Source author record

Yongxin Tong

Yongxin Tong appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

Machine Learning Artificial Intelligence Computer Vision Databases Computation and Language Distributed, Parallel, and Cluster Computing

Catalog footprint

What is connected

7works

6topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2026arXiv

CAFEDistill: Learning Personalized and Dynamic Models through Federated Early-Exit Network Distillation

Personalized Federated Learning (PFL) enables collaboratively model training on decentralized, heterogeneous data while tailoring them to each client's unique distribution. However, existing PFL methods produce static models with a fixed tradeoff between accuracy and efficiency, limiting their applicability in environments where inference requirements vary with contexts and resource availability. Early-exit networks (EENs) offer adaptive inference by attaching intermediate classifiers. Yet integrating them into PFL is challenging due to client-wise heterogeneity and depth-wise interference arising from conflicting exit objectives. Prior studies fail to resolve both conflicts simultaneously, leading to suboptimal performance. In this paper, we propose CAFEDistill, a Conflict-Aware Federated Exit Distillation framework that jointly addresses these conflicts and extends PFL to early-exit networks. Through a progressive, depth-prioritized student coordination mechanism, CAFEDistill mitigates interference among shallow and deep exits while allowing effective personalized knowledge transfer across clients. Furthermore, it reduces communication overhead via a client-decoupled formulation. Extensive evaluations show that CAFEDistill outperforms the state-of-the-arts, achieving higher accuracy and reducing inference costs by 30.79%-46.86%.

preprint2026arXiv

FedSEA-LLaMA: A Secure, Efficient and Adaptive Federated Splitting Framework for Large Language Models

Private data holds promise for improving LLMs due to its high quality, but its scattered distribution across data silos and the high computational demands of LLMs limit their deployment in federated environments. To address this, the transformer-based federated split models are proposed, which offload most model parameters to the server (or distributed clients) while retaining only a small portion on the client to ensure data privacy. Despite this design, they still face three challenges: 1) Peer-to-peer key encryption struggles to secure transmitted vectors effectively; 2) The auto-regressive nature of LLMs means that federated split learning can only train and infer sequentially, causing high communication overhead; 3) Fixed partition points lack adaptability to downstream tasks. In this paper, we introduce FedSEA-LLaMA, a Secure, Efficient, and Adaptive Federated splitting framework based on LLaMA2. First, we inject Gaussian noise into forward-pass hidden states to enable secure end-to-end vector transmission. Second, we employ attention-mask compression and KV cache collaboration to reduce communication costs, accelerating training and inference. Third, we allow users to dynamically adjust the partition points for input/output blocks based on specific task requirements. Experiments on natural language understanding, summarization, and conversational QA tasks show that FedSEA-LLaMA maintains performance comparable to centralized LLaMA2 and achieves up to 8x speedups in training and inference. Further analysis of privacy attacks and different partition points also demonstrates the effectiveness of FedSEA-LLaMA in security and adaptability.

preprint2026arXiv

Replacing Parameters with Preferences: Federated Alignment of Heterogeneous Vision-Language Models

Vision-Language Models (VLMs) have broad potential in privacy-sensitive domains such as healthcare and finance, yet strict data-sharing constraints render centralized training infeasible. Federated Learning mitigates this issue by enabling decentralized training, but practical deployments face challenges due to client heterogeneity in computational resources, application requirements, and model architectures. Under extreme model and data heterogeneity, replacing parameter aggregation with preference-based collaboration offers a more suitable interface, as it eliminates the need for direct parameter or data exchange. Motivated by this, we propose MoR, a federated alignment framework that combines GRPO with Mixture-of-Rewards for heterogeneous VLMs. In MoR, each client locally trains a reward model from local preference annotations, capturing specific evaluation signals without exposing raw data. To combine these heterogeneous supervision signals, MoR introduces a Mixture-of-Rewards mechanism with learned routing, which adaptively fuses client reward models according to the input and alignment objective. The server then optimizes a base VLM using GRPO with a KL penalty to a reference model, enabling preference alignment without requiring client models to share architectures or parameters. Experiments on diverse public vision-language benchmarks demonstrate that MoR consistently outperforms federated alignment baselines in generalization and cross-client adaptability. Our approach provides a scalable solution for privacy-preserving alignment of heterogeneous VLMs under federated settings.

preprint2026arXiv

Taming Noise-Induced Prototype Degradation for Privacy-Preserving Personalized Federated Fine-Tuning

Prototype-based Personalized Federated Learning (ProtoPFL) enables efficient multi-domain adaptation by communicating compact class prototypes, but directly sharing them poses privacy risks. A common defense involves per-example $\ell_2$ clipping before prototype computation to bound sensitivity, followed by isotropic Gaussian noise to enforce Local Differential Privacy (LDP). However, Isotropic Gaussian Prototype Perturbation (IGPP) typically over-perturbs discriminative dimensions and struggles to balance the clipping threshold with representation fidelity. In this paper, we propose VPDR, a client-side privacy plug-in that seamlessly integrates into existing ProtoPFLs. Motivated by the observation that dimension-wise class variance reflects discriminability, we introduce Variance-adaptive Prototype Perturbation (VPP), which allocates less noise to discriminative subspaces, preserving semantic separability while ensuring privacy. We further develop Distillation-guided Clipping Regularization (DCR), which enables feature norms to adaptively concentrate near the predefined clipping threshold while maintaining prediction consistency. Theoretical analysis shows that our groupwise mechanism provides privacy guarantees no weaker than the isotropic baseline under the same privacy constraints. Extensive experiments on multi-domain benchmarks demonstrate that VPDR achieves a superior privacy-utility trade-off, outperforming IGPP in personalized federated fine-tuning without sacrificing robustness against realistic attacks.

preprint2022arXiv

p-Meta: Towards On-device Deep Model Adaptation

Data collected by IoT devices are often private and have a large diversity across users. Therefore, learning requires pre-training a model with available representative data samples, deploying the pre-trained model on IoT devices, and adapting the deployed model on the device with local data. Such an on-device adaption for deep learning empowered applications demands data and memory efficiency. However, existing gradient-based meta learning schemes fail to support memory-efficient adaptation. To this end, we propose p-Meta, a new meta learning method that enforces structure-wise partial parameter updates while ensuring fast generalization to unseen tasks. Evaluations on few-shot image classification and reinforcement learning tasks show that p-Meta not only improves the accuracy but also substantially reduces the peak dynamic memory by a factor of 2.5 on average compared to state-of-the-art few-shot adaptation methods.

preprint2012arXiv

Mining Frequent Itemsets over Uncertain Databases

In recent years, due to the wide applications of uncertain data, mining frequent itemsets over uncertain databases has attracted much attention. In uncertain databases, the support of an itemset is a random variable instead of a fixed occurrence counting of this itemset. Thus, unlike the corresponding problem in deterministic databases where the frequent itemset has a unique definition, the frequent itemset under uncertain environments has two different definitions so far. The first definition, referred as the expected support-based frequent itemset, employs the expectation of the support of an itemset to measure whether this itemset is frequent. The second definition, referred as the probabilistic frequent itemset, uses the probability of the support of an itemset to measure its frequency. Thus, existing work on mining frequent itemsets over uncertain databases is divided into two different groups and no study is conducted to comprehensively compare the two different definitions. In addition, since no uniform experimental platform exists, current solutions for the same definition even generate inconsistent results. In this paper, we firstly aim to clarify the relationship between the two different definitions. Through extensive experiments, we verify that the two definitions have a tight connection and can be unified together when the size of data is large enough. Secondly, we provide baseline implementations of eight existing representative algorithms and test their performances with uniform measures fairly. Finally, according to the fair tests over many different benchmark data sets, we clarify several existing inconsistent conclusions and discuss some new findings.

preprint2012arXiv

Whom to Ask? Jury Selection for Decision Making Tasks on Micro-blog Services

It is universal to see people obtain knowledge on micro-blog services by asking others decision making questions. In this paper, we study the Jury Selection Problem(JSP) by utilizing crowdsourcing for decision making tasks on micro-blog services. Specifically, the problem is to enroll a subset of crowd under a limited budget, whose aggregated wisdom via Majority Voting scheme has the lowest probability of drawing a wrong answer(Jury Error Rate-JER). Due to various individual error-rates of the crowd, the calculation of JER is non-trivial. Firstly, we explicitly state that JER is the probability when the number of wrong jurors is larger than half of the size of a jury. To avoid the exponentially increasing calculation of JER, we propose two efficient algorithms and an effective bounding technique. Furthermore, we study the Jury Selection Problem on two crowdsourcing models, one is for altruistic users(AltrM) and the other is for incentive-requiring users(PayM) who require extra payment when enrolled into a task. For the AltrM model, we prove the monotonicity of JER on individual error rate and propose an efficient exact algorithm for JSP. For the PayM model, we prove the NP-hardness of JSP on PayM and propose an efficient greedy-based heuristic algorithm. Finally, we conduct a series of experiments to investigate the traits of JSP, and validate the efficiency and effectiveness of our proposed algorithms on both synthetic and real micro-blog data.

Institution

Affiliation not imported yet

This author record came from a source that does not expose affiliation metadata. Once the author claims the profile or we enrich the record from another provider, this section will link to the concrete institution.

Topic footprint