arXiv subjects

Runze Li

Publications and source records attributed to Runze Li.

At least 19 recordsLinked to original sources

arXiv2026

Testing Interchangeability in LLM Agent Teams

Production multi-agent systems replace agents constantly, on the assumption that an agent filling a role is interchangeable with any other agent that can do the job. We test that assumption. Eight teams per setting are formed independently from one base model on the same tasks, each agent keeping a private notebook across ten formation episodes; we then trade role-matched agents between teams and measure what changes on held-out tasks. Against a placebo that reproduces the disruption of a roster change without changing who occupies the seat, a swap costs little in task score but raises the communication a team spends per unit of progress by 16 to 63 percent, and in Hanabi a swapped agent is more expensive than an inexperienced one, consistent with interference from conventions learned with its former partner. In Collab-Overcooked, when the agent that sets the agenda is replaced, most of the extra communication comes from the agent that stayed. Three ablations, over base models, decoding temperature and formation length, move the swap penalty alongside one other quantity: how far independently formed teams drift apart. Greedy decoding lowers both; doubling a team's history raises both. In these settings, agents are more fungible in task outcome than in coordination efficiency, with larger swap effects after longer formation histories.

cs.AI↗

arXiv2026

On tail-robust autocovariance matrix estimation for high-dimensional and potentially nonstationary time series

In this paper, we study the autocovariance matrix estimation and inference problems under heavy-tailedness, high-dimensionality, general nonlinear temporal dependence, and potentially nonstationarity of time series. We consider two types of tail-robust autocovariance matrix estimation methods: the element-wise Huber's $M$-estimator and a computationally more efficient element-wise truncated estimator. Both estimators are designed to achieve sharp error bounds in matrix max-norm. The nonasymptotic properties of these estimators are proved based on new variants of Bernstein-type inequalities under functional dependence for the potentially nonstationary processes which may be of independent interest. Moreover, we prove a high-dimensional Gaussian approximation result, as a limiting distribution, for our element-wise truncated autocovariance estimator. A Gaussian multiplier bootstrap result is also given to facilitate the practicality. Our theoretical results are nonasymptotic, which gives explicit error bounds in terms of the sample size, dimensionality, moments, and the strength of temporal dependence. Numerical evidence is provided to support our theoretical results. Finally, we illustrate the benefits of the proposed methodology for detecting change points in monthly macroeconomic data.

stat.ME↗

arXiv2026

UI-Venus-2 Technical Report

Multimodal GUI agents have emerged as a promising paradigm for digital task automation, yet transitioning from benchmark-oriented models to dependable real-world applications remains challenging due to limited environment coverage, brittle task construction, and unreliable reward verification. In this work, we present UI-Venus-2, a general-purpose foundation GUI agent designed to operate across mobile, web, and desktop environments through a unified closed-loop reasoning-action framework. To bridge the gap toward practical deployment, we jointly scale three critical dimensions: (1) Environments, expanding coverage to more than 170 multilingual mobile apps and native desktop operating systems; (2) Tasks, employing a deep-research pipeline for function-grounded instruction generation; and (3) Verification, adopting trace-level and sample-level evaluators with visual keypoints and multi-model voting to ensure reliable RL signals for training. Furthermore, we integrate safety-aware mechanisms to ensure controlled execution of consequential actions. By offering a capable, efficient, and open-source foundation, UI-Venus-2 advances the field toward more generalizable, verifiable, and self-reflective agents for real-world applications.

cs.AI↗

arXiv2026

Differentiable Parametric Simulation and Reconstruction Models in Parnassus

Parnassus is a framework for fast detector simulation and reconstruction, directly mapping truth-level particles onto reconstructed objects. Such models can be built from deep generative networks trained on paired samples, which are fit automatically to a target detector, or from parametric prescriptions of the kind used by Delphes, which are constructed by hand. We remove this asymmetry by making the parametric models fully differentiable so that their parameters can be fit to a target sample by gradient descent. We demonstrate closure by fitting a parametric model to samples from a known configuration of itself, recovering the generating parameters and characterizing the degeneracies among them, and we present a first fit to CMS full simulation. The resulting models are interpretable, inexpensive, and run in the standard Parnassus pipeline which is fully Python based and GPU enabled.

hep-ex↗

arXiv2026

(Mis)Understanding Benign Overfitting in Equity Return Prediction

Highly overparameterized models often predict well despite interpolating training data in complex domains, challenging the classical bias--variance tradeoff. We investigate whether this ``benign overfitting'' phenomenon extends to equity return prediction. Consistent with recent statistical theory, we document two key phenomena: first, a double descent pattern in the ridgeless model's prediction risk; and second, that while the optimal ridge model consistently outperforms its ridgeless counterpart, this performance gap becomes negligible at large parameter-to-observation ratios. Ultimately, however, both models fail to outperform a simple historical average. This empirical evidence aligns with our asymptotic results under the null hypothesis of zero slope coefficients, suggesting that standard equity predictors lack true forecasting power---even within highly flexible, nonlinear machine learning architectures. These findings reconcile modern and classical machine learning in asset pricing: in the absence of a true signal, they asymptotically collapse to the historical average benchmark.

stat.ML↗

arXiv2026

Video Diffusion for Satellite-based High-Dynamical-Fidelity Precipitation (HiDFiP) Field Generation

High spatiotemporal fidelity precipitation products that accurately capture storm spatial organization, propagation, and lifecycle evolution, are essential for advancing hydrometeorological research and operations at regional and global scales. Satellite products offer the only near-global precipitation observations, but they still fall short of reproducing the spatiotemporal structure of ground-based references, due largely to dynamic distortions from the inhomogeneity, intermittency, and indirectness of satellite retrievals. Here we propose a video-diffusion framework for satellite-based High-Dynamical-Fidelity Precipitation (HiDFiP) field generation beyond the space-time coverage of ground-radar, using radar-rich CONUS as a testbed. The framework performs explicit spatiotemporal modeling with IMERG as the primary source and leverages four-dimensional storm-environment information from about 40 ERA5/ERA5-Land atmospheric/land fields to compensate for the temporal information deficit inherent to satellite retrievals. We introduce an extensive metric suite to assess HiDFiP dynamical fidelity in temporal-reconstruction and spatial-transfer settings against MRMS ground-radar precipitation over CONUS. Relative to IMERG and image-wise diffusion baselines, HiDFiP accurately reproduces the storm space-time spectral characteristics; storm timing, location, and directional propagation; precipitation-event episodicity and temporal structure; precipitation-system morphology and spatial organization; and storm-track kinematics and lifecycle evolution. Transferability experiments indicate that HiDFiP generalizes reasonably well to an unseen region. This work advances a video-diffusion paradigm for satellite-based high-spatiotemporal-fidelity precipitation generation and provides an algorithmic and diagnostic foundation for long-term global radar-grade precipitation records.

physics.ao-ph↗

arXiv2026

MirrorCraft: Paired Evaluation under Hidden Rule Changes in Minecraft

With the prosperity of the large language models (LLMs), it has become an interesting topic: how do LLM-based agents work in Minecraft? Unfortunately, most existing benchmarks evaluate them under fixed game mechanics. High performance in these settings does not show whether an agent can continue making progress when familiar recipes, drops, and other rules change. In this paper, we introduce MirrorCraft, a paired benchmark for evaluating agents under hidden rule changes in Minecraft. Each Mirror world is a copy of its paired Vanilla world, with selected server-side rules modified by the corresponding datapack. Terrain, spawn, resource placement, objective, interface, and action budget remain matched within every Vanilla-Mirror pair. MirrorCraft includes five controlled biomes, six rule suites, three progression objectives, two model families, and six agent configurations under a shared Mineflayer interface. We evaluate task progress with deterministic advancement milestones and success rate and use the Rule Intervention Effect (RIE) to measure the performance change between matched Vanilla and Mirror worlds. The experiments show that hidden rule changes have strongly different effects across suites. Among the configurations evaluated without rule descriptions, ReAct achieves the highest pooled Mirror score. Providing the exact rules yields modest gains in average progress and completion across all three objectives. MirrorCraft extends Minecraft evaluation beyond fixed mechanics and provides a controlled setting for studying how agents use gameplay outcomes when the rules of the current world differ from familiar ones.

cs.AI↗

arXiv2026

MAGA: Multi-Platform Self-Fusion of GUI Agents via Structured Action Distillation

Graphical user interface (GUI) agents based on large language models are increasingly deployed across mobile, web, and desktop environments. However, existing agents are typically domain-specific, limiting the deployment and user experience. This motivates the consolidation of specialized models into a single cross-environment policy. Weight merging directly merges domain-specific experts but can corrupt executable actions under expert disagreement, while on-policy distillation (OPD) avoids conflicting teacher supervision yet still treats all response tokens equally during distillation, ignoring that action tokens are the only interface between the environment and the agent. To address this, We introduce MAGA that re-allocates training signal according to the structured action. Based on the correctness of the generated action, it suppresses unnecessary or invalid distillation signals and focuses learning on erroneous actions. Besides, a training-only hint optimizes the supervision signal provided by domain-specific teachers without changing the student input. Across two model scales, MAGA achieves the highest mean success rate, outperforming the strongest baseline by 2.0% at 8B and achieves almost the same average performance with teachers.

cs.AI↗

arXiv2026

Towards anomaly detection searches for new physics signatures including Higgs bosons with weakly supervised machine learning

The Higgs boson, with its universal coupling to mass, provides a broadly applicable portal to sectors beyond the Standard Model and is therefore a natural anchor for anomaly detection (AD) at collider experiments. The Higgs And X Anomaly Detection (HAXAD) strategy offers a principled approach to searching for such anomalies occurring in association with a Higgs boson by combining machine-learning-based feature embedding, background estimation, and weakly supervised classification. This work extends the previous HAXAD approach towards the level of maturity required for application to recorded collider data. A major addition is the introduction and comparison of two new embedding strategies, which in turn shape the background estimation and classification. In addition, a new inference framework is developed, yielding signal-agnostic and signal-specific cross section limits and thereby completing the statistical machinery needed for future AD analyses built on HAXAD. The set of investigated signal models is also significantly expanded, allowing for the evaluation of sensitivity on a much broader phase space. Improvements to the method increase signal sensitivity with respect to the original method, and when benchmarked against an example cut-based search on the same final state, HAXAD matches or exceeds the best individual cut-based limits for a wide variety of considered signal models. These developments strengthen the case for HAXAD as a viable and compelling AD-based search strategy with novel discovery potential at colliders.

hep-ex↗

arXiv2026

Rotation topological states: theory and material realization

The conventional characterization of topological materials relies on topological invariants calculated from the entire set of occupied bands. However, when a system possesses rotational symmetry, the occupied Hilbert space can be decomposed into multiple subspaces labeled by distinct rotation eigenvalues. We show that this decomposition reveals hidden topological states characterized by a novel $\mathbb{Z}_2^n$ topological invariant, where $n$ is the number of subspaces, while the conventional $\mathbb{Z}_2$ invariant may fail to detect the topology hidden in the rotation subspaces. Remarkably, time-reversal symmetry pairs conjugate rotation eigenvalues and guarantees that the two subspaces have the same $\mathbb{Z}_2$ invariants, making the topology always hidden from the conventional global invariant. We formulate the theory of rotation-subspace topology and demonstrate its material realization in bulk CsCl. Using first-principles calculations and symmetry analysis, we show that bulk CsCl, which is diagnosed as topologically trivial by the conventional approach, features a nontrivial $\mathbb{Z}_2^3$ invariant along the $\Gamma$-R path and a nontrivial $\mathbb{Z}_2^4$ invariant along the $\Gamma$-Z and M-R paths, leading to double Weyl points on the (111) and (001) surfaces, respectively. The subspace $\mathbb{Z}_2^n$ invariant proposed here serves as a necessary refinement for symmetry-protected topological phases and will facilitate the identification of a large class of topological states overlooked by existing diagnostics.

cond-mat.mtrl-sci↗

arXiv2026

Beyond Heuristic Tuning: Power-Calibrated LLM Watermarking

Logit-based watermarking is a widely used mechanism for identifying LLM generated content, yet its effectiveness is governed by a fundamental trade-off between detectability and semantic distortion. Existing analyses provide limited guidance for principled hyperparameter selection, leaving practical deployments reliant on heuristic tuning. In this work, we develop a power-calibrated statistical framework that establishes explicit quantitative relationships between watermark hyperparameters, detection power, and distortion. This characterization transforms watermark design into a guided optimization problem. Building on these results, we derive practical parameter selection procedures that achieve optimal tradeoffs under constraints. Extensive experiments across multiple language models and datasets validate the theory and demonstrate that the proposed framework consistently identifies Pareto-optimal points.

stat.ML↗

arXiv2026

Mitigating the Curse of Dimensionality in Uniform Convergence of Deep Neural Networks via Smooth Activations

This paper establishes a theoretical framework for the uniform convergence of smoothly activated deep neural network (DNN) estimators. While standard ReLU networks achieve minimax-optimal rates in the $L^2(P)$ norm for various nonparametric regression tasks, we establish a theoretical lower bound demonstrating that least-squares ReLU estimators can suffer from the curse of dimensionality in their uniform convergence behavior. Motivated by the need for reliable uniform guarantees in downstream tasks requiring worst-case reliability, we address this limitation by analyzing smoothly activated DNNs (smooth DNNs), encompassing both feedforward and residual structures. We establish novel pseudo-dimension bounds, non-asymptotic approximation guarantees, and H\"older-norm bounds for the approximators of these models. Leveraging these results, we derive non-asymptotic uniform convergence rates for smooth DNN estimators across multiple statistical contexts, including Huber, least-squares, quantile, and logistic regression. We prove that smooth DNNs can mitigate the {curse of dimensionality} in uniform convergence by adaptively exploiting the low-dimensional hierarchical composition structure of the target function. Supported by both simulation studies and a real-world application, our results position smooth DNNs as a theoretically grounded and practically viable alternative to ReLU networks for statistical learning tasks requiring uniform guarantees.

cs.LG↗

arXiv2026

ReSkill: Reconciling Skill Creation with Policy Optimization in Agentic RL

Agentic reinforcement learning (RL) enables LLM agents to improve continuously from environment rewards, yet the resulting policies do not systematically accumulate reusable strategies that generalize across tasks. Modular skills can provide such reusable strategies, yet existing skill-augmented RL methods decouple skill creation from policy optimization, risking adopting skills that conflict with the evolving policy. Inspired by Anthropic's Skill Creator, we introduce ReSkill, an RL-in-the-loop skill creation framework that reconciles skill evolution with policy learning. ReSkill exploits the group-wise structure of GRPO to naturally embed three mechanisms with only marginal additional overhead: (1) an assertion-driven skill creator that diagnoses failures from past experience and proposes conditional, trigger-based skill revisions; (2) within-group rollout sampling that enables controlled comparison of skill versions, capturing which version best supports the policy's ongoing learning; and (3) Thompson Sampling with adaptive discounting to balance exploration and exploitation in skill version selection as the policy evolves. Across several domains, ReSkill consistently outperforms existing memory and skill-based RL methods, with the largest gains on unseen tasks. Analysis of the skill lifecycle shows skills being automatically created, tested, refined, and pruned as the policy improves, demonstrating reconciled skill-policy co-evolution.

cs.AI↗

arXiv2026

StepAudio 2.5 Technical Report

Unified audio-language modeling has emerged as a prominent trend in modern speech systems, promising to bring the reasoning capabilities of large language models to auditory tasks. However, existing unified foundations often struggle to match the depth of specialized systems across automatic speech recognition (ASR), text-to-speech synthesis (TTS), and realtime spoken interaction. Bridging this gap remains an open challenge. This report presents StepAudio 2.5, a unified audio-language foundation model that matches or exceeds specialized systems across all three capabilities. Rather than treating these tasks as architecturally distinct, we operate on the premise that once text and audio share a multimodal representational space, task specialization becomes a matter of operational regimes: data construction, optimization targets, and decoding constraints. Guided by this insight, we advance the post-training paradigm from standard supervised learning to task-tailored Reinforcement Learning from Human Feedback (RLHF), using it as the primary mechanism to define complex optimization targets. We leverage this RLHF-centric alignment, alongside specialized decoding, to shape a shared backbone into three distinct operational modes. Concretely, the ASR branch advances transcription efficiency via verifiable multi-token decoding; the TTS branch achieves controllable, expressive synthesis through preference-based RLHF and context-rich supervision; and the Realtime branch realizes low-latency, persona-consistent dialogue via generative reward modeling within an RLHF framework. On standard benchmarks, StepAudio 2.5 achieves state-of-the-art results across ASR, TTS, and Realtime, demonstrating that a singular audio-language foundation can successfully internalize the distinct deployment objectives of speech understanding, generation, and live interaction.

eess.AS↗

arXiv2026

Unified Multimodal Visual Tracking with Dual Mixture-of-Experts

Multimodal visual object tracking can be divided into to several kinds of tasks (e.g. RGB and RGB+X tracking), based on the input modality. Existing methods often train separate models for each modality or rely on pretrained models to adapt to new modalities, which limits efficiency, scalability, and usability. Thus, we introduce OneTrackerV2, a unified multi-modal tracking framework that enables end-to-end training for any modality. We propose Meta Merger to embed multi-modal information into a unified space, allowing flexible modality fusion and robustness. We further introduce Dual Mixture-of-Experts (DMoE): T-MoE models spatio-temporal relations for tracking, while M-MoE embeds multi-modal knowledge, disentangling cross-modal dependencies and reducing feature conflicts. With a shared architecture, unified parameters, and a single end-to-end training, OneTrackerV2 achieves state-of-the-art performance across five RGB and RGB+X tracking tasks and 12 benchmarks, while maintaining high inference efficiency. Notably, even after model compression, OneTrackerV2 retains strong performance. Moreover, OneTrackerV2 demonstrates remarkable robustness under modality-missing scenarios.

cs.CV↗

arXiv2026

World-Value-Action Model: Implicit Planning for Vision-Language-Action Systems

Vision-Language-Action (VLA) models have emerged as a promising paradigm for building embodied agents that ground perception and language into action. However, most existing approaches rely on direct action prediction, lacking the ability to reason over long-horizon trajectories and evaluate their consequences, which limits performance in complex decision-making tasks. In this work, we introduce World-Value-Action (WAV) model, a unified framework that enables implicit planning in VLA systems. Rather than performing explicit trajectory optimization, WAV model learn a structured latent representation of future trajectories conditioned on visual observations and language instructions. A learned world model predicts future states, while a trajectory value function evaluates their long-horizon utility. Action generation is then formulated as inference in this latent space, where the model progressively concentrates probability mass on high-value and dynamically feasible trajectories. We provide a theoretical perspective showing that planning directly in action space suffers from an exponential decay in the probability of feasible trajectories as the horizon increases. In contrast, latent-space inference reshapes the search distribution toward feasible regions, enabling efficient long-horizon decision making. Extensive simulations and real-world experiments demonstrate that the WAV model consistently outperforms state-of-the-art methods, achieving significant improvements in task success rate, generalization ability, and robustness, especially in long-horizon and compositional scenarios. Code is available at https://github.com/Win-commit/WAV.

cs.RO↗

arXiv2026

The Second Challenge on Cross-Domain Few-Shot Object Detection at NTIRE 2026: Methods and Results

Cross-domain few-shot object detection (CD-FSOD) remains a challenging problem for existing object detectors and few-shot learning approaches, particularly when generalizing across distinct domains. As part of NTIRE 2026, we hosted the second CD-FSOD Challenge to systematically evaluate and promote progress in detecting objects in unseen target domains under limited annotation conditions. The challenge received strong community interest, with 128 registered participants and a total of 696 submissions. Among them, 31 teams actively participated, and 19 teams submitted valid final results. Participants explored a wide range of strategies, introducing innovative methods that push the performance frontier under both open-source and closed-source tracks. This report presents a detailed overview of the NTIRE 2026 CD-FSOD Challenge, including a summary of the submitted approaches and an analysis of the final results across all participating teams. Challenge Codes: https://github.com/ohMargin/NTIRE2026_CDFSOD.

cs.CV↗

arXiv2026

Maximum-of-Differences Test for Comparing Multivariate K-Sample Distributions

Comparing $K$-sample distributions is a fundamental problem in data science that arises in a wide variety of fields and applications. In this article, we introduce a maximum-of-differences approach to make such comparisons. Specifically, we first calculate the pairwise distances from the pooled observations of the $K$ samples. We then define the two observations as connected if their distance is less than a pre-specified threshold value. For each observation, we next calculate the ``within" and the ``between" probabilities associated with these two types of connections for the given observation, i.e., with other observations within the same sample and between the given observation and the observations in other samples. Subsequently, we propose a maximum-of-differences (MOD) test that finds the maximum value among the standardized squared differences between the ``within" and the ``between" probabilities of all observations. Accordingly, the proposed test is not only applicable to multivariate data with $K$ samples, but can also be extended to multivariate regression models. Furthermore, we obtain the covariance-adjusted (CA) version of the MOD (CA-MOD) test, which converges to the Type I extreme value distribution under some conditions. Moreover, we demonstrate the asymptotic properties of the two tests under both the null and alternative hypotheses. The performance and usefulness of the tests are illustrated via simulation studies and real examples.

stat.ME↗