~2024

1. Federated Learning

YouTube Video, Flower Summit 2023 | FedOps: Federated Learning Lifecycle Operations Management ↗
YouTube Video, Enhancing Federated Learning with FedOps for the FL Marketplace (Flower AI Summit 2024) ↗
YouTube Video, Poster Presentations Session 1 (Flower AI Summit 2024) ↗
  • FedOps Overview

- FedOps was designed and implemented to enable FL lifecycle operations management by extending the existing MLOps concept.

- FedOps is to manage the entire FL process of creating a global model by deploying and training the local model created in the experimental environment.

FedOps Overview - FedOps was designed and implemented to enable FL lifecycle operations management by extending the existing MLOps concept.
  • FedOps Detail

1. Easily application of data and models to the FL environment  (FL Scalize)

2. Continuously check and manage FL Client/Server state (Manager)

3. Contribution Evaluation and Client Selection to give Incentive to Client (CE/CS & BCFL)

4. Easily deploy FL Client/Server and perform periodic FL rounds (CI/CD/CFL)

5. FL Client/Server monitoring dashboard for FL Lifecycle Visualization (Monitoring)

FedOps Detail 1. Easily application of data and models to the FL environment (FL Scalize) 2. Continuously check and manage FL Client/Server
  • FedOps Demo

- Kubernetes-based Simulation

Client CI/CD Deploying and running client pods using ArgoCD

Client CI/CD
Deploying and running client pods using ArgoCD

Creating FL tasks & CFL Creating FL tasks to support a workflow using Airflow that creates an FL round using Airflow

Creating FL tasks & CFL
Creating FL tasks to support a workflow using Airflow that creates an FL round using Airflow

FL Lifecycle Monitoring Performance and time cost monitoring of global model and local models using ELK

FL Lifecycle Monitoring
Performance and time cost monitoring of global model and local models using ELK

- Distributed Real Device

Docker-based FL D eployment to R eal D evices Monitoring the health of data on your device and the performance and time cost of local mode l
Docker-based FL D eployment to R eal D evices Monitoring the health of data on your device and the performance and time cost of local mode l

Docker-based FL Deployment to Real Devices
Monitoring the health of data on your device and the performance and time cost of local model using ELK

  • Ongoing National Project (2022.7 ~ 2024.12)
    : Ministry of Science and Technology Employment Technology Promotion Agency

- End Goal
  Blockchain Convergence Federated Learning Remote Clinical Trial Service for Digital Healthcare

- Utilization Plan
“Blockchain Convergence Federated Learning” developed through this project is used as a core technology for remote clinical trials targeting Digital Therapeutics(DTx), Digital Patient Monitoring, and Software as Medical Device(SaMD)

- Participating Institutions

Ongoing National Project (2022.7 ~ 2024.12) : Ministry of Science and Technology Employment Technology Promotion Agency - End Goal Blockchai

2. TAVR Medical Twin based on RWD

  • Post-TAVR Complications

Post-TAVR Complications
  • The FL Post-TAVR Complication Prediction System

The FL Post-TAVR Complication Prediction System

3. Sensor-based Behavior Recognition and Risk Detection

  • Real-Time Risk Prediction for Behavioral Management in Psychiatric Wards 

Real-Time Risk Prediction for Behavioral Management in Psychiatric Wards
  •  Comprehensive Symptom Prediction in Inpatients With Acute Psychiatric Disorders 

Comprehensive Symptom Prediction in Inpatients With Acute Psychiatric Disorders