The School of Computer Science is pleased to present…

A Blockchain-based Data Evaluation Technique in Federated Learning 

MSc Thesis Proposal by: Laveen Bhatia 

Date: Friday November 18, 2022

Time: 11:00am-12:30pm

Location:  Essex Hall, Room 122

Reminder: Two-part attendance mandatory, arrive 5-10 minutes prior to event starting – LATECOMERS WILL NOT BE ADMITTED once the door has been closed and the presentation has begun. If the door is closed, please be respectful to Laveen and his committee by NOT knocking on the door for admittance.

Abstract:  

Federated Learning (FL) is a type of distributed Deep Learning framework, in which the model is trained locally on each device and the trained gradients are sent to a central server which aggregates them and creates a global model. This helps ensure the data privacy of the user as the data never leaves the local device. However, a major concern in Federated Learning is ensuring the data quality of trained data. A model trained on poor quality data can have a major impact on its accuracy. In this

proposal, we propose a decentralized approach using blockchain to counter this issue. We use a set of miners in blockchain who act as validators for data. They validate the data quality, check for correct labels, and ensure the model is protected from data poisoning attacks. Both the validators and trainers are awarded for their contribution using smart contracts. We also devise a point system, where a trainer will be awarded more points if they train with good quality data and points will be reduced for poor quality data.

Keywords: Federated Learning, Blockchain, Data Evaluation, Smart Contracts

MSc Thesis Committee:  

Internal Reader: Dr. Pooya Moradian Zadeh

External Reader: Dr. Mohammad Hassanzadeh

Advisor: Dr. Saeed Samet