arXiv subjects

Xiaowen Li

Publications and source records attributed to Xiaowen Li.

At least 19 recordsLinked to original sources

arXiv2026

SequenceO1: End-to-End Ultra-Long (100K) Sequence Modeling in Recommendation with Low-Rank Caching

Modeling long-term user behavior is central to sequential recommendation and billion-scale industrial recommender systems, yet production ranking models operate under strict latency, memory, communication, and training-throughput constraints. At the 100K scale, the challenge extends beyond attention complexity: raw sequence features must be stored, transferred, and repeatedly processed during training and online serving. Existing approaches based on history truncation, multi-stage behavior retrieval, compressed lifelong histories, or train-short/infer-long extrapolation either weaken end-to-end optimization or retain substantial length-dependent cost. We present SequenceO1, an end-to-end framework for ultra-long user behavior sequence modeling, deployed at full traffic on Douyin with histories of up to 100K interactions. SequenceO1 follows a compress-then-reason design. Its Sketch Attention (SA) uses learnable prototypes and prototype-wise normalization to compress the raw history into a fixed-size, target-agnostic user representation. Target-conditioned Stacked Target-to-History Cross Attention (STCA) then models complementary time scales: a recent 10K suffix for short-term interests and the compact sketch for long-term preferences. To make training and inference practical, SequenceO1 combines low-rank user representation caching, multi-request user-level batching, pipeline lift, and a fused FlashSA kernel to amortize feature storage, communication, and computation across targets, training instances, and consecutive requests. Production experiments show consistent offline and online gains, while the compact cached sketch retains most of the benefit of directly scaling end-to-end sequence ranking to 100K. These results provide a practical model-system approach to efficient attention, sequence compression, and scalable long-sequence and long-context recommendation systems.

cs.IR↗

arXiv2026

Adversarial Update-Based Federated Unlearning for Poisoned Model Recovery

Federated learning (FL) is vulnerable to poisoning attacks, where malicious clients upload manipulated updates to degrade the performance of the global model. Although detection methods can identify and remove malicious clients, the model remains affected. Retraining from scratch is effective but costly, and existing unlearning methods remain unsatisfactory in both effectiveness and efficiency. We propose Federated Adversarial Unlearning (FAUN), a lightweight framework that retains only a short window of malicious clients' updates and employs adversarial optimization on a proxy dataset to derive updates that eliminate malicious directions. Applying these updates for a few unlearning rounds, followed by benign fine-tuning, enables fast removal of malicious effects and stable recovery. Experiments on three canonical datasets show that FAUN achieves recovery comparable to retraining while requiring far fewer rounds and reduces attack success rates to near zero, confirming FAUN successfully eliminates the contributions of unlearned clients.

cs.LG↗

arXiv2026

Disciplined Diffusion: Text-to-Image Diffusion Model against NSFW Generation

Text-to-image (T2I) diffusion models have the ability to build high-quality pictures from text prompts, but they pose safety concerns because they can generate offensive or disturbing imagery when provided with harmful inputs. Existing safety filters typically rely on text-based classifiers or image-based checkers that completely block the output upon detecting a threat, issuing an explicit allow/block feedback signal to the user. This binary strategy leaves models vulnerable to adversarial attacks that alter keywords to bypass detection, and it causes high false-alarm rates that degrade the experience for benign users. To address such vulnerabilities, we propose Disciplined Diffusion (DDiffusion), a novel robust text-to-image diffusion that counters Not Safe For Work (NSFW) generation by uncovering implicit malicious semantics in prompt embeddings. DDiffusion leverages a semantic retrieval mechanism to evaluate prompts against concept distributions rather than relying on brittle pairwise similarity. Furthermore, it employs a localization method during the diffusion process to selectively edit only the harmful regions of the generated image. By returning locally sanitized images instead of applying uniform blocking, DDiffusion suppresses malicious content while preserving generation fidelity for benign prompts and avoiding the binary allow-deny signal on which existing probing attacks rely.

cs.CV↗

arXiv2026

Convergence to shock profiles for Burgers equation with singular fast-diffusion and boundary effect

In this paper, we study the asymptotic stability of viscous shock profile for the Burgers equation $u_t +f(u)_x = (\frac{u_{x}}{u^{1-m}})_x$ on the half-space $(0,+\infty)$, subject to the boundary conditions $u|_{x=0}=u_->0$ and $u|_{x=+\infty}=0$. Here, the parameter $\frac{1}{2}<m<1$ measures the strength of fast diffusion. A key challenge arises from the pronounced singularity in the diffusivity $\left(\frac{u_x}{u^{1-m}} \right)_x$ at $u=0$ and the boundary layer. We demonstrate that the long-time behavior of $u$ converges to a shifted shock profile $U(x-st-d(t))$, where $d(t)$ is governed by the boundary layer dynamics at $x=0$ and driven by the initial data $u(x,0)$. To overcome the singularity from fast diffusion compounded by the bad effect of boundary layer for wave stability, some new techniques for weighted energy estimates are introduced artfully.

math.AP↗

arXiv2025

Quenching of polarized jets

Jets produced in association with a $Z^{0}$ or $W^{\pm}$ boson in hadronic collisions are automatically polarized due to the parity violation of weak interaction, making these processes ideal for understanding the spin transfer from polarized partons to polarized hadrons. Furthermore, leveraging this feature, we can also employ the weak-boson-tagged process to study the quenching phenomenon of polarized jets, an aspect that has rarely been discussed in the literature. In this work, we compute the polarization of $\Lambda$ hyperons in $pp$ collisions and investigate the nuclear modification due to the jet-medium interaction in $AA$ collisions. Our results highlight this process as a valuable probe of polarized parton hadronization and of spin-dependent medium effects in the QGP.

hep-ph↗

arXiv2025

Make It Long, Keep It Fast: End-to-End 10K Long User Behavior Sequence Modeling for Billion-Scale Douyin Recommendation

Short-video recommenders such as Douyin must exploit extremely long user behavior histories without breaking latency or cost budgets. We present an end-to-end industrial recommender system that scales long-sequence recommendation modeling to 10K-length histories in production. First, we introduce Stacked Target-to-History Cross Attention (STCA), which replaces history self-attention with stacked cross-attention from the target to the history, reducing complexity from quadratic to linear in sequence length and enabling efficient end-to-end training over long user behavior sequences. Second, we propose Request Level Batching (RLB), a user-centric batching scheme that aggregates multiple targets for the same user/request to share the user-side encoding, substantially lowering sequence-related storage, communication, and compute without changing the learning objective. Third, we design a length-extrapolative training strategy -- train on shorter windows, infer on much longer ones -- so the model generalizes to 10K-scale histories without additional training cost. Across offline and online experiments, we observe predictable, monotonic gains as we scale history length and model capacity, mirroring the scaling law behavior observed in large language models. Deployed at full traffic on Douyin, our system delivers significant improvements on key engagement metrics while meeting production latency, demonstrating a practical path to scaling end-to-end ultra-long sequence recommendation to the 10K regime.

cs.LG↗

arXiv2025

Nuclear modification of $B_c$ mesons in relativistic heavy-ion collisions based on a linear Boltzmann transport model

The nuclear modification factor ($R_\mathrm{AA}$) of $B_c$ mesons in high-energy nuclear collisions provides a novel probe of heavy quark interactions with the quark-gluon plasma (QGP). Based on a linear Boltzmann transport model that incorporates both Yukawa and string types of interactions between heavy quarks and the QGP, we study the production and evolution of heavy quarks and $B_c$ mesons within the same framework. A $B_c$ bound state dissociates while one of its constituent heavy quarks scatters with the QGP with momentum transfer greater than its binding energy. The medium-modified charm and bottom quarks can recombine into $B_c$ mesons, and the medium-modified bottom quarks can also fragment to $B_c$ mesons. We find that most primordial $B_c$ mesons generated from the initial hard collisions dissociate inside the QGP. The production of $B_c$ mesons is primarily driven by the recombination mechanism at low transverse momentum and fragmentation at high transverse momentum. The string interaction dominates over the Yukawa interaction in the nuclear modification of $B_c$ mesons. The participant number dependence of the $B_c$ meson $R_\mathrm{AA}$ is determined by the complicated interplay between the heavy quark yield, energy loss, and the QGP volume. We obtain a reasonable description of the $R_\mathrm{AA}$ of $B_c$ mesons in Pb+Pb collisions at $\sqrt{s_\mathrm{NN}}=5.02$ TeV, and provide predictions for Au+Au collisions at $\sqrt{s_\mathrm{NN}}=200$ GeV.

hep-ph↗

arXiv2025

DiffuEraser: A Diffusion Model for Video Inpainting

Recent video inpainting algorithms integrate flow-based pixel propagation with transformer-based generation to leverage optical flow for restoring textures and objects using information from neighboring frames, while completing masked regions through visual Transformers. However, these approaches often encounter blurring and temporal inconsistencies when dealing with large masks, highlighting the need for models with enhanced generative capabilities. Recently, diffusion models have emerged as a prominent technique in image and video generation due to their impressive performance. In this paper, we introduce DiffuEraser, a video inpainting model based on stable diffusion, designed to fill masked regions with greater details and more coherent structures. We incorporate prior information to provide initialization and weak conditioning,which helps mitigate noisy artifacts and suppress hallucinations. Additionally, to improve temporal consistency during long-sequence inference, we expand the temporal receptive fields of both the prior model and DiffuEraser, and further enhance consistency by leveraging the temporal smoothing property of Video Diffusion Models. Experimental results demonstrate that our proposed method outperforms state-of-the-art techniques in both content completeness and temporal consistency while maintaining acceptable efficiency.

cs.CV↗

arXiv2024

Traveling waves to a logarithmic chemotaxis model with fast diffusion and singularities

This paper is concerned with a chemotaxis model with logarithmic sensitivity and fast diffusion, which possesses strong singularities for the sensitivity at zero-concentration of chemical signal, and for the diffusion at zero-population of cells, respectively. The main purpose is to show the existence of traveling waves connecting the singular zero-end-state, and particularly, to show the asymptotic stability of these traveling waves. The challenge of the problem is the interaction of two kinds of singularities involved in the model: one is the logarithmic singularity of the sensitivity; and the other is the power-law singularity of the diffusivity. To overcome the singularities for the wave stability, some new techniques of weighted energy method are introduced artfully. Numerical simulations are also carried out, which further confirm our theoretical stability results, in particular, the numerical results indicate that the effect of fast diffusion to the structure of traveling waves is essential, which causes the traveling waves much steeper like shock waves. This new phenomenon is a first observation.

math.AP↗

arXiv2024

Transition layers to chemotaxis-consumption models with volume-filling effect

We are interested in the dynamical behaviors of solutions to a parabolic-parabolic chemotaxis-consumption model with a volume-filling effect on a bounded interval, where the physical no-flux boundary condition for the bacteria and mixed Dirichlet-Neumann boundary condition for the oxygen are prescribed. By taking a continuity argument, we first show that the model admits a unique nonconstant steady state. Then we use Helly's compactness theorem to show that the asymptotic profile of steady state is a transition layer as the chemotactic coefficient goes to infinity. Finally, based on the energy method along with a cancellation structure of the model, we show that the steady state is nonlinearly stable under appropriate perturbations. Moreover, we do not need any assumption on the parameters in showing the stability of steady state.

math.AP↗

arXiv2024

Using Generative Artificial Intelligence Creatively in the Classroom and Research: Examples and Lessons Learned

Although generative artificial intelligence (AI) is not new, recent technological breakthroughs have transformed its capabilities across many domains. These changes necessitate new attention from educators and specialized training within the atmospheric and related sciences. Enabling students to use generative AI effectively, responsibly, and ethically is crucial for their academic and professional development. Educators can also use generative AI to develop engaging classroom activities, such as active learning modules and games; however, they must be aware of potential pitfalls and biases. There are also ethical implications in using tools that lack transparency and have a considerable carbon footprint, as well as equity concerns for students who lack access to more sophisticated paid versions of generative AI tools and have deficiencies in prior educational training. This article is written for students and educators alike, particularly those interested in learning more about generative AI in education and research, including its use cases, ethical concerns, and a brief history of its emergence. Sample user prompts are also provided across numerous applications in education and the atmospheric and related sciences. Current solutions addressing broader ethical concerns regarding the use of generative AI in education remain limited; however, this work aims to foster a discussion that could galvanize the education community around shared goals and values.

cs.HC↗

arXiv2024

Randomized Algorithms for Solving Singular Value Decomposition Problems with Matlab Toolbox

This thesis gives an overview of the state-of-the-art randomized linear algebra algorithms for singular value decomposition (SVD), including the presentation of existing pseudo-codes and theoretical error analysis. Our main focus is on presenting numerical experiments illustrating image restoration using various randomized singular value decomposition (RSVD) methods; theoretical error bounds, computed errors, and canonical angles analysis for these RSVD algorithms. This thesis also comes with a newly developed Matlab toolbox that contains implementations and test examples for some of the state-of-the-art randomized numerical linear algebra algorithms.

math.OC↗

arXiv2024

Nonlinear stability of shock profiles to Burgers' equation with critical fast diffusion and singularity

In this paper we propose the first framework to study Burgers' equation featuring critical fast diffusion in form of $u_t+f(u)_x = (\ln u)_{xx}$. The solution possesses a strong singularity when $u=0$ hence bringing technical challenges. The main purpose of this paper is to investigate the asymptotic stability of viscous shocks, particularly those with shock profiles vanishing at the far field $x=+\infty$. To overcome the singularity, we introduce some weight functions and show the nonlinear stability of shock profiles through the weighted energy method. Numerical simulations are also carried out in different cases of fast diffusion with singularity, which illustrate and confirm our theoretical results.

math.AP↗

arXiv2024

Detecting Adversarial Spectrum Attacks via Distance to Decision Boundary Statistics

Machine learning has been adopted for efficient cooperative spectrum sensing. However, it incurs an additional security risk due to attacks leveraging adversarial machine learning to create malicious spectrum sensing values to deceive the fusion center, called adversarial spectrum attacks. In this paper, we propose an efficient framework for detecting adversarial spectrum attacks. Our design leverages the concept of the distance to the decision boundary (DDB) observed at the fusion center and compares the training and testing DDB distributions to identify adversarial spectrum attacks. We create a computationally efficient way to compute the DDB for machine learning based spectrum sensing systems. Experimental results based on realistic spectrum data show that our method, under typical settings, achieves a high detection rate of up to 99\% and maintains a low false alarm rate of less than 1\%. In addition, our method to compute the DDB based on spectrum data achieves 54\%--64\% improvements in computational efficiency over existing distance calculation methods. The proposed DDB-based detection framework offers a practical and efficient solution for identifying malicious sensing values created by adversarial spectrum attacks.

cs.CR↗

arXiv2023

DreaMoving: A Human Video Generation Framework based on Diffusion Models

In this paper, we present DreaMoving, a diffusion-based controllable video generation framework to produce high-quality customized human videos. Specifically, given target identity and posture sequences, DreaMoving can generate a video of the target identity moving or dancing anywhere driven by the posture sequences. To this end, we propose a Video ControlNet for motion-controlling and a Content Guider for identity preserving. The proposed model is easy to use and can be adapted to most stylized diffusion models to generate diverse results. The project page is available at https://dreamoving.github.io/dreamoving

cs.CV↗

arXiv2023

Correlations of dihadron polarization in central, peripheral and ultraperipheral heavy-ion collisions

While jet quenching in relativistic heavy-ion collisions has been extensively studied over decades, the polarization of quenched hadrons has rarely been discussed. It has recently been proposed that the correlations of dihadron polarization in $e^+e^-$ and $pp$ collisions provide a novel probe of the longitudinal spin transfer from hard partons to hadrons without requiring the colliding beams to be polarized. To support realistic experimental measurement of dihadron polarization with sufficient luminosity, we extend the aforementioned study to relativistic heavy-ion collisions by convoluting the vacuum fragmentation of partons with their energy loss inside the quark-gluon plasma (QGP). We find that while the correlation functions of $\Lambda$-$\Lambda$ (or $\Lambda$-$\bar{\Lambda}$) polarization in peripheral collisions is consistent with those in $pp$ collisions, clear enhancement can be seen in central collisions. These correlation functions appear sensitive to different assumptions in the DSV parameterization of parton fragmentation functions, and therefore could place additional constraints on the spin-dependent fragmentation functions of quarks and gluons. The correlation of dihadron polarization has also been explored in ultraperipheral heavy-ion collisions, which provides a cleaner probe of fragmentation functions of quarks produced by energetic photon-photon and photon-pomeron interactions.

hep-ph↗

arXiv2022

Interactions between heavy quarks and tilted QGP fireballs in 200 AGeV Au+Au collisions

Heavy quark observables are applied to probe the initial energy density distribution with violation of longitudinal boost invariance produced in relativistic heavy-ion collisions. Using an improved Langevin model coupled to a (3+1)-dimensional viscous hydrodynamic model, we study the nuclear modification factor ($R_\text{AA}$), directed flow ($v_1$) and elliptic flow ($v_2$) coefficients of heavy mesons and their decayed electrons at the RHIC energy. We find that the counter-clockwise tilt of the nuclear matter in the reaction plane results in a positive (negative) heavy flavor $v_1$ in the backward (forward) rapidity region, whose magnitude increases with the heavy quark transverse momentum. The difference in the heavy flavor $R_\text{AA}$ between different angular regions is also proposed as a complementary tool to characterize the asymmetry of the medium profile. Our model results are consistent with currently available data at RHIC and provide predictions that can be tested by future measurements.

nucl-th↗

arXiv2022

Evolution of global polarization in relativistic heavy-ion collisions within a perturbative approach

Extremely large angular orbital momentum can be produced in non-central heavy-ion collisions, leading to a strong transverse polarization of partons that scatter through the quark-gluon plasma (QGP) due to spin-orbital coupling. We develop a perturbative approach to describe the formation and spacetime evolution of quark polarization inside the QGP. Polarization from both the initial hard scatterings and interactions with the QGP have been consistently described using the quark-potential scattering approach, which has been coupled to realistic initial condition calculation and the subsequent (3+1)-dimensional viscous hydrodynamic simulation of the QGP for the first time. Within this improved approach, we have found that different spacetime-rapidity-dependent initial energy density distributions generate different time evolution profiles of the longitudinal flow velocity gradient of the QGP, which further lead to an approximately 15% difference in the final polarization of quarks collected on the hadronization hypersurface of the QGP. Therefore, in addition to the collective flow coefficients, the hyperon polarization may serve as a novel tool to help constrain the initial condition of the hot nuclear matter created in high-energy nuclear collisions.

nucl-th↗