arXiv subjects

Xinwen Zhang

Publications and source records attributed to Xinwen Zhang.

13 recordsLinked to original sources

arXiv2026

MIMFlow: Integrating Masked Image Modeling with Normalizing Flows for End-to-End Image Generation

Normalizing Flows (NFs) are powerful generative models capable of exact density estimation and sampling. However, their strict invertibility often forces the model to exhaust its capacity on low-level pixel details, hindering the capture of high-level semantic structures. While Masked Image Modeling (MIM) has excelled in representation learning, its integration into generative pipelines has remained largely modular and disjointed. In this paper, we propose MIMFlow, a unified end-to-end framework that jointly optimizes latent semantics, pixel reconstruction, and generative flow. By employing a VAE encoder to infer semantic latent from masked images, MIMFlow achieves a principled decoupling of the generative task: the Normalizing Flow focuses on modeling a simplified, low-frequency semantic manifold, while a specialized decoder handles high-frequency synthesis. This design effectively resolves the inherent capacity bottleneck of NFs, allowing the model to prioritize global structural coherence over redundant noise. Empirical results on ImageNet 256$\times$256 show that MIMFlow-L reaches 71.3\% linear probing accuracy and an FID of 2.50. Despite using only 128 tokens (50\% fewer than standard models), it yields a 32.8\% performance gain over similar-scale NF baselines. Our code is available at https://github.com/MCG-NJU/MIMFlow.

cs.CV↗

arXiv2026

Nonconvex Decentralized Stochastic Bilevel Optimization under Heavy-Tailed Noise

Existing decentralized stochastic optimization methods assume the lower-level loss function is strongly convex and the stochastic gradient noise has finite variance. These strong assumptions typically are not satisfied in real-world machine learning models. For example, learning on language data typically leads to heavy-tailed gradient. To address these limitations, we develop a novel decentralized stochastic bilevel optimization algorithm for the nonconvex bilevel optimization problem under heavy-tailed noise. Specifically, we develop a normalized stochastic variance-reduced bilevel gradient descent algorithm, which does not rely on any clipping operation. Moreover, we establish its convergence rate by innovatively bounding interdependent gradient sequences under heavy-tailed noise for nonconvex decentralized bilevel optimization problems. As far as we know, this is the first decentralized bilevel optimization algorithm with rigorous theoretical guarantees under heavy-tailed noise. The extensive experimental results confirm the effectiveness of our algorithm in handling heavy-tailed noise.

cs.LG↗

arXiv2026

SAM 2++: Tracking Anything at Any Granularity

Due to the varying granularity of target states across different tasks, most existing trackers are tailored to a single task, which specificity limits their generalization, preventing them from effectively utilizing multi-task training data and leading to redundancy in both model design and parameters. Although recent unified vision models share partial architectures across tasks, they usually retain task-specific interfaces and overlook the common tracking principle behind different granularities, leaving a gap for truly unified video tracking. To unify video tracking tasks, we present SAM 2++, a unified framework that can handle target states at different granularities, including masks, boxes, and points, through an integrated design of prompt encoding, output decoding, and memory representation. First, to handle different target granularities, we design task-specific prompts that map diverse task inputs into general prompt embeddings, together with a Unified Decoder that produces task results in a common output form without redesigning the overall pipeline. Next, to satisfy memory matching, the core operation of tracking, we introduce a task-adaptive memory mechanism that unifies memory across different granularities while preserving their distinct state semantics, preventing full parameter sharing from causing interference across granularities. Finally, we introduce Tracking-Any-Granularity, the first large and diverse video tracking dataset with rich annotations at three granularities. It is constructed through a customized data engine with phased manual annotation and model-assisted completion, providing a comprehensive resource for training, benchmarking, and analyzing unified tracking models. Comprehensive experiments confirm that SAM 2++ sets a new state of the art across diverse tracking tasks at different granularities, establishing a unified and robust tracking framework.

cs.CV↗

arXiv2025

Federated Stochastic Minimax Optimization under Heavy-Tailed Noises

Heavy-tailed noise has attracted growing attention in nonconvex stochastic optimization, as numerous empirical studies suggest it offers a more realistic assumption than standard bounded variance assumption. In this work, we investigate nonconvex-PL minimax optimization under heavy-tailed gradient noise in federated learning. We propose two novel algorithms: Fed-NSGDA-M, which integrates normalized gradients, and FedMuon-DA, which leverages the Muon optimizer for local updates. Both algorithms are designed to effectively address heavy-tailed noise in federated minimax optimization, under a milder condition. We theoretically establish that both algorithms achieve a convergence rate of $O({1}/{(TNp)^{\frac{s-1}{2s}}})$. To the best of our knowledge, these are the first federated minimax optimization algorithms with rigorous theoretical guarantees under heavy-tailed noise. Extensive experiments further validate their effectiveness.

cs.LG↗

arXiv2025

On Provable Benefits of Muon in Federated Learning

The recently introduced optimizer, Muon, has gained increasing attention due to its superior performance across a wide range of applications. However, its effectiveness in federated learning remains unexplored. To address this gap, this paper investigates the performance of Muon in the federated learning setting. Specifically, we propose a new algorithm, FedMuon, and establish its convergence rate for nonconvex problems. Our theoretical analysis reveals multiple favorable properties of FedMuon. In particular, due to its orthonormalized update direction, the learning rate of FedMuon is independent of problem-specific parameters, and, importantly, it can naturally accommodate heavy-tailed noise. The extensive experiments on a variety of neural network architectures validate the effectiveness of the proposed algorithm.

cs.LG↗

arXiv2025

AniSora: Exploring the Frontiers of Animation Video Generation in the Sora Era

Animation has gained significant interest in the recent film and TV industry. Despite the success of advanced video generation models like Sora, Kling, and CogVideoX in generating natural videos, they lack the same effectiveness in handling animation videos. Evaluating animation video generation is also a great challenge due to its unique artist styles, violating the laws of physics and exaggerated motions. In this paper, we present a comprehensive system, AniSora, designed for animation video generation, which includes a data processing pipeline, a controllable generation model, and an evaluation benchmark. Supported by the data processing pipeline with over 10M high-quality data, the generation model incorporates a spatiotemporal mask module to facilitate key animation production functions such as image-to-video generation, frame interpolation, and localized image-guided animation. We also collect an evaluation benchmark of 948 various animation videos, with specifically developed metrics for animation video generation. Our entire project is publicly available on https://github.com/bilibili/Index-anisora/tree/main.

cs.GR↗

arXiv2024

Federated Compositional Deep AUC Maximization

Federated learning has attracted increasing attention due to the promise of balancing privacy and large-scale learning; numerous approaches have been proposed. However, most existing approaches focus on problems with balanced data, and prediction performance is far from satisfactory for many real-world applications where the number of samples in different classes is highly imbalanced. To address this challenging problem, we developed a novel federated learning method for imbalanced data by directly optimizing the area under curve (AUC) score. In particular, we formulate the AUC maximization problem as a federated compositional minimax optimization problem, develop a local stochastic compositional gradient descent ascent with momentum algorithm, and provide bounds on the computational and communication complexities of our algorithm. To the best of our knowledge, this is the first work to achieve such favorable theoretical results. Finally, extensive experimental results confirm the efficacy of our method.

cs.LG↗

arXiv2023

SynthMix: Mixing up Aligned Synthesis for Medical Cross-Modality Domain Adaptation

The adversarial methods showed advanced performance by producing synthetic images to mitigate the domain shift, a common problem due to the hardship of acquiring labelled data in medical field. Most existing studies focus on modifying the network architecture, but little has worked on the GAN training strategy. In this work, we propose SynthMix, an add-on module with a natural yet effective training policy that can promote synthetic quality without altering the network architecture. Following the adversarial philosophy of GAN, we designed a mix-up synthesis scheme termed SynthMix. It coherently mixed up aligned images of real and synthetic samples to stimulate the generation of fine-grained features, examined by an associated Inspector for the domain-specific details. We evaluated our method on two segmentation benchmarks among three publicly available datasets, where our method showed a significant performance gain compared with existing state-of-the-art approaches.

eess.IV↗

arXiv2022

Unsupervised Domain Adaptive Fundus Image Segmentation with Few Labeled Source Data

Deep learning-based segmentation methods have been widely employed for automatic glaucoma diagnosis and prognosis. In practice, fundus images obtained by different fundus cameras vary significantly in terms of illumination and intensity. Although recent unsupervised domain adaptation (UDA) methods enhance the models' generalization ability on the unlabeled target fundus datasets, they always require sufficient labeled data from the source domain, bringing auxiliary data acquisition and annotation costs. To further facilitate the data efficiency of the cross-domain segmentation methods on the fundus images, we explore UDA optic disc and cup segmentation problems using few labeled source data in this work. We first design a Searching-based Multi-style Invariant Mechanism to diversify the source data style as well as increase the data amount. Next, a prototype consistency mechanism on the foreground objects is proposed to facilitate the feature alignment for each kind of tissue under different image styles. Moreover, a cross-style self-supervised learning stage is further designed to improve the segmentation performance on the target images. Our method has outperformed several state-of-the-art UDA segmentation methods under the UDA fundus segmentation with few labeled source data.

cs.CV↗

arXiv2022

Imitate then Transcend: Multi-Agent Optimal Execution with Dual-Window Denoise PPO

A novel framework for solving the optimal execution and placement problems using reinforcement learning (RL) with imitation was proposed. The RL agents trained from the proposed framework consistently outperformed the industry benchmark time-weighted average price (TWAP) strategy in execution cost and showed great generalization across out-of-sample trading dates and tickers. The impressive performance was achieved from three aspects. First, our RL network architecture called Dual-window Denoise PPO enabled efficient learning in a noisy market environment. Second, a reward scheme with imitation learning was designed, and a comprehensive set of market features was studied. Third, our flexible action formulation allowed the RL agent to tackle optimal execution and placement collectively resulting in better performance than solving individual problems separately. The RL agent's performance was evaluated in our multi-agent realistic historical limit order book simulator in which price impact was accurately assessed. In addition, ablation studies were also performed, confirming the superiority of our framework.

cs.LG↗

arXiv2018

ResumeNet: A Learning-based Framework for Automatic Resume Quality Assessment

Recruitment of appropriate people for certain positions is critical for any companies or organizations. Manually screening to select appropriate candidates from large amounts of resumes can be exhausted and time-consuming. However, there is no public tool that can be directly used for automatic resume quality assessment (RQA). This motivates us to develop a method for automatic RQA. Since there is also no public dataset for model training and evaluation, we build a dataset for RQA by collecting around 10K resumes, which are provided by a private resume management company. By investigating the dataset, we identify some factors or features that could be useful to discriminate good resumes from bad ones, e.g., the consistency between different parts of a resume. Then a neural-network model is designed to predict the quality of each resume, where some text processing techniques are incorporated. To deal with the label deficiency issue in the dataset, we propose several variants of the model by either utilizing the pair/triplet-based loss, or introducing some semi-supervised learning technique to make use of the abundant unlabeled data. Both the presented baseline model and its variants are general and easy to implement. Various popular criteria including the receiver operating characteristic (ROC) curve, F-measure and ranking-based average precision (AP) are adopted for model evaluation. We compare the different variants with our baseline model. Since there is no public algorithm for RQA, we further compare our results with those obtained from a website that can score a resume. Experimental results in terms of different criteria demonstrate the effectiveness of the proposed method. We foresee that our approach would transform the way of future human resources management.

cs.IR↗

arXiv2012

Multiparty Cloud Computation

With the increasing popularity of the cloud, clients oursource their data to clouds in order to take advantage of unlimited virtualized storage space and the low management cost. Such trend prompts the privately oursourcing computation, called \emph{multiparty cloud computation} (\MCC): Given $k$ clients storing their data in the cloud, how can they perform the joint functionality by contributing their private data as inputs, and making use of cloud's powerful computation capability. Namely, the clients wish to oursource computation to the cloud together with their private data stored in the cloud, which naturally happens when the computation is involved with large datasets, e.g., to analyze malicious URLs. We note that the \MCC\ problem is different from widely considered concepts, e.g., secure multiparty computation and multiparty computation with server aid. To address this problem, we introduce the notion of \emph{homomorphic threshold proxy re-encryption} schemes, which are encryption schemes that enjoy three promising properties: proxy re-encryption -- transforming encrypted data of one user to encrypted data of target user, threshold decryption -- decrypting encrypted data by combining secret key shares obtained by a set of users, and homomorphic computation -- evaluating functions on the encrypted data. To demonstrate the feasibility of the proposed approach, we present an encryption scheme which allows anyone to compute arbitrary many additions and at most one multiplications.

cs.CR↗

arXiv2011

Unleashing the Power of Mobile Cloud Computing using ThinkAir

Smartphones have exploded in popularity in recent years, becoming ever more sophisticated and capable. As a result, developers worldwide are building increasingly complex applications that require ever increasing amounts of computational power and energy. In this paper we propose ThinkAir, a framework that makes it simple for developers to migrate their smartphone applications to the cloud. ThinkAir exploits the concept of smartphone virtualization in the cloud and provides method level computation offloading. Advancing on previous works, it focuses on the elasticity and scalability of the server side and enhances the power of mobile cloud computing by parallelizing method execution using multiple Virtual Machine (VM) images. We evaluate the system using a range of benchmarks starting from simple micro-benchmarks to more complex applications. First, we show that the execution time and energy consumption decrease two orders of magnitude for the N-queens puzzle and one order of magnitude for a face detection and a virus scan application, using cloud offloading. We then show that if a task is parallelizable, the user can request more than one VM to execute it, and these VMs will be provided dynamically. In fact, by exploiting parallelization, we achieve a greater reduction on the execution time and energy consumption for the previous applications. Finally, we use a memory-hungry image combiner tool to demonstrate that applications can dynamically request VMs with more computational power in order to meet their computational requirements.

cs.DC↗