arXiv subjects

Xiao Yu

Publications and source records attributed to Xiao Yu.

At least 19 recordsLinked to original sources

arXiv2026

Reinforcement World Model Learning for LLM-based Agents

Large language models (LLMs) have achieved strong performance in language-centric tasks. However, in agentic settings, LLMs often struggle to anticipate action consequences and adapt to environment dynamics, highlighting the need for world-modeling capabilities in LLM-based agents. We propose Reinforcement World Model Learning (RWML), a self-supervised method that learns action-conditioned world models for LLM-based agents on textual states using sim-to-real gap rewards. Our method aligns simulated next states produced by the model with realized next states observed from the environment, encouraging consistency between internal world simulations and actual environment dynamics in a pre-trained embedding space. Unlike next-state token prediction, which prioritizes token-level fidelity (i.e., reproducing exact wording) over semantic equivalence and can lead to model collapse, our method provides a more robust training signal and is empirically less susceptible to reward hacking than LLM-as-a-judge. We evaluate our method on ALFWorld and $τ^2$ Bench and observe significant gains over the base model, despite using no expert data, task-success rewards, or strong-LLM annotations - only an off-the-shelf embedding model as a distance metric. When combined with task-success rewards, our method outperforms direct task-success reward RL by 6.9 and 5.7 points on ALFWorld and $τ^2$ Bench respectively; it further outperforms expert-data training on ALFWorld and is comparable to it on $τ^2$ Bench.

cs.CL↗

arXiv2026

CRoS: Causality Reward on Scrambled CoTs Improves Accuracy and Reasoning Faithfulness

Reinforcement learning with verifiable rewards (RLVR) has become the standard method for training language models to reason. However, RLVR typically relies solely on outcome correctness, which does not guarantee that the CoT causally contributes to the final answer. Without this connection, outcome rewards may fail to reinforce high-quality reasoning. Existing causality metrics assess CoT causality by perturbing the CoT, such as deleting reasoning steps or injecting errors, and measuring changes in the final answer. However, these perturbations inevitably alter information in each step. Consequently, high causality scores may reflect sensitivity to information changes introduced by perturbation rather than true causality. Such confounding effects further induce undesirable reasoning behaviors when these metrics are used as training rewards, including generating less informative CoTs and reducing exploration and self-correction. To address this problem, we propose CRoS (Causality Reward on Scrambled CoTs), a novel CoT causality metric that perturbs the order of reasoning steps without altering content, thereby preserving information within each step. On Reasoning Gym with Gemma-3-4B, CRoS outperforms existing causality and outcome rewards in out-of-domain accuracy, CoT quality, and CoT faithfulness. Specifically, CRoS achieves 46.8% accuracy on five unseen logic and games datasets, outperforming the strongest baseline (43.8%). It also achieves better CoT quality, consistently outperforming baselines in CoT-Pass@K from K=1 to 128, averaging 43.4% vs. 41.4%. For CoT faithfulness, CRoS achieves a hint acknowledgment score of 34.0% vs. 20.2%.

cs.CL↗

arXiv2026

Diagnosing Training Inference Mismatch in LLM Reinforcement Learning via a Zero-Mismatch Reference

Modern LLM RL systems separate rollout generation from policy optimization. These two stages are expected to produce token probabilities that match exactly. However, implementation differences can make them assign different values to the same sequence under the same model weights, inducing Training-Inference Mismatch (TIM). TIM is difficult to inspect because it is entangled with off-policy drift and common stabilization mechanisms. In this work, we isolate TIM in a zero-mismatch diagnostic setting (VeXact), and show that small token-level numerical disagreements can independently cause training collapse. We further show that TIM changes the effective optimization problem, and identify a set of remedies that could mitigate TIM. Our results suggest that TIM is not benign numerical noise, but a systems-level perturbation that should be treated as a first-order factor in analyzing LLM RL stability.

cs.LG↗

arXiv2026

STEPS: Scene Text Editing with Preserved Style Using Diffusion and Contrastive Style Encoding

We introduce Scene Text Editing with Preserved Style (STEPS), a novel diffusion model architecture for quality text replacement in images. Scene Text Editing (STE), also known as Visual Text Editing, consists of changing the textual content in an image while conserving the original style, e.g. font, colors, orientation, background, etc. STEPS advances the state of the art in STE through directed focus on improved style preservation. We introduce a style encoder for visual text that captures style independently of textual content, and a model architecture that combines the style encoder with multiple semantic conditions (target text characters encoding and rendered glyphs). STEPS achieves superior results to previous STE methods in style preservation, output readability, and subjective quality.

cs.CV↗

arXiv2026

SCOUT: Synergizing Reasoning and Tool-Use for Computer-Use Safety

Computer-use agents (CUAs), while capable of completing computer tasks in everyday and professional workflows, can cause unintended harm even under benign instructions and environments. However, detecting such harm remains challenging. First, it requires careful, task-specific reasoning: verifiers guided only by general safety criteria often overlook many important but subtle harmful behaviors. Second, it requires active investigation: past trajectory screenshots show what the agent did but not always what actually changed in the environment, so LLM-as-a-judge verifiers that rely on screenshots alone may be unable to determine the actual consequences of actions. To address these challenges, we introduce SCOUT, a two-stage agentic safety verifier that synergizes reasoning-intensive rubric generation with tool-intensive evidence gathering. First, our SCOUT rubric generator extensively reasons over the task and the agent's trajectory to determine what successful and safe execution should entail, generating task-specific completion and safety rubrics. Then, our SCOUT probing agent follows these rubrics to interact with the post-execution environment and collect grounded evidence for final safety and completion judgments. We evaluate our framework on two computer-use safety benchmarks. On AutoElicit-Bench, SCOUT achieves 75.4 unsafe F1 and 74.5 completion F1, outperforming LLM-as-a-judge verifiers and naive tool-use verifiers. SCOUT leads on OS-Blind with 76.4% unsafe detection accuracy. Test-time reflection reduces final unsafe execution rates from 30.2% to 17.2% on AutoElicit-Bench. Ablations and analysis show that tool-free rubric generation in SCOUT elicits substantially more reasoning and is crucial for safety detection across verifier backbones, especially non-frontier ones. A preliminary extension to coding tasks shows that SCOUT can support safety verification beyond computer-use.

cs.CL↗

arXiv2026

What Was Once Learned May Need to Be Unlearned: Machine Unlearning for Deprecated API Knowledge in Large Language Models

Large language models (LLMs) for code completion may generate deprecated APIs because their pre-training corpora contain code from historical library versions. Existing approaches use inference-time intervention, model editing, or machine unlearning, but multiple plausible completions make predefined replacements restrictive. Moreover, existing studies rarely verify whether models exhibit the targeted deprecated behavior or evaluate unintended changes to other APIs. We conduct a systematic empirical study of machine unlearning for deprecated API knowledge and construct MUDAPIBench, a behavior-grounded benchmark with over 7,000 model-specific instances derived from 145 deprecated-to-up-to-date API mappings across eight Python libraries. Instances are retained only when the original model generates the target deprecated API. We evaluate eight representative unlearning methods across three code LLMs on deprecated API forgetting, up-to-date API generation, preservation of other and unrelated API behaviors, general code-generation capability, and efficiency. Results show that Gradient Difference (GD) provides the best overall trade-off, suppressing deprecated APIs while preserving other capabilities with moderate computational costs. Further analyses reveal substantial variation across libraries and show that APIs deprecated after the model's training-data cutoff are harder to forget. Layer-wise analyses indicate that GD achieves effective forgetting with comparatively controlled internal changes.

cs.SE↗

arXiv2026

Keep Evaluation Fair: Detecting Data Leakage in Code Generation Benchmarks via Membership Inference Attacks

Code generation benchmarks are widely used to evaluate Large Language Models (LLMs), but benchmark data leakage into training sets can inflate performance and undermine evaluation validity. DetectLeak, a method specifically designed for code generation benchmark leakage detection, relies on perplexity scores to identify likely leaked samples. However, perplexity mainly reflects general familiarity with code patterns and may perform poorly on complex or rare samples. It also overlooks other useful signals, such as code similarity, functional correctness, and semantic representations. To address these limitations, we propose CGMIA (Code-Generation-specific Membership Inference Attack), a method for detecting leakage in code generation benchmarks. CGMIA fine-tunes a shadow model on a subset of benchmark samples to construct labeled member and non-member data. For each sample, it collects the input prompt, generated code, and reference solution, and extracts expert features, including CodeBLEU, edit distance, test pass rate, and perplexity, together with semantic features from CodeBERT embeddings. An integrated learning module combines these features to capture both surface-level memorization signals and deeper behavioral patterns, enabling a classifier to predict whether a sample was included in the target model's training set. Experiments on eight code generation benchmarks show that CGMIA outperforms eight existing membership inference methods in most cases. It also effectively detects known leaked APPS samples in StarCoder-7B's training data.

cs.SE↗

arXiv2026

APPSim-Bench: Bridging Real-world Apps and Reproducible Evaluation for Mobile GUI Agents

Mobile GUI agents can execute tasks from natural-language instructions, but their evaluation remains difficult to make both realistic and reproducible. Existing benchmarks typically trade off these goals: simplified apps lack real-world mobile complexity, whereas live commercial apps introduce uncontrolled variation from recommendations, advertisements, accounts, and changing content. We propose AppSim-Bench, which addresses this trade-off through controllable simulated apps that preserve task-relevant interaction logic while supporting deterministic evaluation. Built through a coding-agent-assisted and human-verified workflow, it contains 557 tasks across 17 high-frequency Chinese and English apps. Its controllable backend data and outcome-based verification remove major sources of environmental stochasticity, enabling reproducible cross-model comparison. Evaluating 19 GUI agents, spanning general-purpose and GUI-specialized systems, we find that autonomous mobile execution remains far from solved. The best model completes only 50.27% of tasks, and 28.55% of tasks are not solved by any agent. Further analysis shows that failures concentrate in longer workflows, numerical reasoning tasks, and inefficient trajectories marked by high action overhead and budget exhaustion. Our project is available at https://github.com/Acrab-Agentic-Labs/AppSim.

cs.AI↗

arXiv2026

How do World Models and Policies Compose in LLM Agents? A Joint Spectral and Behavioral Account

How do LLM agents come to both understand environments they act in and master tasks set within them? Through controlled experiments combining world-model training (next-state prediction) and policy training (reward maximization), we investigate this question. We dissect the resulting models through their additive parameter updates. Geometrically, we find effective world-model updates are low-rank and share an input-feature subspace with policy updates while writing to nearly orthogonal output directions, whether trained separately or sequentially. However, we find that, in projection interventions, the sequential update induces more robustness than separate policy RL when removing the world model's leading input directions, suggesting that it has learned alternative input pathways. Behaviorally, we find the sequentially trained agent explores a wider range of states and actions. Based on this, we ask: does policy training preserve world knowledge as well as it could? We probe this with training-free merging built on the geometrically motivated input basis plus an online world-model loss during policy RL, and show both improve over the untreated baseline. Our findings suggest world knowledge and task-directed ability can be learned in geometrically complementary forms, and that future post-training pipelines should consider how best to engineer the interface between them.

cs.LG↗

arXiv2026

OpenForgeRL: Train Harness-native Agents in Any Environment

Modern AI agents rely on elaborate inference harnesses such as Claude Code, Codex, and OpenClaw to drive multi-turn reasoning, tool use, and access to external systems. While powerful, these complex harnesses also make agents hard to train end-to-end with open infrastructure, whose SFT/RL stacks cannot natively express stateful, multi-process harness inference. To address this, we present OpenForgeRL, an open-source framework for training harness-based agents end-to-end in diverse environments. OpenForgeRL achieves this with a lightweight proxy that serves the harness's model calls while recording them as training data for a standard RL codebase (e.g., veRL), and a Kubernetes orchestrator that runs each rollout in its own remote container, together enabling training on any harness in any environment at scale. By decoupling training and inference, OpenForgeRL allows researchers to easily train, study, and improve agents directly in the real harnesses and environments they are deployed with. We validate our framework across diverse, complex harnesses and environments, spanning tool/claw-based agents and multimodal GUI browser- and computer-use agents. Using only hundreds to a few thousand tasks, OpenForgeClaw reaches 31.7 pass^3 and 55.9 pass@3 on ClawEval and 33.7 on QwenClawBench. OpenForgeGUI reaches 37.7 on OSWorld-Verified, 63.0 on Online-Mind2Web, and 72.3 on WebVoyager. Both outperform open baselines of similar size on nearly all benchmarks, and in the GUI setting match or surpass models several times larger. Beyond benchmarks, we analyze how harness choice (e.g., ZeroClaw, OpenClaw, Codex) and RL shape agent behavior. We find that some harnesses are substantially harder to learn than others, and that RL improves agentic reliability, such as self-verification, tool coverage, and completing multi-step plans, though critical abilities such as error recovery remain weak.

cs.AI↗

arXiv2026

Don't Use a Cannon to Kill a Fly: Lightweight Model Editing for LLMs to Correct Deprecated API Recommendations

Pre-trained or fine-tuned on large code corpora, Large Language Models (LLMs) have demonstrated strong performance in code completion tasks. However, their embedded knowledge is constrained by the timeliness of training data, which often includes code using deprecated APIs. Consequently, LLMs frequently generate deprecated APIs that will no longer be supported in future versions of third-party libraries. While retraining LLMs on updated codebases could refresh their API knowledge, this approach is computationally expensive. Recently, lightweight model editing methods have emerged to efficiently correct specific knowledge in LLMs. However, it remains unclear whether these methods can effectively update deprecated API knowledge and enable edited models to generate up-to-date APIs. To address this gap, we conduct the first systematic study applying 10 state-of-the-art model editing techniques to update deprecated API knowledge in three LLMs: Qwen2.5-Coder, CodeGemma, and DeepSeek-Coder. We introduce EDAPIBench, a dedicated benchmark featuring over 70 deprecated APIs from 8 popular Python libraries, with more than 3,000 editing instances. Our results show that the parameter-efficient fine-tuning method AdaLoRA achieves the best performance in enabling edited models to generate correct, up-to-date APIs, but falls short in Specificity (i.e., the editing influences untargeted knowledge). To resolve this, we propose AdaLoRA-L, which defines "Common API Layers" (layers within the LLMs with high importance across all APIs, storing general knowledge and excluded from editing) and restricts edits exclusively to "Specific API Layers" (layers with high importance only for the target API, storing the API-specific knowledge). Experimental results demonstrate that AdaLoRA-L significantly improves Specificity while maintaining comparable performance across other evaluation metrics.

cs.SE↗

arXiv2026

Invasion Fronts in Shifting Habitats and Competition Systems: A Hamilton-Jacobi Approach and Nonlocal Effects

We review recent developments in the study of spreading phenomena in reaction--diffusion equations arising from ecological invasion models. Motivated by the conjecture of Shigesada and Kawasaki on staged invasions, we discuss how competition systems can lead to effective scalar models with shifting habitats. We present the Hamilton--Jacobi approach for determining spreading speeds and revisit the result of {Li--Bewick--Shang--Fagan} (2014) from this perspective. We then describe the emergence of nonlocally pulled fronts when the shifting habitat connects regions of distinct positive growth rates. Recent results including the works of Lam--Yu (2022) and Lam--Nadin--Yu (2025) are surveyed. We also discuss spreading phenomena in competition systems, predator-prey systems, and the existence of various classes of entire solutions. Finally, we discuss the logarithmic correction for invasion waves in moving environments and prove a new result.

math.AP↗

arXiv2026

Orchard: An Open-Source Agentic Modeling Framework

Agentic modeling aims to transform LLMs into autonomous agents capable of solving complex tasks through planning, reasoning, tool use, and multi-turn interaction with external environments. We present Orchard, an open-source framework for scalable agentic modeling. At its core is Orchard Env, a lightweight Kubernetes-native environment service that provides reusable primitives for sandbox lifecycle management across task domains, agent harnesses, and training stages. On top of Orchard Env, we build three agentic modeling recipes. Orchard-SWE targets software engineering agents. We introduce credit-assignment supervised fine-tuning and a progression of RL signals: Balanced Adaptive Rollout (BAR) for sparse-reward optimization, on-policy distillation (OPD) and rubric-based process reward (RPR) for dense supervision, and historical experience distillation, which compresses rollouts from prior experiments into a compact value model for inference-time reranking. Built on the Qwen3.5-35B-A3B backbone, Orchard-SWE reaches 69.7% with RPR-based RL and 73.0% with value-model reranking on SWE-bench Verified, setting a new state of the art among open-source methods while approaching frontier systems over 10x larger. Orchard-GUI trains a 4B vision-language computer-use agent using only 0.4K distilled trajectories and 2.2K open-ended tasks, achieving 68.4% average success across WebVoyager, Online-Mind2Web, and DeepShop, making it the strongest open-source model while remaining competitive with proprietary systems. Orchard-Claw targets personal assistant agents. Trained with only 0.2K synthetic tasks, it achieves 59.6% pass@3 on Claw-Eval and 73.9% when paired with the stronger ZeroClaw harness. Collectively, these results demonstrate that a lightweight, open, harness-agnostic environment layer enables reusable agentic data, training recipes, and evaluation protocols across domains.

cs.AI↗

arXiv2026

When Models Meet Users: An Empirical Study of Perceptions of General LLMs and Multimodal LLMs on Hugging Face

Large language models (LLMs) have rapidly evolved from general-purpose systems to multimodal models capable of processing text, images, and audio. As both general-purpose LLMs (GLLMs) and multimodal LLMs (MLLMs) gain widespread adoption, understanding user perceptions in real-world settings becomes increasingly important. However, existing studies often rely on surveys or platform-specific data (e.g., Reddit or GitHub issues), which either constrain user feedback through predefined questions or overemphasize failure-driven, debugging-oriented discussions, thus failing to capture diverse, experience-driven, and cross-model user perspectives in practice. To address this issue, we conduct an empirical study of user discussions on Hugging Face, a major model hub with diverse models and active communities. We collect and manually annotate 662 discussion threads from 38 representative models (21 GLLMs and 17 MLLMs), and develop a three-level taxonomy to systematically characterize user concerns. Our analysis reveals that LLM access barriers, generation quality, and deployment and invocation complexity are the most prominent concerns, alongside issues such as documentation limitations and resource constraints. Based on these findings, we derive actionable implications for improving LLM ecosystem.

cs.SE↗

arXiv2026

OSWorld 2.0: Benchmarking Computer Use Agents on Long-Horizon Real-World Tasks

Existing computer-use benchmarks fail to capture the realism, complexity, and long-horizon demands of real-world computer use, limiting their ability to reveal the limitations of frontier agents. We introduce OSWorld 2.0, a benchmark of 108 long-horizon computer-use workflows across everyday and professional tasks, designed to capture complex and challenging real-world phenomena. Each task represents a realistic end-to-end workflow that takes human users a median of about 1.6 hours to complete and requires an average of 318 tool calls with Claude Opus 4.7 using maximum thinking, compared with about 30 in OSWorld 1.0. OSWorld 2.0 targets challenge phenomena that are common in real workflows yet underrepresented in prior benchmarks, spanning interaction-design challenges such as streaming interaction and dynamic environments, as well as agent-pattern challenges such as cross-source reasoning, implicit-state inference, and visual-spatial precision. Tasks are grounded in authentic input artifacts and cross-referenced against realistic stateful user profile data, and include separate safety reports auditing safety-sensitive execution. Under our primary binary-completion metric at 500 steps, Claude Opus 4.8 with maximum thinking and batched tool calls scores best but still completes only 20.6% of tasks at a 54.8% partial score; GPT-5.5 is far more token-efficient yet plateaus near 13%. These results show that current agents are still far from professional-level computer use: rather than stumbling on basic GUI control or coding, they lose track of constraints, miss information that arrives mid-task, guess rather than ask the user, and skip verification, struggling most when a task hinges on hidden state they must recover.

cs.AI↗

arXiv2026

Triospect: A Three-Dimensional Framework for Robust Statistical AI-Generated Text Detection Against Diverse Attacks

Existing AI-generated text detectors are vulnerable to attacks that manipulate textual characteristics. In this study, we propose a novel Triospect Detection Framework by using additional perspectives of content (core ideas) and expression (stylistic elements) within a given text. Experiments on two benchmarks involving 17 attacks, 12 domains, and 17 source models demonstrate that Triospect is robust against these attacks. It improves the strong baseline by a significant margin of 22.3% (AUROC) and 13% (TPR01) on the Humanize-16K after-attack subset, and by 9.1% (AUROC) and 22% (TPR01) on the adversarial RAID. This framework marks a pioneering effort in statistical methods to enhance detection reliability against attacks. We release our data and code at https://github.com/baoguangsheng/triospect.

cs.CL↗

arXiv2026

Strong decays and effective spin-symmetry-breaking corrections in excited charm-strange mesons

We study two-body pseudoscalar-emission decays of excited charm-strange mesons in heavy meson effective field theory, where phenomenological $1/m_c$ corrections are encoded as effective relative shifts between $DP$ and $D^*P$ amplitudes, referred to here as effective spin-symmetry-breaking corrections. Using $D_{s2}^*(2573)$ data to calibrate the $T(3/2^+)$ doublet, we obtain $h'=0.407\pm0.034$ and $ε_T=-0.207\pm0.109$, indicating a natural effective correction of order $20\%$. Applying this input to the $D_{s1}(2460)$ and $D_{s1}(2536)$ system, the Belle and LHCb partial-wave data constrain the mixing angle to $0^\circ<θ_P\lesssim22.0^\circ$ and $0^\circ<θ_P\lesssim14.6^\circ$, respectively, confirming that $D_{s1}(2536)$ is dominantly a $T(3/2^+)$ state with only a small $S(1/2^+)$ admixture. In the radial sector, the pure-$2S$ assignment gives $R_{2700}^{\rm LO}=0.919$, consistent with the observed $D^{*0}K^+/D^0K^+$ ratio of $D_{s1}^*(2700)$, but predicts only $Γ_{\rm ps}[D_{s0}(2590)]\simeq20$ MeV. Allowing mixing between $D_{s1}^*(2700)$ and $D_{s1}^*(2860)$, together with a relative strong phase and effective spin-symmetry-breaking corrections, substantially increases this width while preserving agreement with the vector-state widths and $R_{2700}$. This scenario further gives $R_{1,2860}=0.911$, far from the pure-$X$ leading-order value $R_{1,2860}^{\rm pure\,X}=0.242$, so the spin-one $D^*K/DK$ ratio near $2.86$ GeV offers a clear discriminator between the mixed and unmixed assignments. Overall, this scenario reduces but does not remove the $D_{s0}(2590)$ width tension, leaving room for non-pseudoscalar channels, threshold effects, or coupled-channel dynamics. Reference decay patterns for $D_{s3}^*(2860)$, $D_{s1}(2933)$, and $D_{sJ}(3040)$ are also given.

hep-ph↗

arXiv2026

Signal Temporal Logic Motion Planning via Graphs of Convex Sets

This paper investigates continuous-time motion planning under Signal Temporal Logic (STL) specifications. The goal is to generate smooth robot trajectories that satisfy high-level logical and timing requirements while respecting low-level motion constraints. To this end, we propose an efficient framework that combines timed-automata reasoning with graphs of convex sets (GCS). An STL specification is first represented by a timed automaton, which is then coupled with a convex decomposition of the configuration space to form a joint transition system encoding both task progress and region occupancy. Based on this joint transition system, the STL motion-planning problem is reformulated as a shortest-path problem over a GCS, whose solution induces a smooth Bézier-spline trajectory satisfying the STL specification, smoothness requirements, and velocity bounds. We establish the soundness of the proposed formulation and analyze its computational complexity, showing that, once the timed automaton and convex decomposition are fixed, the convex relaxation scales polynomially with the configuration-space dimension and the Bézier degree. We further develop a compact timed-automaton construction for an expressive STL fragment using dedicated templates and Boolean composition. Numerical experiments on low-dimensional benchmarks, a $3$-D quadrotor, a $30$-DoF humanoid, and a hardware experiment on a UR-3 robot arm demonstrate that the proposed method efficiently solves complex STL motion-planning problems and produces smooth executable trajectories.

cs.RO↗