publications

2026

1. AAAI’26

   

   SMoFi: Step-wise Momentum Fusion for Split Federated Learning on Heterogeneous Data

   Mingkun Yang, Ran Zhu, Qing Wang, and Jie Yang

   In The AAAI Conference on Artificial Intelligence (AAAI), 2026

   Abs HTML

   Split Federated Learning is a system-efficient federated learning paradigm that leverages the rich computing resources at a central server to train model partitions. Data heterogeneity across silos, however, presents a major challenge undermining the convergence speed and accuracy of the global model. This paper introduces Step-wise Momentum Fusion (SMoFi), an effective and lightweight framework that counteracts gradient divergence arising from data heterogeneity by synchronizing the momentum buffers across server-side optimizers. To control gradient divergence over the training process, we design a staleness-aware alignment mechanism that imposes constraints on gradient updates of the serverside submodel at each optimization step. Extensive validations on multiple real-world datasets show that SMoFi consistently improves global model accuracy (up to 7.1%) and convergence speed (up to 10.25×). Furthermore, SMoFi has a greater impact with more clients involved and deeper learning models, making it particularly suitable for model training in resource-constrained contexts.

2025

1. NeurIPS’25

   

   Flick: Empowering Federated Learning with Commonsense Knowledge

   Ran Zhu, Mingkun Yang, Shiqiang Wang, Jie Yang, and Qing Wang

   In The Thirty-ninth Annual Conference on Neural Information Processing Systems (NeurIPS), 2025

   Abs HTML

   Federated Learning (FL) has emerged as a privacy-preserving framework for training models on data generated at the edge. However, the heterogeneity of data silos (e.g., label skew and domain shift) often leads to inconsistent learning objectives and suboptimal model performance. Inspired by the data-driven approach, we propose Flick, a novel data generation framework for heterogeneous FederatedLearning with Commonsense Knowledge from Large Language Models (LLMs). In Flick, the client performs the local data summary to capture client-specific knowledge in textual form. The central server then distills task-relevant, highquality knowledge from the out-of-the-box LLM – guided by cross-client-specific insights – to generate informative text prompts. These prompts direct a generative model in producing synthetic data, enabling global model fine-tuning and local data compensation. This process gradually aligns the label and feature distributions across clients. Extensive results on three datasets demonstrate that Flick improves the global model accuracy by up to 11.43%, and accelerates convergence by up to12.9×, validating its effectiveness in addressing data heterogeneity.

2024

1. IMWUT’24

   

   ShuffleFL: Addressing heterogeneity in multi-device federated learning

   Ran Zhu, Mingkun Yang, and Qing Wang

   In Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies (IMWUT), 2024

   Abs HTML

   Federated Learning (FL) has emerged as a privacy-preserving paradigm for collaborative deep learning model training across distributed data silos. Despite its importance, FL faces challenges such as high latency and less effective global models. In this paper, we propose ShuffleFL, an innovative framework stemming from the hierarchical FL, which introduces a user layer between the FL devices and the FL server. ShuffleFL naturally groups devices based on their affiliations, e.g., belonging to the same user, to ease the strict privacy restriction-"data at the FL devices cannot be shared with others", thereby enabling the exchange of local samples among them. The user layer assumes a multi-faceted role, not just aggregating local updates but also coordinating data shuffling within affiliated devices. We formulate this data shuffling as an optimization problem, detailing our objectives to align local data closely with device computing capabilities and to ensure a more balanced data distribution at the intra-user devices. Through extensive experiments using realistic device profiles and five non-IID datasets, we demonstrate that ShuffleFL can improve inference accuracy by 2.81% to 7.85% and speed up the convergence by 4.11x to 36.56x when reaching the target accuracy.

2. ICLR’24

   

   FedTrans: Client-Transparent Utility Estimation for Robust Federated Learning

   Mingkun Yang^*, Ran Zhu^*, Qing Wang, and Jie Yang

   In The Twelfth International Conference on Learning Representations (ICLR), 2024

   Abs HTML Code

   Federated Learning (FL) is an important privacy-preserving learning paradigm that plays an important role in the Intelligent Internet of Things. Training a global model in FL, however, is vulnerable to the noise in the heterogeneous data across the clients. In this paper, we introduce **FedTrans**, a novel client-transparent client utility estimation method designed to guide client selection for noisy scenarios, mitigating performance degradation problems. To estimate the client utility, we propose a Bayesian framework that models client utility and its relationships with the weight parameters and the performance of local models. We then introduce a variational inference algorithm to effectively infer client utility, given only a small amount of auxiliary data. Our evaluation demonstrates that leveraging FedTrans as a guide for client selection can lead to a better accuracy performance (up to 7.8%), ensuring robustness in noisy scenarios.

3. COMMAG’24

   

   Centimeter-Level Indoor Visible Light Positioning

   Ran Zhu, Maxim Abeele, Jona Beysens, Jie Yang, and Qing Wang

   IEEE Communications Magazine, 2024

   Abs HTML

   Visible light positioning (VLP) based on the received signal strength (RSS) can leverage a dense deployment of LEDs in future lighting infrastructure to provide accurate and energy-efficient indoor positioning. However, its positioning accuracy heavily depends on the density of collected fingerprints, which is labor-intensive. In this work, we propose a data pre-processing method, including data cleaning and data augmentation, to construct reliable and dense fingerprint samples, thereby alleviating the impact of noisy samples as well as reducing labor intensity. Extensive experiments demonstrate that our proposed method achieves an average positioning error of 1.7 cm, utilizing a sparse dataset that reduces the fingerprint collection effort by 98 percent. Running a tinyML-based model for VLP on the Arduino Nano microcontroller, we also show the possibilities for deploying RSS fingerprint-based VLP systems on resource-constrained embedded devices for real-world applications.

2023

1. SECON’23

   

   FedNaWi: Selecting the Befitting Clients for Robust Federated Learning in IoT Applications

   Ran Zhu, Mingkun Yang, Jie Yang, and Qing Wang

   In 20th Annual IEEE International Conference on Sensing, Communication, and Networking (SECON), 2023

   Abs HTML

   Federated Learning (FL) is an important privacy-preserving learning paradigm that is expected to play an essential role in the future Intelligent Internet of Things (IoT). However, model training in FL is vulnerable to noise and the statistical heterogeneity of local data across IoT clients. In this paper, we propose FedNaWi, a “Go Narrow, Then Wide” client selection method that speeds up the FL training, achieves higher model performance, while requiring no additional data or sensitive information transfer from clients. Our method first selects reliable clients (i.e., going narrow) which allows the global model to quickly improve its performance and then includes less reliable clients (i.e., going wide) to exploit more IoT data of clients to further improve the global model. To profile client utility, we introduce a unified Bayesian framework to model the client utility at the FL server, assisted by a small amount of auxiliary data. We conduct extensive evaluations with 5 state-of-the-art FL methods, on 3 IoT tasks and under 7 different types of label and feature noise. We build an FL testbed with 38 IoT nodes (20 nodes run on Raspberry Pi 4B and 18 nodes run on Jetson Nano) for the evaluation. Our results show that FedNaWi improves the FL accuracy substantially and significantly reduces energy consumption. In particular, FedNaWi improves the accuracy from 35% to 75% in the non-IID Dirichlet setting, and reduces the average energy consumption by 55%.

2. Neurocomputing’23

   

   ContextAVO: Local context guided and refining poses for deep visual odometry

   Rujun Song, Ran Zhu, Zhuoling Xiao, and Bo Yan

   Neurocomputing, 2023

   Abs HTML

   Learning-based monocular visual odometry (VO) has lately drawn significant attention for its robustness to camera parameters and environmental variations. The correlation of ego-motion in the local time dimension, denoted as the local context, is crucial for alleviating accumulated errors of VO problems. Unlike most current learning-based methods, our approach, called ContextAVO, focuses on the effectiveness of local contexts to improve the estimation recovered from consecutive multiple optical flow snippets. To retain the pose consistency in the temporal domain, we design the Context-Attention Refining component to adaptively ameliorate current inference by exploiting the continuity of camera motions and aligning corresponding observations with local contexts. Besides, we employ the multi-length window to make ContextAVO more suitable for general scenarios and less dependent on the fixed length of the input snippet. Extensive experiments on outdoor KITTI, Malaga, ApolloScape, and indoor TUM RGB-D datasets have demonstrated that our approach efficiently produces competitive results against classic algorithms. It outperforms state-of-the-art methods by large margins, improving up to 7.40% and 48.56% for translational and rotational estimation, respectively.

2022

1. Neurocomputing’22

   

   ContextAVO: Local context guided and refining poses for deep visual odometry

   Ran Zhu, Mingkun Yang, Wang Liu, Rujun Song, Bo Yan, and 1 more author

   Neurocomputing, 2022

   Abs HTML

   The technology for Visual Odometry (VO) that estimates the position and orientation of the moving object through analyzing the image sequences captured by on-board cameras, has been well investigated with the rising interest in autonomous driving. This paper studies monocular VO from the perspective of Deep Learning (DL). Unlike most current learning-based methods, our approach, called DeepAVO, is established on the intuition that features contribute discriminately to different motion patterns. Specifically, we present a novel four-branch network to learn the rotation and translation by leveraging Convolutional Neural Networks (CNNs) to focus on different quadrants of optical flow input. To enhance the ability of feature selection, we further introduce an effective channel-spatial attention mechanism to force each branch to explicitly distill related information for specific Frame to Frame (F2F) motion estimation. Experiments on various datasets involving outdoor driving and indoor walking scenarios show that the proposed DeepAVO outperforms the state-of-the-art monocular methods by a large margin, demonstrating competitive performance to the stereo VO algorithm and verifying promising potential for generalization.

2. Sensors Journal’22

   

   Symmetrical-Net: Adaptive zero velocity detection for ZUPT-aided pedestrian navigation system

   Mingkun Yang, Ran Zhu, Zhuoling Xiao, and Bo Yan

   IEEE Sensors Journal, 2022

   Abs HTML

   Inertial navigation system (INS) is a practical method for indoor pedestrian navigation without pre-installation of infrastructure. Based on the fundamentals of human bipedal motion, zero velocity update (ZUPT) is a pervasive approach to tackle the accumulated error of inertial measurement units (IMU). While zero velocity detection plays a vital role in the algorithm, existing fixed-threshold methods to pick these pseudo-measurements of error-state Kalman Filter (ESKF) have the doubtful capability to fit various individuals in different motions. To address this issue, we propose the Symmetrical-Net leveraging deep Recurrent Convolutional Neural Networks (RCNNs) to detect the zero velocity interval adaptively. Additionally, two RCNNs are constructed in the symmetrical framework considering bidirectional IMU readings, which further improves the performance of the model. A comprehensive evaluation containing 87 different trajectories from 27 individuals has been conducted. The results show that the detection accuracy is up to 99.5% and 96.5% on the training and validation sets, respectively. It is verified that the precise and robust Symmetrical-Net can be a viable approach for the ZUPT-aided INS system.

2019

1. IEEE Access’19

   

   Efficient Human Activity Recognition Solving the Confusing Activities Via Deep Ensemble Learning

   Ran Zhu, Zhuoling Xiao, Ying Li, Mingkun Yang, Yawen Tan, and 3 more authors

   IEEE Access, 2019

   Abs HTML

   The ubiquity of smartphones and their rich set of on-board sensors has created many exciting new opportunities, where smartphones are used as powerful computing platforms to sense and analyze pervasive data. One important application of mobile sensing is activity recognition based on smartphone inertial sensors, which is a fundamental building block for a variety of scenarios, such as indoor pedestrian tracking, mobile health care, and smart cities. Although many approaches have been proposed to address the human activity recognition problem, several challenges are still present: 1) people’s motion modes are very different for different individuals; 2) there is only a very limited amount of training data; 3) human activities can be arbitrary and complex, thus handcrafted feature engineering often fails to work; and 4) the recognition accuracy tends to be limited due to confusing activities. To tackle those challenges, in this paper, we propose a human activity recognition framework based on convolutional neural networks (CNNs) with two convolutional layers using the smartphone-based accelerometer, gyroscope, and magnetometer. To solve the confusion between highly similar activities like going upstairs and walking, this paper presents a novel ensemble model of CNN to further improve the identification accuracy. The extensive experiments have been conducted using 235 977 sensory samples from a total of 100 subjects. The results have shown that the classification accuracy of the proposed model can be up to 96.11%, which proves the effectiveness of the proposed model.