Sinchhean Phea の履歴(No.18)

Efficient Learning Algorithm for Privacy Preserving Collaboration Across Multi-Edge Nodes in AIRBiS

AIRBiS Overview†

AIRBiS

Background†

AIRBiS project goal is to do pneumonia detection and analysis by using Convolutional Neural Network(CNN) on lung x-ray images. But, due to the nature of the distribution of data amongst hospitals and privacy concerns for hospitals as well as patients, a conventional approach of centralized training would not work in this kind of setting. To solve this problem, Federated learning is adopted to collaborate across multi-edge nodes without sharing the raw data and achieve good decentralized training results.

↑

Research Goal†

Research and implementation of CNN (software) based on federated learning model that helps preserve the data privacy of users. It also reduces the burden of the aggregator (e.g., our FPGA-cluster, which is located in the hospital)

Implement the communication protocol between server and clients (hospital), which allows the local training and data (parameters) exchange.

Implement the user interface at client side, which controls the whole pipeline including the training, inference and communication procedures.

↑

Research Schedule†

Date	Task
☑️June 1 - June 30, 2020	Run demos, Read papers and books, Understand the basic concepts of CNN and federated learning
☑️July 1 - July 31, 2020	Do basic CNN and federated learning tutorials & implement own simple CNN and federated learning system
August 1 - September 30, 2020	Implement federated learning with the prototype CNN
September 20 - October 10, 2020	Prepare for mid-term presentation
October 1 - November 20, 2020	Implement communication between FL system with UI & Optimization
October 10 - November 20, 2020	Write paper

↑

Researech Progress†

06/29/2020: pdf
07/31/2020: pdf

↑

References†

My Shared Google Drive