Research
Distributed learning · Adaptive networks
Learning to collaborate.
Adapting to heterogeneity.
My research connects heterogeneous federated learning with multi-center medical imaging and adaptive control in Open RAN. These frameworks explore how clients share knowledge, how policies adapt, and how independent controllers work together.
클라이언트 간 데이터·구조 이질성을 고려한 연합학습과 다기관 의료영상 분석에서, 동적인 Open RAN 환경의 정책 적응과 xApp 협업까지 연구합니다.
Learning and orchestration for Open RAN
From adapting UE policies to coordinating network, edge, and xApp controllers.
Network–edge coordination through federated runtime contracts
Mediates independent radio, routing, batching, and compute proposals against end-to-end service objectives. Personalized impact models screen actions; independent closed-loop tests validate complete versioned execution policies.
Core idea Bind local execution authority to calibrated evidence, with bounded intervention and explicit invalidation.
Pandora
SubmittedPandora: Personalized Adaptive Network-Driven Open RAN Orchestration with Federated Contracts
IEEE INFOCOM 2027 Main Conference · 2027
Coordinates resource, steering, and QoS xApps through site-personalized feasible action regions. Federated impact models and calibrated risk screening inform contracts; weighted projection minimally adjusts incompatible joint proposals.
Core idea Learn where xApps can operate together, while preserving their own policies.
St-INTEL
Under ReviewSt-INTEL: Stackelberg-Intent Enhanced Learning based Resource Allocation in 5G Open RAN
Computer Communications - Special Issue: Intent-Based Agentic AI for Next-Generation Networking · 2026
Connects intent-aware MILP benchmarks and scarcity prices to UE-local DDQN request policies. Federated updates share experience across heterogeneous users, while periodic and event-triggered optimization refreshes network guidance.
Core idea Turn operator intent into slow-timescale guidance for fast, feasible request–grant control.
TFL-CORAN
First RevisionTransfer-enhanced Federated Learning with Dynamic Clustering for Traffic Management in 5G Open RAN
Computer Communications · 2026
Combines UE-local reinforcement learning with dynamic client grouping at the non-RT RIC. Graph embeddings and soft memberships personalize federated policies, while model transfer supports handovers and newly connected UEs.
Core idea Adapt traffic-control policies as radio conditions, client groups, and UE locations change.
Federated learning under heterogeneity
From variation in client updates to graph-based and relation-aware collaboration.
CarPe-FL
PublishedCausal Representation-Based Personalized Federated Learning with Causal Graph Consensus for Medical Imaging
Electronics - Special Issue: AI-Driven Medical Image/Video Processing · 2026
Uses latent proxy graphs to describe differences between medical institutions. Structure-aware clustering and acyclic graph consensus guide personalized representation learning, while local heads preserve site-specific prediction behavior.
Core idea Make shared structural dependencies a coordination prior for multi-center medical imaging.
FedHyDRA
Under ReviewFederated Learning with Dual-Scale Hybrid Divergence and Relation-Aware Embedding for Structured Non-IID Data
Applied Soft Computing · 2026
Models structured client heterogeneity through complementary label and feature distribution summaries. Hybrid divergence, graph representation learning, and soft community memberships guide collaboration under partial participation.
Core idea Represent overlapping client relationships beyond a single measure of distribution mismatch.
FedGCD
PublishedFedGCD: Federated Learning Algorithm with GNN based Community Detection for Heterogeneous Data
Journal of Internet Computing and Services (JICS) · 2023
Uses a graph view of federated clients to identify communities with related data characteristics. Community detection and membership-aware aggregation organize collaboration in heterogeneous, non-IID settings.
Core idea Make client relationships part of federated model aggregation.
FedVar
PublishedFedVar: Federated Learning Algorithm with Weight Variation in Clients
International Technical Conference on Circuits/Systems, Computers and Communications (ITC-CSCC) · 2022
Studies variation in client model weights as a signal for server-side aggregation. Statistics of client updates provide a lightweight way to account for heterogeneous local training behavior.
Core idea Use client-weight variation to inform aggregation under non-IID data.