arXiv subjects

Guopeng Li

Publications and source records attributed to Guopeng Li.

At least 37 records · Page 2Linked to original sources

arXiv2025

Step-GUI Technical Report

Recent advances in multimodal large language models unlock unprecedented opportunities for GUI automation. However, a fundamental challenge remains: how to efficiently acquire high-quality training data while maintaining annotation reliability? We introduce a self-evolving training pipeline powered by the Calibrated Step Reward System, which converts model-generated trajectories into reliable training signals through trajectory-level calibration, achieving >90% annotation accuracy with 10-100x lower cost. Leveraging this pipeline, we introduce Step-GUI, a family of models (4B/8B) that achieves state-of-the-art GUI performance (8B: 80.2% AndroidWorld, 48.5% OSWorld, 62.6% ScreenShot-Pro) while maintaining robust general capabilities. As GUI agent capabilities improve, practical deployment demands standardized interfaces across heterogeneous devices while protecting user privacy. To this end, we propose GUI-MCP, the first Model Context Protocol for GUI automation with hierarchical architecture that combines low-level atomic operations and high-level task delegation to local specialist models, enabling high-privacy execution where sensitive data stays on-device. Finally, to assess whether agents can handle authentic everyday usage, we introduce AndroidDaily, a benchmark grounded in real-world mobile usage patterns with 3146 static actions and 235 end-to-end tasks across high-frequency daily scenarios (8B: static 89.91%, end-to-end 52.50%). Our work advances the development of practical GUI agents and demonstrates strong potential for real-world deployment in everyday digital interactions.

cs.CV↗

arXiv2025

Astraea: A State-Aware Scheduling Engine for LLM-Powered Agents

Large Language Models (LLMs) are increasingly being deployed as intelligent agents. Their multi-stage workflows, which alternate between local computation and calls to external network services like Web APIs, introduce a mismatch in their execution pattern and the scheduling granularity of existing inference systems such as vLLM. Existing systems typically focus on per-segment optimization which prevents them from minimizing the end-to-end latency of the complete agentic workflow, i.e., the global Job Completion Time (JCT) over the entire request lifecycle. To address this limitation, we propose Astraea, a service engine designed to shift the optimization from local segments to the global request lifecycle. Astraea employs a state-aware, hierarchical scheduling algorithm that integrates a request's historical state with future predictions. It dynamically classifies requests by their I/O and compute intensive nature and uses an enhanced HRRN policy to balance efficiency and fairness. Astraea also implements an adaptive KV cache manager that intelligently handles the agent state during I/O waits based on the system memory pressure. Extensive experiments show that Astraea reduces average JCT by up to 25.5\% compared to baseline methods. Moreover, our approach demonstrates strong robustness and stability under high load across various model scales.

cs.CL↗

arXiv2025

Shallow-water convergence of the intermediate long wave equation in $L^2$

We continue our study on the convergence issue of the intermediate long wave equation (ILW) on both the real line and the circle. In particular, we establish convergence of the scaled ILW dynamics to that of the Korteweg-de Vries equation (KdV) in the shallow-water limit at the $L^2$-level. Together with the recent work by the first three authors and D. Pilod (2024) on the deep-water convergence in $L^2$, this work completes the well-posedness and convergence study of ILW on both geometries within the $L^2$-framework. Our proof equally applies to both geometries and is based on the following two ingredients: the complete integrability of ILW and the normal form method. More precisely, by making use of the Lax pair structure and the perturbation determinant for ILW, recently introduced by Harrop-Griffths, Killip, and Vişan (2025), we first establish weakly uniform (in small depth parameters) equicontinuity in $L^2$ of solutions to the scaled ILW, providing a control on the high frequency part of solutions. Then, we treat the low frequency part by implementing a perturbative argument based on an infinite iteration of normal form reductions for KdV.

math.AP↗

arXiv2025

Global well-posedness of the energy-critical stochastic Hartree nonlinear wave equation

We consider the Cauchy problem for the stochastic Hartree nonlinear wave equations (SHNLW) with a cubic convolution nonlinearity and an additive stochastic forcing on the Euclidean space. Our goal in this paper is two-fold. (i) We study the defocusing energy-critical SHNLW on $\mathbb{R}^d$, for $d \geq 5$, and prove that they are globally well-posed with deterministic initial data in the energy space. (ii) Next, we consider the well-posedness of the defocusing energy-critical SHNLW with randomized initial data below the energy space. In particular, when $d=5$, we prove it is almost surely globally well-posed. As a byproduct, by removing the stochastic forcing our result covers the study of the (deterministic) Hartree nonlinear wave equation (HNLW) with randomized initial data below the energy space. The main ingredients in the globalization argument involve the probabilistic perturbation approach by Bényi-Oh-Pocovnicu (2015) and Pocovnicu (2017), time integration by parts trick of Oh-Pocovnicu (2016), and an estimate of the Hartree potential energy.

math.AP↗

arXiv2025

Probabilistic well-posedness of dispersive PDEs beyond variance blowup I: Benjamin-Bona-Mahony equation

We investigate a possible extension of probabilistic well-posedness theory of nonlinear dispersive PDEs with random initial data beyond variance blowup. As a model equation, we study the Benjamin-Bona-Mahony equation (BBM) with Gaussian random initial data. By introducing a suitable vanishing multiplicative renormalization constant on the initial data, we show that solutions to BBM with the renormalized Gaussian random initial data beyond variance blowup converge in law to a solution to the stochastic BBM forced by the derivative of a spatial white noise. By considering alternative renormalization, we show that solutions to the renormalized BBM with the frequency-truncated Gaussian initial data converges in law to a solution to the linear stochastic BBM with the full Gaussian initial data, forced by the derivative of a spatial white noise. This latter result holds for the Gaussian random initial data of arbitrarily low regularity. We also establish analogous results for the stochastic BBM forced by a fractional derivative of a space-time white noise.

math.AP↗

arXiv2025

Optimal divergence rate of the focusing Gibbs measures

We study Gibbs measures on the $d$-dimensional torus with $L^2$-(super)critical focusing interaction potentials. We establish a precise divergence rate of the partition function as we remove regularization, where the optimal constant is given by (i) (the negative of) the minimum value of the Hamiltonian given an $L^2$-constraint in the $L^2$-critical case and (ii) the optimal constant for certain Bernstein's inequality in the mass-supercritical case. In particular, our result in the $L^2$-critical case precisely quantifies the phase transition of the focusing Gibbs measure at the critical $L^2$ threshold, previously studied by Lebowitz, Rose, and Speer (1988) and Sosoe, Tolomeo, and the fourth author (2022).

math.PR↗

arXiv2025

NextStep-1: Toward Autoregressive Image Generation with Continuous Tokens at Scale

Prevailing autoregressive (AR) models for text-to-image generation either rely on heavy, computationally-intensive diffusion models to process continuous image tokens, or employ vector quantization (VQ) to obtain discrete tokens with quantization loss. In this paper, we push the autoregressive paradigm forward with NextStep-1, a 14B autoregressive model paired with a 157M flow matching head, training on discrete text tokens and continuous image tokens with next-token prediction objectives. NextStep-1 achieves state-of-the-art performance for autoregressive models in text-to-image generation tasks, exhibiting strong capabilities in high-fidelity image synthesis. Furthermore, our method shows strong performance in image editing, highlighting the power and versatility of our unified approach. To facilitate open research, we will release our code and models to the community.

cs.CV↗

arXiv2025

Step1X-Edit: A Practical Framework for General Image Editing

In recent years, image editing models have witnessed remarkable and rapid development. The recent unveiling of cutting-edge multimodal models such as GPT-4o and Gemini2 Flash has introduced highly promising image editing capabilities. These models demonstrate an impressive aptitude for fulfilling a vast majority of user-driven editing requirements, marking a significant advancement in the field of image manipulation. However, there is still a large gap between the open-source algorithm with these closed-source models. Thus, in this paper, we aim to release a state-of-the-art image editing model, called Step1X-Edit, which can provide comparable performance against the closed-source models like GPT-4o and Gemini2 Flash. More specifically, we adopt the Multimodal LLM to process the reference image and the user's editing instruction. A latent embedding has been extracted and integrated with a diffusion image decoder to obtain the target image. To train the model, we build a data generation pipeline to produce a high-quality dataset. For evaluation, we develop the GEdit-Bench, a novel benchmark rooted in real-world user instructions. Experimental results on GEdit-Bench demonstrate that Step1X-Edit outperforms existing open-source baselines by a substantial margin and approaches the performance of leading proprietary models, thereby making significant contributions to the field of image editing.

cs.CV↗

arXiv2025

Fuse Before Transfer: Knowledge Fusion for Heterogeneous Distillation

Most knowledge distillation (KD) methodologies predominantly focus on teacher-student pairs with similar architectures, such as both being convolutional neural networks (CNNs). However, the potential and flexibility of KD can be greatly improved by expanding it to novel Cross-Architecture KD (CAKD), where the knowledge of homogeneous and heterogeneous teachers can be transferred flexibly to a given student. The primary challenge in CAKD lies in the substantial feature gaps between heterogeneous models, originating from the distinction of their inherent inductive biases and module functions. To this end, we introduce an assistant model as a bridge to facilitate smooth feature knowledge transfer between heterogeneous teachers and students. More importantly, within our proposed design principle, the assistant model combines the advantages of cross-architecture inductive biases and module functions by merging convolution and attention modules derived from both student and teacher module functions. Furthermore, we observe that heterogeneous features exhibit diverse spatial distributions in CAKD, hindering the effectiveness of conventional pixel-wise mean squared error (MSE) loss. Therefore, we leverage a spatial-agnostic InfoNCE loss to align features after spatial smoothing, thereby improving the feature alignments in CAKD. Our proposed method is evaluated across some homogeneous model pairs and arbitrary heterogeneous combinations of CNNs, ViTs, and MLPs, achieving state-of-the-art performance for distilled models with a maximum gain of 11.47% on CIFAR-100 and 3.67% on ImageNet-1K. Our code and models will be released.

cs.CV↗

arXiv2025

You Think, You ACT: The New Task of Arbitrary Text to Motion Generation

Text to Motion aims to generate human motions from texts. Existing settings rely on limited Action Texts that include action labels, which limits flexibility and practicability in scenarios difficult to describe directly. This paper extends limited Action Texts to arbitrary ones. Scene texts without explicit action labels can enhance the practicality of models in complex and diverse industries such as virtual human interaction, robot behavior generation, and film production, while also supporting the exploration of potential implicit behavior patterns. However, newly introduced Scene Texts may yield multiple reasonable output results, causing significant challenges in existing data, framework, and evaluation. To address this practical issue, we first create a new dataset HUMANML3D++ by extending texts of the largest existing dataset HUMANML3D. Secondly, we propose a simple yet effective framework that extracts action instructions from arbitrary texts and subsequently generates motions. Furthermore, we also benchmark this new setting with multi-solution metrics to address the inadequacies of existing single-solution metrics. Extensive experiments indicate that Text to Motion in this realistic setting is challenging, fostering new research in this practical direction.

cs.CV↗

arXiv2025

On the deep-water and shallow-water limits of the intermediate long wave equation from a statistical viewpoint

(Due to the limit on the number of characters for an abstract set by arXiv, the full abstract can not be displayed here. See the abstract in the paper.) We study convergence problems for the intermediate long wave equation (ILW), with the depth parameter $δ> 0$, in the deep-water limit ($δ\to \infty$) and the shallow-water limit ($δ\to 0$) from a statistical point of view. In particular, we establish convergence of invariant Gibbs dynamics for ILW in both the deep-water and shallow-water limits. For this purpose, we first construct the Gibbs measures for ILW, $0 < δ< \infty$. As they are supported on distributions, a renormalization is required. With the Wick renormalization, we carry out the construction of the Gibbs measures for ILW. We then prove that the Gibbs measures for ILW converge in total variation to that for the Benjamin-Ono equation (BO) in the deep-water limit. In the shallow-water regime, after applying a scaling transformation, we prove that, as $δ\to 0$, the Gibbs measures for the scaled ILW converge weakly to that for the Korteweg-de Vries equation (KdV). We point out that this second result is of particular interest since the Gibbs measures for the scaled ILW and KdV are mutually singular (whereas the Gibbs measures for ILW and BO are equivalent). We also discuss convergence of the associated dynamical problem. Lastly, we point out that our results also apply to the generalized ILW equation in the defocusing case, converging to the generalized BO in the deep-water limit and to the generalized KdV in the shallow-water limit. In the non-defocusing case, however, our results can not be extended to a nonlinearity with a higher power due to the non-normalizability of the corresponding Gibbs measures.

math.AP↗

arXiv2025

Global well-posedness of the energy-critical stochastic nonlinear Schrödinger equation on the three-dimensional torus

We study the Cauchy problem of the defocusing energy-critical stochastic nonlinear Schrödinger equation (SNLS) on the three dimensional torus, forced by an additive noise. We adapt the atomic spaces framework in the context of the energy-critical nonlinear Schrödinger equation, and employ probabilistic perturbation arguments in the context of stochastic PDEs, establishing the global well-posedness of the defocusing energy-critical quintic SNLS in the energy space. It is the first global well-posedness result for the periodic SNLS in a critical space.

math.AP↗

arXiv2025

Compact $T(1)$ theorem à la Stein

We prove a compact $T(1)$ theorem, involving quantitative estimates, analogous to the quantitative classical $T(1)$ theorem due to Stein. We also discuss the $C_c^\infty$-to-$CMO$ mapping properties of non-compact Calderón-Zygmund operators.

math.FA↗

arXiv2025

Unconditional deep-water limit of the intermediate long wave equation in low-regularity

In this paper, we establish the unconditional deep-water limit of the intermediate long wave equation (ILW) to the Benjamin-Ono equation (BO) in low-regularity Sobolev spaces on both the real line and the circle. Our main tool is new unconditional uniqueness results for ILW in $H^s$ when $s_0<s\leq \frac 14$ on the line and $s_0<s< \frac 12$ on the circle, where $s_0 = 3-\sqrt{33/4}\approx 0.1277$. Here, we adapt the strategy of Moşincat-Pilod (2023) for BO to the setting of ILW by viewing ILW as a perturbation of BO and making use of the smoothing property of the perturbation term.

math.AP↗

arXiv2024

Energy Score-based Pseudo-Label Filtering and Adaptive Loss for Imbalanced Semi-supervised SAR target recognition

Automatic target recognition (ATR) is an important use case for synthetic aperture radar (SAR) image interpretation. Recent years have seen significant advancements in SAR ATR technology based on semi-supervised learning. However, existing semi-supervised SAR ATR algorithms show low recognition accuracy in the case of class imbalance. This work offers a non-balanced semi-supervised SAR target recognition approach using dynamic energy scores and adaptive loss. First, an energy score-based method is developed to dynamically select unlabeled samples near to the training distribution as pseudo-labels during training, assuring pseudo-label reliability in long-tailed distribution circumstances. Secondly, loss functions suitable for class imbalances are proposed, including adaptive margin perception loss and adaptive hard triplet loss, the former offsets inter-class confusion of classifiers, alleviating the imbalance issue inherent in pseudo-label generation. The latter effectively tackles the model's preference for the majority class by focusing on complex difficult samples during training. Experimental results on extremely imbalanced SAR datasets demonstrate that the proposed method performs well under the dual constraints of scarce labels and data imbalance, effectively overcoming the model bias caused by data imbalance and achieving high-precision target recognition.

cs.CV↗

arXiv2024

Global well-posedness of the energy-critical stochastic nonlinear wave equations

We consider the Cauchy problem for the defocusing energy-critical stochastic nonlinear wave equations (SNLW) with an additive stochastic forcing on $\mathbb{R}^{d}$ and $\mathbb{T}^{d}$ with $d \geq 3$. By adapting the probabilistic perturbation argument employed in the context of the random data Cauchy theory by Bényi-Oh-Pocovnicu (2015) and Pocovnicu (2017) and in the context of stochastic PDEs by Oh-Okamoto (2020), we prove global well-posedness of the defocusing energy-critical SNLW. In particular, on $\mathbb{T}^d$, we prove global well-posedness with the stochastic forcing below the energy space.

math.AP↗

arXiv2024

Deep-water and shallow-water limits of the intermediate long wave equation

In this paper, we study the low regularity convergence problem for the intermediate long wave equation (ILW), with respect to the depth parameter $δ>0$, on the real line and the circle. As a natural bridge between the Korteweg-de Vries (KdV) and the Benjamin-Ono (BO) equations, the ILW equation is of physical interest. We prove that the solutions of ILW converge in the $H^s$-Sobolev space for $s>\frac12$, to those of BO in the deep-water limit (as $δ\to\infty$), and to those of KdV in the shallow-water limit (as $δ\to 0$). This improves previous convergence results by Abdelouhab, Bona, Felland, and Saut (1989), which required $s>\frac32$ in the deep-water limit and $s\geq2$ in the shallow-water limit. Moreover, the convergence results also apply to the generalised ILW equation, i.e.~with nonlinearity $\partial_x (u^k)$ for $k\geq 2$. Furthermore, this work gives the first convergence results of generalised ILW solutions on the circle with regularity $s\geq \frac34$. Overall, this study provides mathematical insights for the behaviour of the ILW equation and its solutions in different water depths, and has implications for predicting and modelling wave behaviour in various environments.

math.AP↗

arXiv2024

Deep-water limit of the intermediate long wave equation in $L^2$

We study the well-posedness issue of the intermediate long wave equation (ILW) on both the real line and the circle. By applying the gauge transform for the Benjamin-Ono equation (BO) and adapting the $L^2$ well-posedness argument for BO by Molinet and the fourth author (2012), we prove global well-posedness of ILW in $L^2$ on both the real line and the circle. In the periodic setting, this provides the first low regularity well-posedness of ILW. We then establish convergence of the ILW dynamics to the BO dynamics in the deep-water limit at the $L^2$-level.

math.AP↗