Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Trust and Resilience in Federated Learning Through Smart Contracts Enabled Decentralized Systems

Lorenzo Cassano, Jacopo D'Abramo, Siraj Munir, Stefano Ferretti

TL;DR

This paper addresses trust, reliability, and privacy in federated learning by integrating IPFS-based storage and a smart contract-driven coordination layer on a permissioned blockchain. It evaluates two weight update schemes, FedAvg and FedProx, under varying collaborator participation and failure scenarios on two healthcare imaging datasets (Alzheimer and Brain Tumor), and measures model accuracy, F1, ROC, as well as system metrics. The results show that FedProx is close to centralized performance and more robust to nonuniform participation, while IPFS and the smart contract introduce manageable overheads in gas and delays, enabling auditable and tamper-resistant training. The work demonstrates a viable path toward trustworthy FL in privacy-sensitive domains and outlines future directions such as LoRA, MPC, and decentralized leadership.

Abstract

In this paper, we present a study of a Federated Learning (FL) system, based on the use of decentralized architectures to ensure trust and increase reliability. The system is based on the idea that the FL collaborators upload the (ciphered) model parameters on the Inter-Planetary File System (IPFS) and interact with a dedicated smart contract to track their behavior. Thank to this smart contract, the phases of parameter updates are managed efficiently, thereby strengthening data security. We have carried out an experimental study that exploits two different methods of weight aggregation, i.e., a classic averaging scheme and a federated proximal aggregation. The results confirm the feasibility of the proposal.

Trust and Resilience in Federated Learning Through Smart Contracts Enabled Decentralized Systems

TL;DR

This paper addresses trust, reliability, and privacy in federated learning by integrating IPFS-based storage and a smart contract-driven coordination layer on a permissioned blockchain. It evaluates two weight update schemes, FedAvg and FedProx, under varying collaborator participation and failure scenarios on two healthcare imaging datasets (Alzheimer and Brain Tumor), and measures model accuracy, F1, ROC, as well as system metrics. The results show that FedProx is close to centralized performance and more robust to nonuniform participation, while IPFS and the smart contract introduce manageable overheads in gas and delays, enabling auditable and tamper-resistant training. The work demonstrates a viable path toward trustworthy FL in privacy-sensitive domains and outlines future directions such as LoRA, MPC, and decentralized leadership.

Abstract

In this paper, we present a study of a Federated Learning (FL) system, based on the use of decentralized architectures to ensure trust and increase reliability. The system is based on the idea that the FL collaborators upload the (ciphered) model parameters on the Inter-Planetary File System (IPFS) and interact with a dedicated smart contract to track their behavior. Thank to this smart contract, the phases of parameter updates are managed efficiently, thereby strengthening data security. We have carried out an experimental study that exploits two different methods of weight aggregation, i.e., a classic averaging scheme and a federated proximal aggregation. The results confirm the feasibility of the proposal.
Paper Structure (16 sections, 1 equation, 10 figures, 5 tables)

This paper contains 16 sections, 1 equation, 10 figures, 5 tables.

Table of Contents

  1. Introduction
  2. System Architecture
  3. System Interaction Overview
  4. The Federated Learning Smart Contract
  5. On the Security and Reliability of the Protocol
  6. Experimental Evaluation
  7. The Federated Learning Model and Weight Update Methods
  8. Employed datasets
  9. Alzheimer Dataset
  10. Brain Tumor Dataset
  11. Results
  12. FL performance: Alzheimer Results
  13. FL performance: Brain Tumor Results
  14. Federated Learning Smart Contract Gas Consumption
  15. IPFS Delays
  16. ...and 1 more sections

Figures (10)

  • Figure 1: System diagram
  • Figure 2: ROC curves for Alzheimer dataset. Number of Collaborators = 10
  • Figure 3: Alzheimer dataset. Accuracy and F1 score performances with a varying number of Collaborators
  • Figure 4: Alzheimer dataset. Accuracy and F1 score performances with node failures
  • Figure 5: ROC curves for Brain Tumor dataset. Number of Collaborators = 5
  • ...and 5 more figures