Research

Research Overview

Our research centers on Edge AI, aiming to fulfill various user demands on edge nodes (e.g., mobile and IoT devices) through scalable, trustworthy, and efficient AI/ML services. Grounded in both theoretical and empirical foundations, we develop new AI/ML algorithms (applicable to both vision and language tasks) tailored to the following research areas:

• Distributed/Federated Learning 

• Personalized, Robust, and Trustworthy AI

• Data/Resource-Efficient Learning

Detailed descriptions of each research area are provided below.

1. Distributed/Federated Learning 

With the explosive growth in the number of smartphones and IoT devices, a vast volume of data is being collected by distributed users at the edge. However, these clients may not want to share their own privacy-sensitive data (e.g., medical data) to others. This strongly motivates the need for distributed/federated learning solutions that enable collaborative model training. In the current era of foundation models, there is also a growing demand to fine-tune large language models (LLMs) across distributed clients. We focus on both the algorithmic and theoretical aspects of improving large-scale distributed/federated learning.

• [ICLR'25] Decentralized Sporadic Federated Learning: A Unified Algorithmic Framework with Convergence Guarantees [paper]

• [ICLR'25] Asynchronous Federated Reinforcement Learning with Policy Gradient Updates: Algorithm Design and Convergence Analysis [paper]

• [NeurIPS'24] Hierarchical Federated Learning with Multi-Timescale Gradient Correction [paper]

• [NeurIPS'23] StableFDG: Style and Attention Based Learning for Federated Domain Generalization [paper]

• [NeurIPS'21] Few-Round Learning for Federated Learning [paper]

• [NeurIPS'21] Sageflow: Robust Federated Learning against Both Stragglers and Adversaries [paper]

• [NeurIPS'20] Election Coding for Distributed Learning: Protecting SignSGD against Byzantine Attacks [paper]

2. Personalized, Robust, and Trustworthy AI

Providing personalized, robust, and trustworthy AI services to every edge user is one of the most rewarding challenges in the current AI/ML era. In particular, models should be personalized to each user's preference while also effectively handling out-of-distribution scenarios (i.e., generalization/robustness to unseen domains and adversarial attacks). One of our key directions is to utilize foundation models (e.g., LLMs) to effectively address these challenges by simultaneously capturing personalization and generalization capabilities.

• [ICLR'25] Unlocking the Potential of Model Calibration in Federated Learning [paper]

• [ICLR'25] Adaptive Energy Alignment for Accelerating Test-Time Adaptation [paper]

• [AAAI'25] Rethinking the Starting Point: Collaborative Pre-Training for Federated Downstream Tasks [paper]

• [AAAI'24] Consistency-Guided Temperature Scaling using Style and Content Information for Out-of-Domain Calibration [paper]

• [INFOCOM'23] SplitGP: Achieving Both Generalization and Personalization in Federated Learning [paper]

• [ICML'23] Test-Time Style Shifting: Handling Arbitrary Styles in Domain Generalization [paper]

3. Data/Resource-Efficient Learning

In practice, it is crucial to facilitate efficient training and inference on data/resource-constrained edge devices. Specifically, each user should be able to handle issues related to limited data, limited labels, and limited computation resources during training. In this direction, our goal is to develop efficient algorithms (e.g., efficient fine-tuning, model/gradient compression, few-shot/active learning) to facilitate lightweight learning and inference using large-scale foundation models.

• [ICLR'25] PRISM: Privacy-Preserving Improved Stochastic Masking for Federated Generative Models [paper]

• [ICML'24] Achieving Lossless Gradient Sparsification via Mapping to Alternative Space in Federated Learning [paper]

• [NeurIPS'23] NEO-KD: Knowledge-Distillation-Based Adversarial Training for Robust Multi-Exit Neural Networks [paper]

• [ICLR'23] Warping the Space: Weight Space Rotation for Class-Incremental Few-Shot Learning [paper]

• [ICLR'23] Active Learning for Object Detection with Evidential Deep Learning and Hierarchical Uncertainty Aggregation [paper]

+ AI/ML Optimization over Edge Networks

Finally, we have also been actively exploring the above three aspects of AI/ML over wireless communication networks (e.g., 6G satellite networks). This research direction has a significant demand for the delivery of intelligent AI/ML services in real-world network edge.

• [MobiHoc'25]  Local-Cloud Inference Offloading for LLMs in Multi-Modal, Multi-Task, Multi-Dialogue Settings

• [ToN'25] Federated Learning over Hierarchical Wireless Networks: Training Latency Minimization via Submodel Partitioning

• [JSAC'24] Orchestrating Federated Learning in Space-Air-Ground Integrated Networks: Adaptive Data Offloading and Seamless Handover

• [JSAC'24] Cooperative Federated Learning over Ground-to-Satellite Integrated Networks: Joint Local Computation and Data Offloading

• [TMC'24] Federated Split Learning with Joint Personalization-Generalization for Inference-Stage Optimization in Wireless Edge Networks 

• [JSAC'21] FedMes: Speeding Up Federated Learning with Multiple Edge Servers

• [INFOCOM'21] TiBroco: A Fast and Secure Distributed Learning Framework for Tiered Wireless Edge Networks

• [TWC'21] Hierarchical Broadcast Coding: Expediting Distributed Learning at the Wireless Edge