge transfer across applications. Here we introduce GenFAR, a modular deep learning framework that learns general, clinically informed features from brain MRIs. We trained this modular architecture on 49,246 individuals across 11 cohorts, using 17 diverse classification and regression tasks spanning cognition, clinical, diagnosis, demographics, and biomarkers. This yields aggregated, focused feature sets that capture rich, clinically- and biologically-relevant brain representations. We developed a sequential learning approach where tasks progressively build on previously learned representations. Through an analysis of 5,000 task sequences, we identified an optimal sequence length of six tasks and introduced a Donor Score metric to quantify each task's contribution to downstream performance. This analysis revealed five consistently strong donor tasks (Age, AD/MCI, MMSE, Hypertension, Hyperlipidemia) that formed the base of our sequential model. We demonstrated the utility of our learned representation, in various tasks beyond those included in the training set, to serve as the foundation for specialized secondary predictors. We further showed that using the learned feature representation can substantially increase the sample efficiency of secondary deep learning training tasks and models, as well as improve their accuracy. △ Less

Submitted 12 August, 2026; originally announced August 2026.

Not All Visual Tokens Are Equally Safe to Remove:Consequence-Sensitive Visual Token Compression

Authors: Jingbo Wen, Liang He, Mingyu Cao, Haoyu Wang, Minxuan Hu, Kangning Cui, Xilu Wang

Abstract: Visual token compression for vision--language models (VLMs) has largely relied on criteria such as attention, redundancy, and uncertainty to maximize average accuracy under a fixed compute budget, implicitly assuming that all errors carry equal cost. However, the consequence of an incorrect prediction on downstream tasks is rarely symmetric: misreading an invoice amount can be far more costly than… ▽ More Visual token compression for vision--language models (VLMs) has largely relied on criteria such as attention, redundancy, and uncertainty to maximize average accuracy under a fixed compute budget, implicitly assuming that all errors carry equal cost. However, the consequence of an incorrect prediction on downstream tasks is rarely symmetric: misreading an invoice amount can be far more costly than misclassifying a background color. Motivated by this, we introduce consequence-sensitive visual token compression, which allocates visual computation across requests according to their potential error costs. Our method follows a calibrate-then-allocate procedure, estimating consequence-specific error-budget curves offline and applying the calibrated token budgets online using consequence signals available from question or task information. On a controlled within-task benchmark, high- and low-consequence questions are drawn from the same document images, so content alone cannot reveal which questions are costly to get wrong. In this setting, our method reduces high-stakes errors from 0.300 to 0.133 under the same total token budget, whereas a content-driven allocator performs no better than uniform allocation. Measuring how error rates change with token budget across different cost ratios, we derive an allocation frontier: uniform allocation is optimal when errors are equally costly, and token transfer toward high-consequence questions becomes increasingly beneficial as the cost gap grows. This allocation principle generalizes well across three dense vision-language benchmarks, two budget realization mechanisms (token deletion and resolution reallocation), two VLM architectures, and multiple token selection strategies. On a realistic mixed workload, consequence-sensitive allocation reduces cost-weighted error by 38% while achieving approximately 21% lower latency than full-resolution inference. △ Less

Submitted 10 August, 2026; originally announced August 2026.

From Relevance to Execution Utility: Reward-Aware Dynamic Execution Gating for Skill-Based LLM Agents

Authors: Liang He, Jingbo Wen, Hongyu Gu, Hao Li, Haoyu Wang, Yixiong Chen, Kangning Cui, Xilu Wang

Abstract: Agent skills are increasingly used to equip large language model (LLM) agents with reusable procedural knowledge. Although recent work has substantially improved skill retrieval due to the increasing skill libraries, retrieving a plausible skill bundle does not guarantee that executing it is worthwhile. Since every skill-conditioned rollout is computationally expensive, deciding whether a retrieve… ▽ More Agent skills are increasingly used to equip large language model (LLM) agents with reusable procedural knowledge. Although recent work has substantially improved skill retrieval due to the increasing skill libraries, retrieving a plausible skill bundle does not guarantee that executing it is worthwhile. Since every skill-conditioned rollout is computationally expensive, deciding whether a retrieved bundle should be executed has become an increasingly important challenge. To this end, we introduce the Reward-Aware Dynamic Execution Gate (RADEG), a lightweight, retriever-agnostic decision layer between skill retrieval and agent execution. RADEG learns a low-cost surrogate model that predicts the execution utility of a query--bundle pair before the expensive rollout is launched. To obtain informative supervision while controlling for task difficulty, we locally perturb each retrieved bundle by deleting, adding, or replacing one skill, producing matched same-query rollouts that isolate the effect of bundle composition on verifier reward. During deployment, RADEG updates only a warm-started logistic head as new verifier feedback becomes available, enabling inexpensive adaptation of the execute/skip boundary without retraining either the retriever or the agent. Under a query-level held-out evaluation on 288 collected rollouts, RADEG substantially reduces unnecessary agent executions while preserving a large fraction of the downstream verifier reward. It consistently outperforms relevance-based and random gating across different execution budgets, demonstrating that execution-aware surrogate modeling provides a practical and cost-effective complement to skill retrieval. △ Less

Submitted 10 August, 2026; originally announced August 2026.

ARMOR: Accelerating RTL Simulation by Mitigating the Front-End Bottleneck Using Node Compression

Authors: Jiaping Tang, Jianan Mu, Zhiteng Chao, Jingzhong Wen, Jing Ye, Huawei Li

Abstract: RTL simulation is indispensable in chip design. High-performance simulators typically lower each node in the RTL graph into an instruction sequence. Although this per-node lowering enables aggressive compiler optimizations, it dramatically increases the code footprint, severely exceeding instruction cache capacity and causing front-end bottlenecks. Our profiling reveals that over 50% of pipeline s… ▽ More RTL simulation is indispensable in chip design. High-performance simulators typically lower each node in the RTL graph into an instruction sequence. Although this per-node lowering enables aggressive compiler optimizations, it dramatically increases the code footprint, severely exceeding instruction cache capacity and causing front-end bottlenecks. Our profiling reveals that over 50% of pipeline stalls are caused by the CPU front-end, becoming a key performance bottleneck in state-of-the-art RTL simulators. However, reaping the optimization benefits of fully unrolling the RTL graph while simultaneously reducing the code footprint to mitigate front-end bottlenecks remains highly challenging. In this paper, we propose ARMOR, an efficient RTL simulator designed to alleviate the front-end bottleneck through node compression. The key idea is to exploit the data parallelism exposed by the unrolling RTL graph and the insufficient bit-space utilization revealed by per-node lowering, leveraging bit-level data parallelism to compress multiple nodes simultaneously, so that a single instruction sequence can serve multiple nodes instead of one per node. To achieve profitable node compression, we first propose a module-aware isomorphic subgraph identification method that leverages structural isomorphism across module instances to systematically identify compression opportunities at the subgraph level. We then propose an alignment-aware dense packing strategy that groups nodes into packs according to dataflow dependencies while preserving data reuse, complemented by greedy merging strategies to enhance bit-space utilization. Finally, we implement a unified bit-level parallelism scheme to support bit-level parallel execution of compressed nodes. Experimental results show that ARMOR achieves 1.6x speedup on CPU designs and 2.7x speedup on AI accelerators compared to state-of-the-art simulators. △ Less

Submitted 9 August, 2026; originally announced August 2026.

Comments: Accepted by 2026 59th IEEE/ACM International Symposium on Microarchitecture (MICRO)

Biorthogonal-only Floquet Dynamical Quantum Phase Transitions

Authors: Jiangrong Wen, Qidong Yuan, Zi-Xiang Hu, Jian-Jun Dong

Abstract: Non-Hermitian dynamical quantum phase transitions (DQPTs) are intrinsically sensitive to the choice of inner product under nonunitary time evolution. Although the biorthogonal formulation based on associated states provides a normalized Loschmidt echo with a probabilistic interpretation, previous studies have found biorthogonal and self-normal DQPTs to occur in the same parameter regimes, suggesti… ▽ More Non-Hermitian dynamical quantum phase transitions (DQPTs) are intrinsically sensitive to the choice of inner product under nonunitary time evolution. Although the biorthogonal formulation based on associated states provides a normalized Loschmidt echo with a probabilistic interpretation, previous studies have found biorthogonal and self-normal DQPTs to occur in the same parameter regimes, suggesting that the two forms of dynamical criticality are concomitant. Here we demonstrate that this is not the case. In an exactly solvable periodically driven non-Hermitian Su-Schrieffer-Heeger chain, we uncover a finite biorthogonal-only Floquet DQPT regime, where the biorthogonal Loschmidt rate becomes nonanalytic while the self-normal Loschmidt rate remains smooth. The critical conditions are obtained analytically, showing that the onset of biorthogonal Floquet DQPTs is locked to the exceptional lines of the effective Floquet Hamiltonian, whereas self-normal criticality has no corresponding spectral boundary. Moreover, for each critical momentum, the biorthogonal DQPT exhibits a pair of critical times within every driving period, whereas the self-normal DQPT exhibits only one. Our results establish a fundamental distinction between biorthogonal and self-normal DQPTs, thereby opening a route toward new nonequilibrium quantum phenomena in non-Hermitian systems. △ Less

Submitted 7 August, 2026; originally announced August 2026.

Comments: 6 pages, 4 figures

CalibForge: Adversarial Solver Calibration for Scaling Learnable Terminal Tasks

Authors: Fanzhe Meng, Guoxin Chen, Jiale Zhao, Shuang Sun, Zhiyu Lin, Wayne Xin Zhao, Ruihua Song, Ji-Rong Wen, Kai Jia

Abstract: Training terminal agents requires executable and verifiable tasks that are not merely solvable, but appropriately challenging for learning. Executable validation establishes feasibility, yet does not reveal how a task behaves relative to a given solver setting. In this paper, we present CalibForge, an autonomous terminal-task synthesis system that uses verified solver behavior to revise candidate… ▽ More Training terminal agents requires executable and verifiable tasks that are not merely solvable, but appropriately challenging for learning. Executable validation establishes feasibility, yet does not reveal how a task behaves relative to a given solver setting. In this paper, we present CalibForge, an autonomous terminal-task synthesis system that uses verified solver behavior to revise candidate tasks through adversarial solver calibration. Multi-solver calibration targets disagreement within a heterogeneous solver pool, whereas contrastive solver calibration targets a designated strong-pass/weak-fail relation; both operationalize a solver-relative learnable zone anchored in demonstrated solvability. Using CalibForge, we construct 5,431 calibrated terminal tasks. Our ablations show that both strategies yield more effective supervision than authoring and validation alone or ordinary single-solver feedback. Models trained on the full collection achieve 32.58% and 47.57% on Terminal-Bench 2.0. The largest improvements over the corresponding base model reach 24.71 percentage points on Terminal-Bench 2.0, 27.68 points on SWE-bench Pro, and 30.04 points on Doc2Repo. Together, these results support solver-relative learnability as a practical target for constructing effective and transferable agent training data. △ Less

Submitted 6 August, 2026; originally announced August 2026.

Comments: Dataset: https://huggingface.co/datasets/AweAI-Team/CalibForge. Repository: https://github.com/AweAI-Team/CalibForge

Strongly Enhanced Charge-Density Waves and Correlated Insulating State in Atomically Thin 1$T$-TaS$_2$

Authors: Gan Liu, Yulu Liu, Qiling Luo, Zhentao Huang, Kenji Watanabe, Takashi Taniguchi, Meiyu Wang, Jinsheng Wen, Yi Lu, Xiaoxiang Xi

Abstract: We investigate thickness-dependent charge-density-wave (CDW) transitions in 1$T$-TaS$_2$ using temperature-dependent Raman spectroscopy and electrical transport. Raman measurements show that the incommensurate, nearly commensurate, and commensurate CDW phases persist down to the monolayer limit. As the thickness is reduced, the transition temperatures increase, accompanied by an orders-of-magnitud… ▽ More We investigate thickness-dependent charge-density-wave (CDW) transitions in 1$T$-TaS$_2$ using temperature-dependent Raman spectroscopy and electrical transport. Raman measurements show that the incommensurate, nearly commensurate, and commensurate CDW phases persist down to the monolayer limit. As the thickness is reduced, the transition temperatures increase, accompanied by an orders-of-magnitude rise in sheet resistance and a sharp reduction in the carrier localization length. The first-order hysteretic CCDW-NCCDW transition is uniquely absent in the monolayer. Calculations suggest that the enhanced CDW in thin layers originates from strengthened Coulomb interactions due to reduced out-of-plane screening, particularly in the nonlocal component. These findings highlight the cooperative roles of electron correlation, electron-phonon interaction, and interlayer coupling in shaping the ground state and transition dynamics of atomically thin 1$T$-TaS$_2$, opening pathways for engineering correlated phases in two-dimensional CDW systems. △ Less

Submitted 5 August, 2026; originally announced August 2026.

Journal ref: Phys. Rev. Lett. 137, 066502 (2026)

LLaDA MoE v2: Scaling Mixture-of-Experts Diffusion Language Models

Authors: Fengqi Zhu, Shaoxuan Xu, Jingyang Ou, Zebin You, Yipeng Xing, Huabin Liu, Xiaolu Zhang, Jun Zhou, Zhenzhong Lan, Yankai Lin, Wayne Xin Zhao, Jianguo Li, Chongxuan Li, Ji-Rong Wen

Abstract: Diffusion language models (dLLMs) offer an alternative to autoregressive (AR) language modeling, yet the scaling behavior of Mixture-of-Experts (MoE) dLLMs remains poorly understood. We systematically characterize how optimization hyperparameters, compute allocation, and architecture scale for MoE dLLMs, identifying quantitative differences from scaling trends previously reported for AR models. Sp… ▽ More Diffusion language models (dLLMs) offer an alternative to autoregressive (AR) language modeling, yet the scaling behavior of Mixture-of-Experts (MoE) dLLMs remains poorly understood. We systematically characterize how optimization hyperparameters, compute allocation, and architecture scale for MoE dLLMs, identifying quantitative differences from scaling trends previously reported for AR models. Specifically, for optimization, the optimal nominal batch size grows faster, while the optimal learning rate decays more rapidly with compute. For model--data allocation, IsoFLOP analysis reveals a slight data-side tilt: the optimal token budget grows faster than activated model-side computation. For MoE architecture, larger scales increasingly favor larger expert pools at fixed activated capacity, while moderate expert granularity remains consistently effective and the preferred fraction of activated capacity assigned to shared experts remains stable across scales. Guided by these findings, we train LLaDA MoE v2, a 30B-A3B dLLM, from scratch on 23.5T tokens. With approximately 65\% as many pretraining tokens as Qwen3, LLaDA MoE v2 approaches Qwen3 on several knowledge, reasoning, and coding benchmarks. After supervised fine-tuning alone, it outperforms SDAR Chat on seven of eight reasoning and coding benchmarks and remains close to Qwen3 on several tasks. These results establish practical scaling laws and design principles for MoE dLLMs. △ Less

Submitted 4 August, 2026; originally announced August 2026.

Single Canonical Prompts Underestimate LLM Safety's Surface-Form Sensitivity

Authors: Yongxi Zhou, Junwei Yao, Yuanzhe Liu, Zihan Dong, Wenbo Ye, Jiaxi Wen, Lai Yun Choi

Abstract: A benchmark score is a measurement instrument, yet most benchmarks read each item at a single canonical surface form. We ask whether that reading is faithful: when an item's intent is held fixed and only its meaning-preserving surface form varies, does the canonical-form score estimate model behavior well, and how much of any variation is decoding/judge noise rather than signal? We instantiate thi… ▽ More A benchmark score is a measurement instrument, yet most benchmarks read each item at a single canonical surface form. We ask whether that reading is faithful: when an item's intent is held fixed and only its meaning-preserving surface form varies, does the canonical-form score estimate model behavior well, and how much of any variation is decoding/judge noise rather than signal? We instantiate this in safety, a high-stakes setting with no gold label to average toward. To avoid prior confounds, we pre-author the reformulations (refusal-free, mostly non-LLM: machine back-translation and a Matrix-Language-Frame code-switch generator) so an identical surface form reaches every model, score all responses with one human-anchored, vendor-neutral judge (Claude, kappa = 0.86 vs. human on unsafe compliance, stable across languages, cross-checked by GPT-4o), and verify intent preservation. On 370 seeds x 5 surface forms x 5 models, no single transformation is uniformly most dangerous (6 of 20 per-transformation McNemar tests survive correction, most protective). Yet evaluating only the canonical prompt underestimates unsafe compliance: the union of unsafe outcomes across forms exceeds even the worst single form by 3.3-12.9 pp, with bootstrap 95% CIs excluding zero for all five models, and 5-13% of seeds safe on canonical are unsafe under some reformulation -- above a zero stochasticity floor (canonical resampled five times at temperature 0 gives 0/370 new exposures). The size of this gap is model-dependent (largest on Gemini 2.5 Pro). One form recovers only ~53% of a model's observed unsafe surface and about three reach 85% -- a redundancy characterization of this form set, not of a defined population. A benign control (XSTest) suggests the instability is bidirectional, though the benign and harmful pools are not item-matched. We release the dataset, code, and per-response labels. △ Less

Submitted 1 August, 2026; originally announced August 2026.

Comments: 9 pages, 3 tables

ACM Class: I.2.7; K.4.2

AMTFV: Agentic Mathematical Tool-Flow Verification for LLM Self-Correction

Authors: Rui Zou, Yutao Zhu, Mengqi Wei, Ji-Rong Wen

Abstract: Large language models have demonstrated strong mathematical problem-solving capabilities, yet reliably verifying their candidate answers remains challenging. Existing representative methods mainly revise outputs through natural-language reflection or assist verification by directly generating verification programs; the former may not reliably support exact computation, whereas the latter premature… ▽ More Large language models have demonstrated strong mathematical problem-solving capabilities, yet reliably verifying their candidate answers remains challenging. Existing representative methods mainly revise outputs through natural-language reflection or assist verification by directly generating verification programs; the former may not reliably support exact computation, whereas the latter prematurely couples mathematical modeling with low-level implementation. We propose AMTFV (Agentic Mathematical Tool-Flow Verification). By introducing Mathematical Tool Flow (MTF) as an interrupt--execute--resume interface, AMTFV decouples verification modeling from concrete execution and supports exact computation through a mathematical toolbox. Specifically, the verification agent first constructs a verification workflow, encodes the mathematical objects and computational intent requiring reliable execution in an MTF request, and sends it to the mathematical toolbox agent. The latter parses the request, generates executable calls, and dispatches them to the backend for exact computation. Tool outputs then support candidate-answer adjudication, answer revision, and verification-workflow revision. We evaluate AMTFV on five challenging mathematical reasoning datasets with seven model configurations from DeepSeek, GPT, and Gemini. Experimental results show that AMTFV outperforms the representative baselines evaluated in this study overall; under an individual model configuration, it improves average accuracy over the strongest baseline by up to 8.3 percentage points, with larger gains on samples of medium and high verification complexity. △ Less

Submitted 31 July, 2026; originally announced July 2026.

Comments: 19 pages, 9 figures

Log-F-penalized Conditional Logistic Regression for Sparse Data

Authors: Ying Yu, Jiying Wen, Jinko Graham, Brad McNeney

Abstract: We investigate penalized likelihood methods for estimation and inference in conditional logistic regression. The standard conditional maximum likelihood estimator is known to be biased away from zero in small or sparse matched case-control studies. A widely used remedy is Firth's penalized likelihood approach, which has good frequentist operating characteristics but provides limited control over t… ▽ More We investigate penalized likelihood methods for estimation and inference in conditional logistic regression. The standard conditional maximum likelihood estimator is known to be biased away from zero in small or sparse matched case-control studies. A widely used remedy is Firth's penalized likelihood approach, which has good frequentist operating characteristics but provides limited control over the degree of shrinkage applied to individual regression coefficients. We develop point and interval estimators by penalizing the conditional likelihood with independent log-$F$ distributions. The log-\(F\)-penalized approach allows analysts to calibrate shrinkage using interpretable prior assumptions about plausible effect sizes. We also provide practical guidance for calibrating the amount of shrinkage and show that the method can be implemented through data augmentation using standard conditional logistic regression software. We illustrate the methods using data from (i) a study of maternal exposure to diethylstilbestrol and the risk of vaginal cancer in daughters, and (ii) a genetic association study of type 2 diabetes. We then compare the log-$F$-penalized approach with Firth's penalized likelihood method in a simulation study. In simulations, the log-$F$-penalized estimators had confidence-interval coverage comparable to that of Firth's method and lower mean squared error, with similar type~1 error rates and power. These results support the use of log-$F$-penalized conditional logistic regression for inference in sparse matched and stratified studies. △ Less

Submitted 30 July, 2026; originally announced July 2026.

Comments: 21 pages, 5 figures

ClawRec: A Claw-Native Recommender System

Authors: Chenghao Wu, Kesha Ou, Xiaolei Wang, Bowen Zheng, Bingqian Li, Enze Liu, Wayne Xin Zhao, Weitao Li, Long Zhang, Sheng Chen, Ji-Rong Wen

Abstract: Recommender systems have become integral to navigating the modern digital ecosystem. Yet most deployed systems remain confined within single-platform boundaries, observing localized interaction traces and ranking items from isolated candidate spaces. This design is poorly suited to real-world tasks that unfold through searches, content consumption, and comparisons across multiple information sourc… ▽ More Recommender systems have become integral to navigating the modern digital ecosystem. Yet most deployed systems remain confined within single-platform boundaries, observing localized interaction traces and ranking items from isolated candidate spaces. This design is poorly suited to real-world tasks that unfold through searches, content consumption, and comparisons across multiple information sources. Claw-style personal agents, with persistent access to authorized cross-platform context, create an opportunity for recommendation to operate around the user rather than any single platform. In this paper, we introduce Claw-native recommender systems, a new paradigm that moves beyond platform-local ranking to produce unified, complementary recommendation slates spanning diverse sources and content forms. To instantiate this paradigm, we present ClawRec, the first recommender system designed to operate natively in this environment. ClawRec maintains an evidence-linked, temporally structured user state that connects cross-platform behaviors with cross-source recommendations. It organizes retrieval around functional source roles and selects candidates according to their marginal utility, producing non-redundant slates aligned with the user's active task. To enable rigorous evaluation, we introduce ClawRec-SimBench, a benchmark constructed from sequences of concrete life events and cross-platform behavior trajectories. Experiments show that ClawRec outperforms the strongest baselines, achieving an NDCG@20 of 0.6134 (+0.1126) and a Hit@20 of 0.6944 (+0.0854), while also improving user state quality and temporal alignment. Our code and dataset are available at https://github.com/RUCAIBox/ClawRec. △ Less

Submitted 26 July, 2026; originally announced July 2026.

QuantiSpect: A Structure-Aware Lightweight 3D CNN Pre-Decoder for Scalable Surface Code Quantum Error Correction

Authors: Pan Gao, Xu-Sheng Xu, Ji-Ze Han, Jing-Wei Wen, Ling Qian, Xudong Lv, Run-Qing Zhang, Xiao-Xiao Hu, Gui-Lu Long

Abstract: Real-time decoding is a critical bottleneck for large-scale fault-tolerant quantum computing. AI-based neural pre-decoders locally correct most physical errors before passing residual syndromes to a global decoder, enabling sub-microsecond latencies. However, existing architectures carry significant overhead from dense 3D convolutions. We present QuantiSpect, a lightweight 3D convolutional neural… ▽ More Real-time decoding is a critical bottleneck for large-scale fault-tolerant quantum computing. AI-based neural pre-decoders locally correct most physical errors before passing residual syndromes to a global decoder, enabling sub-microsecond latencies. However, existing architectures carry significant overhead from dense 3D convolutions. We present QuantiSpect, a lightweight 3D convolutional neural network (CNN) pre-decoder for the rotated surface code, built on the decoding pipeline of Chamberland et al. The key idea is to replace the dense 3D convolutions with three parallel branches in each residual block: a depthwise spatial branch, a depthwise temporal branch, and a grouped spatio-temporal branch, followed by a squeeze-and-excitation channel gate. This reflects the structure of surface code errors, where spatial and temporal syndrome correlations are partially separable. On a unified 4xA100 GPU benchmark, QuantiSpect matches the receptive field of the Accurate baseline at R=13 while using ~2.71x fewer parameters (0.663M vs 1.80M) and ~2.84x fewer per-voxel convolutional MACs. It matches Accurate's circuit-level threshold and accuracy at moderate and large code distances, reduces the logical error rate by up to ~1.85x relative to uncorrelated PyMatching at d=13, p=0.5%, and speeds up the PyMatching decode by up to 3.11x at d=23. We also explored enlarging the receptive field by adding blocks. Even at R=21, the model uses only 1.18M parameters, fewer than both the R=13 Accurate baseline (1.80M) and the R=17 dense model (4.22M), despite its larger receptive field. This expanded variant significantly outperforms the Accurate model, raising the circuit-level threshold to ~0.80% and further reducing the logical error rate. Together, both variants show that a structure-aware factorized design is an effective, parameter-efficient alternative to a dense one for decoding the surface code. △ Less

Submitted 5 August, 2026; v1 submitted 20 July, 2026; originally announced July 2026.

Comments: 22 pages, 10 figures

DAUPNet: Domain-Aware Uncertainty Modeling for Reliable Prototype Discrimination in Cross-Domain Few-Shot Semantic Segmentation

Authors: Lei Yuan, Zhongxu Hu, Jingyi Wen, Pengxing Yi

Abstract: Cross-domain few-shot semantic segmentation (CD-FSS) has predominantly been formulated as learning domain-invariant representations or improving support-query correspondence. Nevertheless, large domain shifts still make prototype matching unreliable: inconsistent hierarchical responses corrupt the support representation, deterministic prototypes cannot express boundary and appearance ambiguity, an… ▽ More Cross-domain few-shot semantic segmentation (CD-FSS) has predominantly been formulated as learning domain-invariant representations or improving support-query correspondence. Nevertheless, large domain shifts still make prototype matching unreliable: inconsistent hierarchical responses corrupt the support representation, deterministic prototypes cannot express boundary and appearance ambiguity, and treating prototypes with different reliability equally during optimization weakens foreground-background separation. We therefore propose DAUPNet, a unified framework that reformulates cross-domain prototype matching as uncertainty-aware prototype discrimination. DAUPNet first harmonizes hierarchical support-query features to provide stable evidence, then represents foreground and background prototypes probabilistically, and finally uses their estimated uncertainty to regulate contrastive optimization. On four standard target domains, DAUPNet achieves 72.6% and 76.7% average mIoU in the 1-shot and 5-shot settings, respectively, including substantial gains on the two medical domains. These results demonstrate that modeling prototype uncertainty and incorporating it into optimization provides a robust and interpretable approach to CD-FSS under severe domain shift. The code is available at https://github.com/madness-Lei/DAUPNet △ Less

Submitted 14 July, 2026; originally announced July 2026.

SearchOS-V1: Towards Robust Open-Domain Information-Seeking Agent Collaboration

Authors: Yuyao Zhang, Junjie Gao, Zhengxian Wu, Jiaming Fan, Jin Zhang, Shihan Ma, Yao Yao, Weiran Qi, Chuyan Jin, Guiyu Ma, Xingzhong Xu, Kai Yang, Ji-Rong Wen, Zhicheng Dou

Abstract: Recent advances in Tool-Integrated Large Language Models have made web search a core capability of information-seeking agents. However, as interaction histories grow, agents increasingly struggle to track task progress. When search attempts fail to yield useful evidence, current single- and multi-agent systems can become trapped in repetitive loops, wasting search budgets and ultimately compromisi… ▽ More Recent advances in Tool-Integrated Large Language Models have made web search a core capability of information-seeking agents. However, as interaction histories grow, agents increasingly struggle to track task progress. When search attempts fail to yield useful evidence, current single- and multi-agent systems can become trapped in repetitive loops, wasting search budgets and ultimately compromising the quality and completeness of the final output. We introduce SearchOS, a system-level multi-agent framework that turns fragile, implicit search progress into explicit, persistent, and shared state. First, we formulate open-domain information seeking as relational schema completion with grounded citations, where agents discover entities, populate attributes across linked tables, and anchor each value to source evidence. Then we design Search-Oriented Context Management (SOCM), which externalizes the evolving state into Frontier Task, an Evidence Graph, a Coverage Map, and Failure Memory. Built on SOCM, SearchOS applies a pipeline-parallel scheduling mechanism that overlaps the execution of sub-agents and continuously refills freed slots with tasks targeting unresolved coverage gaps to improve utilization and throughput. To schedule and control the execution of search agents, SearchOS introduces a Search Tool Middleware Harness that intercepts model and tool interactions to record grounded evidence and react to stalls or budget exhaustion, and provides a reusable hierarchical skill system comprising strategy and access skills to augment the agents' search process and avoid repeating failed search patterns across runs. On WideSearch and GISA, SearchOS leads all metrics among the evaluated single- and multi-agent baselines, paving the way toward robust information-seeking collaboration. △ Less

Submitted 16 July, 2026; originally announced July 2026.

Comments: Code is available at https://github.com/antins-labs/SearchOS

Sizable Ligand-Mediated Bond-Dependent Interactions in a Spin-1 Triangular Antiferromagnet NiI$_2$

Authors: Hao Xu, Weiqin Zhu, Shufan Cheng, Yanyan Shangguan, Song Bao, Junbo Liao, Bo Zhang, Zihang Song, Shuai Dong, Maofeng Wu, Stanislav E. Nikitin, Travis J. Williams, Changsong Xu, Jinsheng Wen

Abstract: The bond-dependent anisotropic Kitaev interactions are the key for the Kitaev model, which has attracted intense interest for its potential to host quantum-spin-liquid states and fractional excitations. However, experimental realizations of such interactions remain scarce. Here, we investigate the magnetic excitations of NiI$_2$, a van der Waals magnet with spin $S=1$. By combining inelastic neutr… ▽ More The bond-dependent anisotropic Kitaev interactions are the key for the Kitaev model, which has attracted intense interest for its potential to host quantum-spin-liquid states and fractional excitations. However, experimental realizations of such interactions remain scarce. Here, we investigate the magnetic excitations of NiI$_2$, a van der Waals magnet with spin $S=1$. By combining inelastic neutron scattering, magnetization measurements, magnetic structure analysis, first-principles calculations, and linear-spin-wave simulations, we identify a minimal model that features substantial Kitaev and off-diagonal $Γ$ interactions, which together stabilize the canted magnetic ground state and open a gap in the spin-wave spectrum. Notably, these interactions arise from strong spin-orbit coupling on the ligand ions, despite the quenched orbital moment of the magnetic Ni$^{2+}$ ions. Our results provide compelling experimental evidence for the ligand-driven Kitaev mechanism. This demonstrates a concrete pathway to generating strong bond-dependent anisotropy in systems where the magnetic ions themselves have weak spin-orbit coupling, thereby substantially broadening the range of potential Kitaev materials. △ Less

Submitted 16 July, 2026; originally announced July 2026.

Comments: 8 pages, 4 figures

Belief-reality separation lives in routing over a shared value slot in language models

Authors: Oliver Steele, Jiangtao Wen, Yuxing Han

Abstract: Capable language models hold what a character believes apart from what is true: told "Anna believes the cup is blue; in reality it is red," they answer blue about Anna and red about the world. Where in the computation does that separation live? We show it rests on two separable mechanisms at two positions. A generic value slot binds the attributed value. A router at the query position selects whic… ▽ More Capable language models hold what a character believes apart from what is true: told "Anna believes the cup is blue; in reality it is red," they answer blue about Anna and red about the world. Where in the computation does that separation live? We show it rests on two separable mechanisms at two positions. A generic value slot binds the attributed value. A router at the query position selects which frame, the character's belief or reality, a query reads out. Two routes fill the slot: an asserted belief, whose value the text supplies, binds in directly; a derived belief, whose value must be inferred from what the character could see, arrives by a visibility-gated lookback. A subspace trained on either route steers the other, and only the derived route depends on described visibility. The slot itself carries no belief-reality tag: intervening on it moves a reality readout as strongly as a belief one. The separation lives instead in a dissociated pair of routing subspaces, which flip a query between frames without injecting the donor's value. These results hold across three architectures, on stimuli de-confounded against theory-of-mind-benchmark shortcuts; the behavior itself emerges between 3B and 7B across five model families. This paper develops the single belief-reality axis in depth; a companion paper shows the same slot-and-router format is shared across the other non-actual contexts a sentence can open (counterfactual, fictional, temporal). △ Less

Submitted 20 July, 2026; v1 submitted 11 July, 2026; originally announced July 2026.

Comments: 21 pages, 6 figures, 6 tables. v2: cite the released Mental Spaces Corpus (dataset DOI); switch to ACL bibliography style; minor copy-editing. No change to results

One mechanism for many mental spaces: a shared router over a value slot in language models

Authors: Oliver Steele, Jiangtao Wen, Yuxing Han

Abstract: Language builds discourse contexts other than the actual: a painting, a belief, a memory, a hypothetical. Each is a mental space in which the same entity can take a different value, as when a flower is red in reality but purple in a portrait. Formal semantics keeps these contexts apart because their logics differ (modal, temporal, doxastic, depictive). Fauconnier's mental-space theory, by contrast… ▽ More Language builds discourse contexts other than the actual: a painting, a belief, a memory, a hypothetical. Each is a mental space in which the same entity can take a different value, as when a flower is red in reality but purple in a portrait. Formal semantics keeps these contexts apart because their logics differ (modal, temporal, doxastic, depictive). Fauconnier's mental-space theory, by contrast, treats them as one space-building operation. We ask which of these a transformer language model implements, and find a mechanistic version of Fauconnier's unification. The model uses one router/slot format across the inventory: a reusable value slot stores attributed content, and a causally manipulable router (the space index) selects which space is read. A subspace trained with Distributed Alignment Search to control one space type, counterfactual, belief, fictional, or temporal, also controls the others, well above a random floor, on three model families. Belief, which formal semantics marks as a distinct case, is not specially separated. The router is low-rank, composes additively with entity identity, and acts through a few late-layer heads. Two further results show the mechanism drives inference and composes: a subspace trained on a rule-derived conclusion flips what the model infers while dissociating from what it reports, and composing space-builders mints a fresh router over the shared slot. This paper establishes the cross-type generality. A companion paper develops belief in depth, because of its special status in philosophy, psychology, and linguistics (epistemology, theory of mind, and propositional attitude reports). △ Less

Submitted 20 July, 2026; v1 submitted 11 July, 2026; originally announced July 2026.

Comments: 26 pages, 5 figures, 9 tables. v2: cite the released Mental Spaces Corpus (dataset DOI); switch to ACL bibliography style; minor copy-editing. No change to results

Robo-ValueRL: Reliable Value Estimation for Offline-to-Online Reinforcement Learning

Authors: Wenke Xia, Pei Ren, Wenbo Yu, Yizhuo Zhang, Jifan Li, Yixue Zhang, Yinuo Zhao, Qingyang Gao, Jianlong Fu, Jian Tang, Ji-Rong Wen, Zhengping Che, Di Hu

Abstract: Offline-to-online reinforcement learning is promising for generalizable robotic manipulation, yet its full-stack complexity obscures reproduction and diagnosis. Within such systems, value estimation plays a central role in prioritizing heterogeneous data for policy improvement. Despite its importance, the central question remains underexplored: how value-function reliability shapes policy optimiza… ▽ More Offline-to-online reinforcement learning is promising for generalizable robotic manipulation, yet its full-stack complexity obscures reproduction and diagnosis. Within such systems, value estimation plays a central role in prioritizing heterogeneous data for policy improvement. Despite its importance, the central question remains underexplored: how value-function reliability shapes policy optimization in offline-to-online reinforcement learning. To answer this question, we propose Robo-ValueRL, a unified framework that enables reliable value estimation and systematically traces its downstream effects on policy pretraining and online improvement. Concretely, Robo-ValueRL learns a history-conditioned value estimator and evaluates its reliability through global-progress and local-preference metrics. These resulting value estimates are propagated into quality-conditioned consistency-policy pretraining and a residual adaptation module on online rollouts, providing a unified testbed for analyzing how value reliability shapes downstream policy performance. Across 240 hours of offline demonstrations and over 3,000 online rollout trajectories, our extensive experiments show that downstream performance is strongly associated with value reliability. Reliable value functions provide better action-quality estimates, allowing value-guided offline RL to scale more effectively than quality-agnostic behavior cloning, and stabilize online improvement by prioritizing high-quality rollout data. Integrating reliable value guidance through offline pretraining with online improvement, our system achieves 86% success on millimeter-level precise chip insertion and 84% on generalizable block disassembly. We hope these findings highlight the importance of value-guided data utilization for effective policy improvement from heterogeneous robotic experience. △ Less

Submitted 10 July, 2026; originally announced July 2026.

Comments: Please refer to our website: https://gewu-lab.github.io/Robo-ValueRL/

WebSwarm: Recursive Multi-Agent Orchestration for Deep-and-Wide Web Search

Authors: Xiaoshuai Song, Liancheng Zhang, Kangzhi Zhao, Yutao Zhu, Zhongyuan Wang, Guanting Dong, Jinghan Yang, Han Li, Kun Gai, Ji-Rong Wen, Zhicheng Dou

Abstract: Large language model (LLM)-based web search agents are transforming information seeking from simple factoid question answering into complex, deep-and-wide search and research-oriented tasks. A single ReAct-style agent is constrained by one long trajectory and limited context, making it difficult to handle depth and coverage simultaneously. Existing multi-agent systems improve search coverage throu… ▽ More Large language model (LLM)-based web search agents are transforming information seeking from simple factoid question answering into complex, deep-and-wide search and research-oriented tasks. A single ReAct-style agent is constrained by one long trajectory and limited context, making it difficult to handle depth and coverage simultaneously. Existing multi-agent systems improve search coverage through parallel execution and aggregation, but still exhibit clear limitations in recursive depth, collaboration adaptability, and evidence-grounded expansion. We propose WebSwarm, a progressive recursive delegation framework that jointly constructs task decomposition, recursive expansion, and agent collaboration during inference. WebSwarm dynamically instantiates agentic search nodes, each coupling a local objective with a search mode that specifies how the node should organize search and collaboration. Each node can either solve its objective itself or further delegate child nodes; after solving, it returns evidence and results upward, enabling parent nodes to further expand, revise, or aggregate the search process. To guide this process, WebSwarm first probes how task-relevant information is organized on the web to ground subsequent node expansion, and reuses process-level experience across homogeneous sibling nodes. Experiments on BrowseComp-Plus, WideSearch, DeepWideSearch, and GISA show that WebSwarm consistently outperforms single-agent and multi-agent baselines on deep, wide, and interleaved deep-and-wide tasks. Further analyses of ablation, task difficulty, web tool efficiency, and model generalization explain WebSwarm's effectiveness and provide insights for multi-agent search systems. △ Less

Submitted 9 July, 2026; originally announced July 2026.

Comments: Work in progress

SoccerNet 2026 Challenges Results

Authors: Anthony Cioppa, Silvio Giancola, Håkan Ardö, Mohamad Dalal, Jan Held, Jérémie Ochin, Jiayuan Rao, Karen Sanchez, Renaud Vandeghen, Artur Xarles, Olivier Barnich, Albert Clapés, Mathieu Delvaux, Sergio Escalera, Bernard Ghanem, Cédric Hons, Antoine Houet, Sotiris Manitsaris, Tom Michel, Pierre Miralles, Thomas B. Moeslund, Mikael Nilsson, Bogdan Stanciulescu, Marc Van Droogenbroeck, Yanfeng Wang , et al. (80 additional authors not shown)

Abstract: The SoccerNet 2026 Challenges constitute the sixth annual edition of the SoccerNet open benchmarking effort, dedicated to advancing computer vision research in sports video understanding. This year's challenges span five vision-based tasks: (1) Ball Action Anticipation, predicting the timing and class of ball-related actions within a short future window from a preceding observation window; (2) Pla… ▽ More The SoccerNet 2026 Challenges constitute the sixth annual edition of the SoccerNet open benchmarking effort, dedicated to advancing computer vision research in sports video understanding. This year's challenges span five vision-based tasks: (1) Ball Action Anticipation, predicting the timing and class of ball-related actions within a short future window from a preceding observation window; (2) Player-Centric Ball Action Spotting, temporally localizing and classifying ball-related actions while assigning each action to the acting player through team affiliation and jersey number; (3) Novel View Synthesis, rendering images from unobserved camera poses in multi-view football scenes; (4) Spiideo SoccerNet Synloc, localizing athletes in real-world pitch coordinates from a single calibrated static-camera image; and (5) Visual Question Answering, answering multiple-choice questions about football broadcasts across text, image, and video inputs. For each task, participants were provided with annotated data, a unified evaluation protocol, and a public baseline. This edition saw broad participation, with 427 teams submitting 1,129 entries across the five tasks and 28 teams contributing reviewed technical reports. This paper describes each task and its evaluation protocol, presents the challenge leaderboards, and summarizes the leading submissions, with the aim of documenting the current state of each task as measured on held-out challenge data. △ Less

Submitted 8 July, 2026; originally announced July 2026.

Comments: 40 pages

Inverse Low-Dimensional Manifold Reconstruction Framework for Spatiotemporal Reconstruction of Compressible Physical Fields

Authors: Qiang Liu, Feng Ma, Wei Zhu, Xiyu Jia, Jianmin Xue, Jun Wen, Gaojun Fu

Abstract: Compressible physical fields are widely present in the real physical world, but current artificial intelligence lacks an understanding mechanism for the non-differentiable features in compressible physical fields. Addressing the limitations of existing deep learning architectures in handling global non-differentiable features, we propose the Inverse Low-Dimensional Manifold reconstruction framewor… ▽ More Compressible physical fields are widely present in the real physical world, but current artificial intelligence lacks an understanding mechanism for the non-differentiable features in compressible physical fields. Addressing the limitations of existing deep learning architectures in handling global non-differentiable features, we propose the Inverse Low-Dimensional Manifold reconstruction framework (ILDM). This framework couples the Non-differentiable Approximation Function (NAF) for capturing non-differentiable features in compressible flows with the Smooth Fluid Reconstruction (SFR) module tailored for smooth fluid regions. Extensive evaluations across 1D and 2D benchmarks, including Riemann problems and double Mach reflection, demonstrate that ILDM significantly outperforms cPINN and R-adaptive DeepONet. Specifically, ILDM achieves superior localization of non-differentiable interfaces and maintains robust super-resolution performance even with low-resolution inputs, establishing a physically consistent and scalable paradigm for data-driven fluid dynamics. △ Less

Submitted 8 July, 2026; originally announced July 2026.

Comments: 25 pages with 6 figures

VisTCP: A Visualization Framework to Construct Knowledge-Graph-Based Representation for Traditional Chinese Painting

Authors: Zhiguang Zhou, Fengling Zheng, Miaoxin Hu, Lina You, Jin Wen, Huan Liu, Wei Zhang, Dekun Qian, Yuhua Liu, Wei Chen, Yigang Wang, Yong Wang

Abstract: Structured representation can characterize semantic objects and relationships in images. It provides a possible effective way for the semantic understanding of Traditional Chinese Paintings (TCPs) to better support archaeology and art history research. However, most image-oriented structured representation methods perform poorly on TCPs, due to two major challenges: 1) the objects and events of TC… ▽ More Structured representation can characterize semantic objects and relationships in images. It provides a possible effective way for the semantic understanding of Traditional Chinese Paintings (TCPs) to better support archaeology and art history research. However, most image-oriented structured representation methods perform poorly on TCPs, due to two major challenges: 1) the objects and events of TCPs exhibit substantial differences from modern natural images, which results in semantic misunderstandings of TCPs; and 2) it is difficult to achieve accurate identification of ancient objects and events in TCPs, even for domain experts.In this paper, we propose VisTCP, a visualization framework that combines a TCP-oriented intelligent model and expert knowledge, which enables art historians to achieve trustworthy structured representations of TCPs in a human-in-the-loop manner. Firstly, we conduct a pilot study with three domain experts to build a semantic taxonomy of TCPs. Then, expert-annotated data are used to train a TCP-oriented structured representation model, which can automatically extract meaningful objects and their relationships in TCPs. To inform users of the model uncertainty, we design a joint embedding visualization view to show the differences between expert annotations and model predictions. This allows users to refine the structured representation based on their domain knowledge, enabling iterative optimization of the model. Finally, we conduct a case study, a usage scenario, and expert interviews on a real dataset to demonstrate the effectiveness of VisTCP in supporting the structured representation and semantic understanding of TCPs. △ Less

Submitted 7 July, 2026; originally announced July 2026.

From Stacking Disorder to Cubic Order: Ice Crystallization from Deeply Supercooled Water

Authors: Yulin Lin, Weimin Guo, Suvo Banik, Tao Zhou, Thomas E. Gage, Lei Yu, Maksim A. Sultanov, Martin Holt, Subramanian Sankaranarayanan, Peng Zhang, Ilke Arslan, Jianguo Wen

Abstract: Crystallization far from equilibrium can generate morphologies that defy classical crystal habits, yet the microscopic mechanisms linking atomic-scale disorder to emergent macroscopic order remain elusive. Here we use in situ cryogenic transmission electron microscopy with a membrane-encapsulated microdroplet platform to directly visualize the freezing of deeply supercooled water at molecular reso… ▽ More Crystallization far from equilibrium can generate morphologies that defy classical crystal habits, yet the microscopic mechanisms linking atomic-scale disorder to emergent macroscopic order remain elusive. Here we use in situ cryogenic transmission electron microscopy with a membrane-encapsulated microdroplet platform to directly visualize the freezing of deeply supercooled water at molecular resolution. We show that homogeneous nucleation produces stacking-disordered ice composed of mixed hexagonal and cubic sequences, in which cubic ice initially exists only as isolated monolayers. The gradual thickening of these cubic layers constitutes the key kinetic mechanism that governs the entire crystallization pathway. As thickening proceeds, nanoscale, defect-free cubic ice germs nucleate on the basal planes of the disordered lattice. These faceted cubic germs act as facet-registered kinetic seeds that enforce cubic twinning and sequentially multiply growth branches. This kinetic pathway reproducibly generates robust eight-branched dendrites with global cubic (octahedral) symmetry, even though each branch remains highly stacking-disordered. At later stages, latent heat release drives a crossover to the thermodynamically favored hexagonal phase; remarkably, the pre-established global cubic symmetry is retained. These results reveal how strong kinetic driving forces convert microscopic disorder into emergent macroscopic symmetry, providing a general framework for understanding and controlling rapid crystallization far from equilibrium. △ Less

Submitted 3 July, 2026; originally announced July 2026.

A unified approach to cross-intersection problems with applications to Hilton--Milner type theorems and stability

Authors: Jie Wen, Benjian Lv

Abstract: We develop a new approach to cross-intersection problems in extremal set theory. The method builds on the iterative procedure introduced by Kupavskii and Zakharov (2024) and the $t$-cover method. It provides a flexible framework for deriving extremal and stability results for cross $t$-intersecting families. Our approach applies to a variety of combinatorial objects. As an application, we prove a… ▽ More We develop a new approach to cross-intersection problems in extremal set theory. The method builds on the iterative procedure introduced by Kupavskii and Zakharov (2024) and the $t$-cover method. It provides a flexible framework for deriving extremal and stability results for cross $t$-intersecting families. Our approach applies to a variety of combinatorial objects. As an application, we prove a product version of the seminal Erdős--Ko--Rado theorem for sufficiently spread set systems. Two families $\mathcal{F}$ and $\mathcal{G}$ of $k$-subsets of $[n]$ are called cross $t$-intersecting if $|F\cap G|\geq t$ for all $F\in\mathcal{F}$ and $G\in\mathcal{G}$. We determine the families maximizing $\min\{|\mathcal{F}|, |\mathcal{G}|\}$ for large $n$ and all $t\ge2$, generalizing results of Mörs (1985) and Füredi (1995) for cross $1$-intersecting families. We then determine the families maximizing $|\mathcal{F}||\mathcal{G}|$ under the condition $\max\{|\cap_{F\in\mathcal{F}}F|,|\cap_{G\in\mathcal{G}}G|\}<t$ for large $n$. This improves the bound obtained by Frankl and Wang (2024), and provides a characterization of extremal configurations. For a family $\mathcal{F}$ of subsets of $[n]$, we introduce its $t$-diversity $γ_t(\mathcal{F})$, defined as the minimum number of sets from $\mathcal{F}$ not containing a fixed $t$-subset. This serves as a natural generalization of the important notion of diversity for $t=1$. We obtain a stability result via $γ_t$, and determine the maximum of $\min\{γ_t(\mathcal{F}),γ_t(\mathcal{G})\}$ for cross $t$-intersecting families $\mathcal{F}$ and $\mathcal{G}$. These yield new results for $t$-intersecting families, including a stability theorem towards a conjecture of Ellis, Keller and Lifshitz (2019), which may also be regarded as a $t$-intersection version, for large $n$, of an influential theorem of Frankl (1987). △ Less

Submitted 3 July, 2026; originally announced July 2026.

MSC Class: 05D05

Double-Helix Active Geometry: LiDAR-Anchored Multi-View Depth with Selective Abstention

Authors: Jinwen Wen

Abstract: Consumer depth sensors such as the LiDAR scanner on recent iPhones provide metric range, but their useful range is short and their returns are sparse. We present DH-Active, a lightweight, training-free geometry back-end that treats the sensor as a metric ruler rather than the sole source of depth. Near-field returns anchor the metric relative pose of two views through PnP; visually trackable sampl… ▽ More Consumer depth sensors such as the LiDAR scanner on recent iPhones provide metric range, but their useful range is short and their returns are sparse. We present DH-Active, a lightweight, training-free geometry back-end that treats the sensor as a metric ruler rather than the sole source of depth. Near-field returns anchor the metric relative pose of two views through PnP; visually trackable samples without a valid depth return are then triangulated under that pose. A parallax/reprojection gate abstains wherever the geometry is ill-conditioned, leaving an explicit hole and a selective score instead of forcing an estimate. The measured core front end, including spiral sampling, sparse back-projection, and hole taxonomy but excluding preprocessing and multi-view recovery, runs at 1.11 ms median latency on CPU (OpenCV using 14 threads), about 38 times faster than a DINOv2-L visual branch on GPU in our timing setup. Across two iPhone captures and the public TUM RGB-D and ARKitScenes benchmarks, held-out depth is recovered at 1.4 to 6.7 percent median relative error. In a controlled ARKitScenes protocol that uses only returns within 2 m to set scale and an independent laser scan as ground truth, DH-Active achieves 64.2 percent scene-median coverage of evaluable far-field candidates at 13.4 percent scene-median relative error; direct triangulation from the device trajectory is not usable. We also report the alternatives that failed in our tests: single-frame defocus, classical focus-stack depth, defocus-LiDAR fusion, point-to-point ICP over a good visual-inertial track, and attention-to-holes resampling. A 1.26 B learned model remains more accurate after oracle scale alignment. The contribution here is narrower: metric sparse depth, explicit abstention, zero learned parameters, and near-millisecond CPU cost. △ Less

Submitted 28 June, 2026; originally announced July 2026.

Comments: 10 pages, 3 figures, 7 tables

Hawk: Harnessing Hardware-Aware Knowledge for High-Performance NPU Kernel Generation

Authors: Junyi Wen, Ruiyan Zhuang, Yongjia Xu, Pengtu Li, Rui Zou, Hongyi Chen, Chingman Wan, Puxu Yang, Wuhui Chen, Yanlin Wang

Abstract: Developing high-performance kernels for Neural Processing Units (NPUs) is a critical industry bottleneck, requiring developers to manually navigate implicit hardware constraints and strict memory hierarchies. While large language models offer immense automation potential, they fail catastrophically on NPUs due to a fundamental lack of hardware-specific priors. Naively transplanting code snippets f… ▽ More Developing high-performance kernels for Neural Processing Units (NPUs) is a critical industry bottleneck, requiring developers to manually navigate implicit hardware constraints and strict memory hierarchies. While large language models offer immense automation potential, they fail catastrophically on NPUs due to a fundamental lack of hardware-specific priors. Naively transplanting code snippets from similar NPU kernels may pass the compiler, but it consistently triggers runtime crashes and performance degradation by blindly violating underlying hardware constraints. To overcome this, we introduce Hawk, a training-free framework that harnesses hardware-aware knowledge through three core modules: (1) Run-Time Knowledge Synthesis Module, which employs a Triple-Part Executable Knowledge Representation to inherently couple the error context with executable semantics; (2) Bottleneck-Aware Knowledge Retrieval Module, which implements a 2D-Retrieval paradigm to project queries into orthogonal syntactic and hardware-aligned semantic spaces; and (3) Effect-Driven Knowledge Distillation Module, which leverages LLM-driven semantic arbitration to continuously distill the knowledge by pruning errors and consolidating redundancies based on the empirical execution feedback. Extensive evaluations on real-world NPU workloads demonstrate that Hawk elevates generation accuracy from 49.4% to 80.0%, while achieving up to a 2.2x execution speedup over state-of-the-art baselines. △ Less

Submitted 2 July, 2026; v1 submitted 1 July, 2026; originally announced July 2026.

A Task-State Representation for Long-Horizon Mobile GUI Agents

Authors: Yujie Zheng, Zikang Liu, Xin Zhao, Ji-Rong Wen

Abstract: While long-horizon mobile GUI agents typically rely on thought-action-observation loops, they struggle to separate persistent task states from transient screen observations. As execution histories grow, this entanglement imposes a severe context burden, causing agents to forget initial requirements, hallucinate progress, or repeatedly interact with stale interfaces. To address this, we introduce T… ▽ More While long-horizon mobile GUI agents typically rely on thought-action-observation loops, they struggle to separate persistent task states from transient screen observations. As execution histories grow, this entanglement imposes a severe context burden, causing agents to forget initial requirements, hallucinate progress, or repeatedly interact with stale interfaces. To address this, we introduce Task-State Representation (TSR), a training-free framework that explicitly decouples task state from sensory input. Acting as a lightweight external wrapper, TSR maintains three structured components: a global instruction summary, a dynamic progress tracker for subgoals, and a transition-aware action verifier. By continuously updating through pre- and post-action visual comparisons, TSR effectively guides the agent's reasoning without requiring architectural modifications. Experiments across four mobile GUI benchmarks validate TSR's effectiveness, yielding up to a 12 absolute point increase in success rate on complex cross-application and memory-intensive tasks. △ Less

Submitted 1 July, 2026; originally announced July 2026.

Comments: Preprint. 9 pages, 3 figures

NeuroCogMap Reveals Cognitive Organization of Large Language Models

Authors: Zhongxiang Sun, Haolang Lu, Qiang Ma, Qi Li, Qipeng Wang, Liang Pang, Chenyu Liu, Qiankun Li, Hao Sun, Kun Wang, Yi Zeng, Jun Xu, Guoqi Li, Ji-Rong Wen

Abstract: Understanding how complex cognitive functions are organized within artificial systems is central to interpreting large language models (LLMs) and relating them to biological cognition. Yet although LLMs exhibit broad cognitive-like behaviours, it remains unclear whether their internal representations form reproducible functional systems that explain behaviour, failure and links to human cognition.… ▽ More Understanding how complex cognitive functions are organized within artificial systems is central to interpreting large language models (LLMs) and relating them to biological cognition. Yet although LLMs exhibit broad cognitive-like behaviours, it remains unclear whether their internal representations form reproducible functional systems that explain behaviour, failure and links to human cognition. Here we present NeuroCogMap, a cognitive neuroscience-inspired framework that organizes internal features of LLMs into functional parcels and links them to interpretable functions, cognitive capabilities and a cognitive hierarchy. These parcels form a stable and semantically coherent organization that is partly conserved across models and functionally linked to model outputs. Within this organization, major LLM failures, including hallucination, bias, refusal failure and sycophancy, correspond to distinct disruptions in representational and behavioural-control systems, yielding internal signatures for mechanism-guided detection and targeted intervention. Beyond model behaviour, NeuroCogMap improves prediction of human cortical responses during naturalistic language comprehension, with the strongest correspondence in higher-order association cortex. At the cognitive level, its internal signatures expose latent strategies that guide refinements of classical models of human decision-making. Together, these findings establish NeuroCogMap as a system-level framework for mapping functional organization in artificial systems and for relating this organization to human cortical function and cognitive behaviour. △ Less

Submitted 30 June, 2026; originally announced July 2026.

Comments: 79 pages, 6 main figures, 5 extended figures

Positivity-preserving dynamical low-rank methods for the Vlasov equation

Authors: Katharina Kormann, Murtazo Nazarov, Junjie Wen

Abstract: In this manuscript, we introduce positivity-preserving correction methods for low-rank approximations of the Vlasov equation. The key idea is to formulate structural properties, including positivity-preservation, as constraints and to seek a minimal correction term that is added to the low-rank solution, by solving a quadratic programming problem. As a result, the corrected solution satisfies the… ▽ More In this manuscript, we introduce positivity-preserving correction methods for low-rank approximations of the Vlasov equation. The key idea is to formulate structural properties, including positivity-preservation, as constraints and to seek a minimal correction term that is added to the low-rank solution, by solving a quadratic programming problem. As a result, the corrected solution satisfies the constraints and preserve these properties, while remaining close to the original low-rank solution. Two positivity-preserving schemes are proposed in this work, and one of them also preserves the total mass and momentum of the system. We apply the proposed methods to a Vlasov--Poisson and Vlasov--Poisson-BGK employing a spectral discretization in space and an explicit Runge--Kutta scheme in time. Numerical experiments demonstrate the effectiveness of the proposed methods. △ Less

Submitted 30 June, 2026; originally announced June 2026.

MSC Class: 35Q83; 65F55; 65K10

Detector-Conditioned Source-Space Nulls and Null-Mask Loss in a Programmable Two-Slit Interferometer

Authors: Jianming Wen

Abstract: Afshar's double-slit experiment probes wave--particle complementarity by placing a wire grid at the dark fringes of a downstream interference pattern while retaining an imaging basis that appears to preserve which-path information. Here we propose and analyze a time-reversed Young--Afshar configuration in which the corresponding null test is transferred from the downstream field plane to the sourc… ▽ More Afshar's double-slit experiment probes wave--particle complementarity by placing a wire grid at the dark fringes of a downstream interference pattern while retaining an imaging basis that appears to preserve which-path information. Here we propose and analyze a time-reversed Young--Afshar configuration in which the corresponding null test is transferred from the downstream field plane to the source-label plane of a time-reversed Young interferometer. In this reciprocal geometry, a point-addressable source illuminates a double slit, while the detector remains fixed. The observed fringe is therefore not a single-shot spatial intensity pattern, but a detector-conditioned response reconstructed by scanning the source coordinate. Consequently, a null in this pattern is not a node of a freely propagating field; it is a source label for which the coherent two-slit transfer amplitude to the selected detector vanishes. A mask placed at such source-plane labels is invisible to that detector when both slits are open, yet becomes visible when either slit is opened alone. We develop the scalar Fresnel model, derive the source-space null condition, introduce a detector-conditioned null-mask loss, and examine how this loss evolves under a tunable which-path marker. The source-space visibility and path distinguishability satisfy the standard duality relation, so no violation of complementarity is implied. The essential new feature is instead a reciprocal, detector-conditioned form of complementarity: Afshar's field-space transparency is replaced by response-function transparency in a reconstructed source basis. △ Less

Submitted 26 June, 2026; originally announced June 2026.

UAV-MapFusion: RTK-Aligned Uncertainty-Aware Coarse-to-Fine Multi-Session UAV Mapping

Authors: Feng Pan, Chunran Zheng, Bing Xue, Yukang Cui, Jiayu Wen, Zhiyu Chen, Wei Wang

Abstract: Large-scale point cloud maps are essential for robotics and spatial intelligence tasks. UAVs provide an efficient means for large-scale map acquisition; however, due to limited flight endurance and onboard storage, mapping a large-scale scene within a single flight remains difficult. Existing multi-session map merging methods can extend the mapping range, yet in UAV scenarios they still struggle t… ▽ More Large-scale point cloud maps are essential for robotics and spatial intelligence tasks. UAVs provide an efficient means for large-scale map acquisition; however, due to limited flight endurance and onboard storage, mapping a large-scale scene within a single flight remains difficult. Existing multi-session map merging methods can extend the mapping range, yet in UAV scenarios they still struggle to simultaneously suppress long-range drift and preserve local geometric accuracy. To address this issue, an uncertainty-aware multi-session point cloud map merging and coarse-to-fine optimization system is proposed. The proposed method first performs initial multi-session map merging based on a scene graph, and then incorporates RTK observations through an RTK spatiotemporal alignment module, where temporal offsets are estimated using Dynamic Time Warping (DTW), and continuous RTK constraints are recovered using Multi-Output Gaussian Processes (MOGP) under incomplete sampling and frame dropouts. On this basis, a unified uncertainty-aware factor graph is constructed, and local geometric accuracy is further improved through iterative plane-factor refinement. Experiments on real-world datasets validate the effectiveness and robustness of the proposed method. To facilitate further research and development in the community, our code and dataset will be publicly released. △ Less

Submitted 25 June, 2026; originally announced June 2026.

Comments: 8 pages, 5 figures, accepted by IEEE Robotics and Automation Letters (RA-L)

A Free Sphere Reverses the Rebound Direction of a Near-Wall Cavitation Bubble

Authors: Chun-Zhu Ren, Jun Wen, Hai-Bao Hu, A-Man Zhang, Xiao Huang

Abstract: A near-wall cavitation bubble is generally expected to acquire a wallward Kelvin-impulse bias and to rebound or jet toward the wall. Here we show that this canonical direction can be reversed by a wall-supported free sphere. High-speed imaging reveals a transition from away-from-wall to wallward rebound as the initial bubble--sphere separation is increased. By reconstructing the Kelvin impulse on… ▽ More A near-wall cavitation bubble is generally expected to acquire a wallward Kelvin-impulse bias and to rebound or jet toward the wall. Here we show that this canonical direction can be reversed by a wall-supported free sphere. High-speed imaging reveals a transition from away-from-wall to wallward rebound as the initial bubble--sphere separation is increased. By reconstructing the Kelvin impulse on a closed bubble boundary that includes both the visible free interface and the bubble-side contact closure, we find that the reversal is not governed primarily by the instantaneous velocity of the sphere. Instead, sphere displacement creates a contact closure on which the bubble-source contribution supplies an away-from-wall impulse. This contact-source impulse competes with a wallward background formed by the wall-image source and the quadrupolar component of the sphere-induced field. The resulting balance yields a calibrated geometric criterion, $\mathcal{M}_K$, and, in the comparable-size bubble--sphere regime, reduces to a contact number $a_z z_b/R_K^2$. These results identify a contact-geometric mechanism by which a movable particle can redirect the first-cycle jet and rebound bias of a near-wall cavitation bubble. △ Less

Submitted 24 June, 2026; originally announced June 2026.

Comments: 16 pages, 5 figures; includes Supplemental Material

Improved Large Language Diffusion Models

Authors: Shen Nie, Qiyang Min, Shaoxuan Xu, Zihao Huang, Yuxuan Song, Yong Shan, Yankai Lin, Wayne Xin Zhao, Chongxuan Li, Ji-Rong Wen

Abstract: Modern large language models are predominantly trained with autoregressive factorization and causal attention. We present \emph{iLLaDA}, an 8B masked diffusion language model trained from scratch with fully bidirectional attention. iLLaDA keeps the masked diffusion objective throughout pre-training and supervised fine-tuning (SFT), scaling pre-training to 12T tokens and fine-tuning on a 25B-token… ▽ More Modern large language models are predominantly trained with autoregressive factorization and causal attention. We present \emph{iLLaDA}, an 8B masked diffusion language model trained from scratch with fully bidirectional attention. iLLaDA keeps the masked diffusion objective throughout pre-training and supervised fine-tuning (SFT), scaling pre-training to 12T tokens and fine-tuning on a 25B-token instruction corpus for 12 epochs. We further use variable-length generation for efficiency and introduce confidence-based scoring for multiple-choice evaluation. Compared with LLaDA, iLLaDA improves broadly across general, mathematical, and code benchmarks; for example, iLLaDA-Base improves by 21.6 points on BBH and 14.9 points on ARC-Challenge, while iLLaDA-Instruct improves by 14.5 points on MATH and 16.5 points on HumanEval. Despite its non-autoregressive training, iLLaDA also remains competitive with Qwen2.5 7B on several benchmarks. These results show that fully bidirectional diffusion training from scratch is a competitive path toward strong language models. Model weights and codes: https://github.com/ML-GSAI/LLaDA. △ Less

Submitted 23 June, 2026; originally announced June 2026.

Self-Recognition Finetuning can Prevent and Reverse Emergent Misalignment

Authors: Arush Tagade, Shaoheng Zhou, Jiaxin Wen, Shi Feng

Abstract: Emergent misalignment (EM) has been linked to the activation of misaligned persona vectors and evil character traits, suggesting that EM operates through disruption of the model's aligned character rather than direct learning of harmful content. Motivated by this connection, we study self-generated text recognition (SGTR) finetuning as a character-targeted intervention that is distinct from existi… ▽ More Emergent misalignment (EM) has been linked to the activation of misaligned persona vectors and evil character traits, suggesting that EM operates through disruption of the model's aligned character rather than direct learning of harmful content. Motivated by this connection, we study self-generated text recognition (SGTR) finetuning as a character-targeted intervention that is distinct from existing in-training defenses. We conduct two-stage finetuning experiments across three models (GPT-4.1, Qwen2.5-32B-Instruct, Seed-OSS-36B-Instruct) and multiple EM datasets to compare SGTR finetuning against benign finetuning baselines (correct domain-specific data, general knowledge, and word counting) to find it an effective defense in both reversal and prevention settings. We find that all interventions produce comparable EM reversal, but only when restoring capabilities that EM had degraded. For prevention, only SGTR finetuning consistently reduces misalignment without exacerbating any individual metric, suggesting that character fortification specifically drives prevention. We provide further evidence for EM's relation to the LLM's default character by showing that EM finetuning induces diversity into the LLM's identity self-reports, artificially corrupting self-recognition exacerbates misalignment caused by EM finetuning, and that removing the model's identity-bearing system prompt substantially reduces the effect of EM finetuning. Together, these findings reframe EM not as the adoption of a coherent misaligned persona but as the destabilization of aligned character. △ Less

Submitted 3 June, 2026; originally announced June 2026.

Comments: 18 pages, 11 figures

PhoneBuddy: Training Open Models for Agentic Phone Use

Authors: Zhengyang Tang, Xin Lai, Pengyuan Lyu, Xinyuan Wang, Tianyi Bai, Chenxin Li, Yiduo Guo, Huawen Shen, Yuxuan Liu, Junyi Li, Zhengyao Fang, Yang Ding, Yi Zhang, Weinong Wang, Xingran Zhou, Liang Wu, Fei Tang, Sunqi Fan, Shangpin Peng, Zheng Ruan, Anran Zhang, Benyou Wang, Ji-Rong Wen, Rui Yan, Chengquan Zhang , et al. (1 additional authors not shown)

Abstract: Phones are becoming an important execution surface for general-purpose agents, but training open models for reliable phone use remains difficult because the environment that matters at deployment, real devices running real apps, is slow, stateful, side-effectful, and hard to reset or verify, while scalable mock environments only approximate real behavior. We present PhoneBuddy, a training recipe a… ▽ More Phones are becoming an important execution surface for general-purpose agents, but training open models for reliable phone use remains difficult because the environment that matters at deployment, real devices running real apps, is slow, stateful, side-effectful, and hard to reset or verify, while scalable mock environments only approximate real behavior. We present PhoneBuddy, a training recipe and open-model line for agentic phone use that combines a real-app environment with a mock-app environment, PhoneWorld, which reconstructs runnable mock apps from real GUI usage structure. PhoneBuddy first builds a shared supervised fine-tuning stage from trajectories collected in both environments, then compares real-app RL against mixed RL across both environments. Across a 150-task human evaluation on real phones spanning apps, mini-apps, and cross-app workflows, task success rate improves from 36.67\% after supervised fine-tuning to 40.67\% after real-app RL and 45.33\% after mixed RL. On AndroidWorld, the same progression rises from 60.3\% to 77.2\% to 83.2\%. These results show that mock-app training is not a replacement for real-app RL, but a complementary source of scalable, resettable, and automatically checked interaction. The gains are strongest on app and mini-app tasks, while long-horizontal cross-app workflows remain an important open challenge. △ Less

Submitted 23 June, 2026; v1 submitted 22 June, 2026; originally announced June 2026.

Code as Anchor, Memory and Metaphor as Support: Learner Experiences with Multi-View Visualizations

Authors: Naaz Sibia, Jessica Wen, Amber Richardson, Yashika Jain, Khushi Malik, Bogdan Simion, Carolina Nobre, Angela Zavaleta Bernuy, Andrew Petersen, Michael Liut

Abstract: Program visualizations are widely used to support novice programmers, yet students often ignore or resist well-designed visual scaffolds. Research on multiple external representations (MERs) offers cognitive design principles for coordinating views, but less is known about what shapes learners' engagement with available representations. We conducted a within-subjects study with 19 undergraduates… ▽ More Program visualizations are widely used to support novice programmers, yet students often ignore or resist well-designed visual scaffolds. Research on multiple external representations (MERs) offers cognitive design principles for coordinating views, but less is known about what shapes learners' engagement with available representations. We conducted a within-subjects study with 19 undergraduates who had completed CS1 and CS2. Students completed think-aloud tasks, reflective interviews, and webcam-based gaze tracking while using a multi-representational probe with synchronized code, memory, and metaphor views, and Python Tutor, across scope, while loops, and linked lists. Gaze analysis showed that students spent nearly half their time focused on code despite available visual scaffolds. Students without prior experience anchored even more heavily in code and engaged minimally with metaphor views. Interviews identified three factors shaping selective engagement: agency, as students sought control over cognitive effort rather than simply having it reduced; representational fit, as identical designs differed in whether they felt helpful or overwhelming; and legitimacy, as some students avoided metaphorical scaffolds they perceived as childish or insufficiently rigorous for university-level work. These findings suggest that multi-representational tools in computing education require attention to affective and social factors alongside cognitive design. Practical considerations include positioning visualizations as verification instruments, offering toggleable abstraction levels, and framing tools to signal disciplinary legitimacy. More broadly, the themes help explain why cognitively sound visualization tools may fail to engage the students they are designed to help. △ Less

Submitted 17 June, 2026; originally announced June 2026.

Comments: Pre-Print of a paper to be published at the International Computing Education Research (ICER) conference 2026

FAST-LIVGO: A Degeneracy-Robust LiDAR-Inertial-Visual-GNSS Fusion Odometry

Authors: Zhiyu Chen, Chunran Zheng, Jiayu Wen, XiaoLei Zhang, Jiaming Xu, Feng Pan, Yukang Cui

Abstract: Robust state estimation and mapping in long-term, large-scale, and highly dynamic environments remains a key challenge in robotics. Existing LiDAR-Inertial-Visual Odometry (LIVO) systems achieve strong local accuracy but suffer from accumulated drift over long distances and may fail in geometrically degraded or textureless scenes. Meanwhile, GNSS-aided fusion frameworks often rely on LiDAR or visu… ▽ More Robust state estimation and mapping in long-term, large-scale, and highly dynamic environments remains a key challenge in robotics. Existing LiDAR-Inertial-Visual Odometry (LIVO) systems achieve strong local accuracy but suffer from accumulated drift over long distances and may fail in geometrically degraded or textureless scenes. Meanwhile, GNSS-aided fusion frameworks often rely on LiDAR or visual odometry for state prediction and outlier rejection, making them vulnerable when odometry degenerates. To address these limitations, we propose a tightly coupled LiDAR-Inertial-Visual-GNSS fusion framework based on an Error-State Iterated Kalman Filter. An online spatiotemporal alignment module using Dynamic Time Warping is introduced for highly dynamic conditions. To better exploit GNSS precision, we develop observation models based on Doppler shifts and fixed-anchor Time-Differenced Carrier Phase, providing millimeter-level relative constraints without augmenting historical anchor states. We further design a degeneracy-aware dual-mode outlier rejection strategy that switches between LIVO-prior-guided rejection and GNSS-aided recovery according to the LIVO degeneracy level. Experiments on the public M3DGR dataset and a custom 20~m/s fixed-wing UAV dataset demonstrate that our system reduces accumulated drift and map ghosting, outperforming state-of-the-art methods in accuracy and robustness. △ Less

Submitted 23 June, 2026; v1 submitted 17 June, 2026; originally announced June 2026.

Comments: Accepted for presentation at the 2026 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2026)

Speaking the Language of Science: Toward a General-Purpose Generative Foundation Model for the Natural Sciences

Authors: Mingyang Li, Yurou Liu, Jieping Ye, Bing Su, Ji-Rong Wen, Zheng Wang

Abstract: In this report, we present LOGOS (Language Of Generative Objects in Science), a scientific generative language model that unifies heterogeneous tasks across the natural sciences within a single autoregressive framework based on a shared scientific grammar. It encodes diverse scientific objects and their spatial interactions as token sequences over a common vocabulary. By representing spatial conta… ▽ More In this report, we present LOGOS (Language Of Generative Objects in Science), a scientific generative language model that unifies heterogeneous tasks across the natural sciences within a single autoregressive framework based on a shared scientific grammar. It encodes diverse scientific objects and their spatial interactions as token sequences over a common vocabulary. By representing spatial contact and constraint patterns as discrete tokens, the model captures complex structural interactions in a purely sequential manner, without relying on explicit coordinates or geometric neural networks. This unified representation enables a wide range of downstream tasks to be formulated consistently as next-token prediction in the same grammar space, creating strong alignment between continued multi-domain pre-training and downstream objectives. Across diverse tasks, LOGOS consistently matches or outperforms domain-specific baselines, providing preliminary evidence for the feasibility of "one model fits all" in the natural sciences. We train LOGOS models at different scales (1B, 3B, and 8B parameters) and find a consistent positive correlation between model size and performance. This suggests that the future of AI for Science (AI4S) may not lie in building an independent technical stack that is separated from large language models (LLMs). Instead, it may depend on deeply aligning scientific foundation models with LLMs through shared architectures, shared training paradigms, and shared inference infrastructure, so that LLMs can truly become a new entry point for AI4S. We release the model weights and associated resources to facilitate further research. △ Less

Submitted 15 June, 2026; originally announced June 2026.

Autonomous End-to-End SOH Prediction Services for Battery Systems via Temporal-Contrastive Representation Learning

Authors: Junting Wen, Dan Li, Qihao Quan, Xiwen Wang, Hang Yang, Zhaohong Meng, Zigui Jiang, Changlin Yang, Tianle Liu, Diego Muñoz-Carpintero, Jian Lou

Abstract: Accurate state of health (SOH) estimation is a critical diagnostic service for lithium-ion battery management. However, reliance on labor-intensive manual feature engineering and opaque black-box models hinders scalable industrial deployment. To address this, we introduce TC-SOH: a modular, plug-and-play service architecture for autonomous, end-to-end SOH prediction. TC-SOH employs a temporal-cont… ▽ More Accurate state of health (SOH) estimation is a critical diagnostic service for lithium-ion battery management. However, reliance on labor-intensive manual feature engineering and opaque black-box models hinders scalable industrial deployment. To address this, we introduce TC-SOH: a modular, plug-and-play service architecture for autonomous, end-to-end SOH prediction. TC-SOH employs a temporal-contrastive mechanism and a cross-window prediction pretext task to extract degradation-relevant representations directly from raw operational data. To improve transparency, we connect model efficacy with representation diagnostics: visualization, sensitivity analysis, redundancy analysis, bidirectional probing, future-SOH probing, and temporal shuffling show that learned features overlap with selected expert descriptors while retaining additional SOH-relevant variation, and that ordered temporal context improves subsequent-SOH prediction. Across four public datasets, TC-SOH outperforms the considered physics-informed and data-driven baselines, reducing MAPE by 1.91 times and RMSE by 2.13 times. △ Less

Submitted 15 June, 2026; originally announced June 2026.

Tyler: Typed Latent Reasoning for Language Models -- When to Think, What to Compute, and How Much to Allocate

Authors: Hanyu Lin, Min Cai, Jiawei Wen, Haodi Zhang

Abstract: Chain-of-thought (CoT) prompting improves reasoning in large language models (LLMs) by externalizing intermediate computation as discrete text tokens, but this textual interface also introduces redundancy and inference overhead. Latent reasoning offers a promising alternative by carrying part of the computation in continuous representations. However, existing methods typically predefine when laten… ▽ More Chain-of-thought (CoT) prompting improves reasoning in large language models (LLMs) by externalizing intermediate computation as discrete text tokens, but this textual interface also introduces redundancy and inference overhead. Latent reasoning offers a promising alternative by carrying part of the computation in continuous representations. However, existing methods typically predefine when latent computation is invoked and how it is allocated during decoding, leaving a key problem unresolved: when to invoke latent computation, what type of computation to perform, and how much budget to allocate. We propose \textbf{Ty}ped \textbf{L}at\textbf{e}nt \textbf{R}easoning (Tyler), a typed and budget-aware framework for latent reasoning during autoregressive decoding. Tyler learns a policy that, at each decoding step, chooses between emitting a text token and switching to a latent computation module specialized for a particular reasoning function. Once invoked, an operator maps the current reasoning state into latent tokens that support global planning, local state updates, or reusable procedural abstraction. Across extensive experiments on three backbone LLMs, Tyler improves accuracy by up to 14.49 points over CoT and by up to 4.30 points over the strongest competing baseline. It further generalizes across diverse reasoning domains and achieves the best final-stage performance with the lowest forgetting. △ Less

Submitted 15 June, 2026; originally announced June 2026.

Comments: website: https://typed-latent-reasoning.github.io

Trainable Quantum Channels as Computational Primitives for Quantum Learning

Authors: Jingwei Wen, Runqing Zhang, Xusheng Xu, Xiaoxiao Hu, Jize Han, Ling Qian, Tiejun Wang, Shijie Wei, Guilu Long

Abstract: Variational quantum learning is traditionally constrained to unitary dynamics, often treating quantum channels as detrimental noise. In this work, we reformulate the quantum channels as trainable computational primitives and establish a non-unitary quantum machine learning framework grounded in open-system dynamics. We demonstrate that the outputs of channel-enhanced quantum models form a structur… ▽ More Variational quantum learning is traditionally constrained to unitary dynamics, often treating quantum channels as detrimental noise. In this work, we reformulate the quantum channels as trainable computational primitives and establish a non-unitary quantum machine learning framework grounded in open-system dynamics. We demonstrate that the outputs of channel-enhanced quantum models form a structured superposition of multiple functional components. Each component is governed by an effective observable whose spectrum can be adaptively modulated during training, a significant departure from the spectral invariance in unitary transformations. Moreover, the proposed framework generalizes conventional unitary quantum models by retaining them as a special case while introducing additional non-unitary degrees of freedom. Furthermore, we reveal that trainable quantum channels enrich the optimization geometry through ensemble-averaged gradient and additional optimization directions induced by the Kraus operators. Extensive experiments on classification tasks using trainable amplitude-damping and phase-damping channels confirm enhanced optimization dynamics and predictive performance. In addition, we experimentally validate the proposed framework through hardware inference using ten-qubit quantum models implemented on the superconducting quantum processor, confirming its practical feasibility and hardware compatibility. Our work provides a principled approach for leveraging quantum channels as trainable resources and advances the design of high-performance quantum learning architectures. △ Less

Submitted 10 August, 2026; v1 submitted 14 June, 2026; originally announced June 2026.

OneFocus: Enabling Real-World X-ray Security Screening with a Unified Vision-Language Model

Authors: Jiali Wen, Hongxia Gao, Litao Li, Yixin Chen, Kaijie Zhang, Qianyun Liu, Xiaoqin Wen

Abstract: X-ray contraband detection is critical for security in large-scale logistics and transportation, yet conventional detectors struggle to adapt to emerging contraband types and lack fundamental visual understanding. Vision-language models (VLMs) offer strong generalization but are hindered by the scarcity of high-quality X-ray image-caption data. To bridge this critical gap, we present MMXray, a met… ▽ More X-ray contraband detection is critical for security in large-scale logistics and transportation, yet conventional detectors struggle to adapt to emerging contraband types and lack fundamental visual understanding. Vision-language models (VLMs) offer strong generalization but are hindered by the scarcity of high-quality X-ray image-caption data. To bridge this critical gap, we present MMXray, a meticulously curated benchmark of 52,124 image-caption pairs spanning 28 fine-grained classes of X-ray contraband. To enrich MMXray with realistic occlusion patterns, we further introduce CleanDET, a dedicated synthesis dataset containing clean foreground contraband images from 28 categories and background images with diverse density levels, together with AnyContraSyn, a controllable synthesis method designed to operate on CleanDET. We also develop OnePipe, an extensible pipeline for systematic data curation. Built on MMXray, we propose OneFocus, a unified VLM that supports four core tasks: visual question answering, contraband localization, classification, and image understanding. OneFocus achieves state-of-the-art performance in X-ray contraband understanding and demonstrates robust cross-domain generalization, establishing a strong vision-language baseline for security screening. △ Less

Submitted 14 June, 2026; originally announced June 2026.

Comments: 17 pages, 10 figures

Temporal Difference Learning for Diffusion Models

Authors: Qizhen Ying, Yangchen Pan, Victor Adrian Prisacariu, Junfeng Wen

Abstract: Diffusion models are typically trained with objectives that focus on local denoising targets at individual time steps (or adjacent pairs), which do not enforce consistency between predictions along the denoising trajectory. This lack of cross-time consistency can degrade performance, especially for few-step samplers. We introduce a temporal difference (TD) objective that penalizes inconsistency of… ▽ More Diffusion models are typically trained with objectives that focus on local denoising targets at individual time steps (or adjacent pairs), which do not enforce consistency between predictions along the denoising trajectory. This lack of cross-time consistency can degrade performance, especially for few-step samplers. We introduce a temporal difference (TD) objective that penalizes inconsistency of the model's multi-step progress along the denoising path. By reformulating the diffusion process as a Markov reward process and casting denoising as a policy evaluation problem in reinforcement learning, we derive a unified TD approach that applies to both discrete- and continuous-time diffusion formulations. We further propose a principled sample-based reweighting method that stabilizes training. Empirically, we show that using our TD training can significantly improve sample quality measured by FID, with stronger advantages when the number of sampling steps is small, highlighting its practical utility under low-computation-budget scenarios. We provide ablation studies to justify our design choices, including pairwise loss reweighting, regularization weight, and one-step stride. Overall, our TD approach can be a general drop-in that enforces cross-time consistency and improves generation quality across different diffusion generative models. △ Less

Submitted 12 June, 2026; originally announced June 2026.

Comments: 15 pages, 4 figures. Accepted at ICML 2026

Double-Helix Vision (DH-V2): A Geometry-Based Visual Sampler for Bandwidth-Constrained Perception

Authors: Jinwen Wen

Abstract: We present Double-Helix Vision (DH), a geometry-based visual sampler that compresses 2D images into compact 1D signals using paired golden-ratio-inspired spiral trajectories. Rather than processing every pixel uniformly, DH employs two phase-shifted helices (Alpha and Beta, offset by 180 degrees) to sample the image with biologically-inspired foveation: high density at the center, sparse coverage… ▽ More We present Double-Helix Vision (DH), a geometry-based visual sampler that compresses 2D images into compact 1D signals using paired golden-ratio-inspired spiral trajectories. Rather than processing every pixel uniformly, DH employs two phase-shifted helices (Alpha and Beta, offset by 180 degrees) to sample the image with biologically-inspired foveation: high density at the center, sparse coverage at the periphery. At 4K resolution, DH achieves a 1,433x compression ratio (99.93% reduction) while preserving the geometric structure of the scene. The full perception pipeline -- including spatial mapping, temporal collision detection, and intra-frame structural disparity estimation -- runs in 0.52 ms at 1080p on CPU-only hardware, with no neural network dependencies. On CIFAR-10 at extreme sampling budgets (K=128 points per helix), DH achieves a +6.03% accuracy gain over uniform random sampling. A JSON-serializable Robotics API is provided, delivering sub-millisecond spatial perception reports in 2.7 KB packets. Code and benchmarks are available under the MIT License. △ Less

Submitted 9 June, 2026; originally announced June 2026.

Comments: 5 pages, 3 figures, 5 tables. Code and benchmarks: https://github.com/JackJ-C/double-helix-vision-tool