Abstract:Data privacy is a major concern in industries such as healthcare or finance. The requirement to safeguard privacy is essential to prevent data breaches and misuse, which can have severe consequences for individuals and organisations. Federated learning is a distributed machine learning approach where multiple participants collaboratively train a model without compromising the privacy of their data. However, a significant challenge arises from the differences in feature spaces among participants, known as non-IID data. This research introduces a novel federated learning framework employing fuzzy cognitive maps, designed to comprehensively address the challenges posed by diverse data distributions and non-identically distributed features in federated settings. The proposal is tested through several experiments using four distinct federation strategies: constant-based, accuracy-based, AUC-based, and precision-based weights. The results demonstrate the effectiveness of the approach in achieving the desired learning outcomes while maintaining privacy and confidentiality standards.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is dealing with the non - IID (non - independently and identically distributed) data in Federated Learning (FL), especially when both the feature space and the sample space are different. Specifically, the paper introduces a new federated learning framework combined with Fuzzy Cognitive Maps (FCMs) to address the following challenges: 1. **Data privacy protection**: In sensitive industries such as healthcare and finance, ensuring data privacy is crucial. Traditional centralized machine - learning methods require sharing actual data, which may lead to data leakage and abuse. Federated learning can perform collaborative training without sharing data. 2. **Non - IID data**: Data from different participants may come from different sources, resulting in different distributions in the feature space and the sample space. This data heterogeneity poses challenges to the training of federated learning models. 3. **Feature space differences**: Data sets of different participants may have different feature spaces or feature distributions, making it difficult for traditional horizontal or vertical federated learning methods to handle effectively. To solve these problems, the paper proposes the following innovations: - **Square Federated Learning combining horizontal and vertical federated learning**: By simultaneously handling changes in the sample space and the feature space, this method can perform effective model training when there are significant differences among participants. - **FCM learning method based on Particle Swarm Optimization (PSO)**: Specifically designed for data sets containing shared and non - IID features, it can better adapt to data differences among different participants. - **Four aggregation strategies**: Including weighted aggregation methods based on constants, accuracy, AUC, and precision to optimize the performance of the federated model. ### Formula summary - **Loss function of federated learning**: \[ L^*=\sum_{i = 1}^{n}\kappa_i\cdot L(D_i,\Psi) \] where \(\Psi\) is the model parameter, \(D_i\) is the data set of the \(i\)-th participant, \(L^*\) is the loss function of the federated model, \(L_i(\cdot)\) is the local model loss function of each participant, and \(\kappa_i\) is the weight of each participant. - **Parameter - average aggregation**: \[ \Phi'_j=\frac{1}{n}\sum_{i = 1}^{n}\Phi_{ji} \] - **Weighted aggregation based on accuracy**: \[ \Phi'_j=\sum_{i = 1}^{n}\frac{\text{acc}_{ji}}{\sum_{k = 1}^{n}\text{acc}_{jk}}\cdot\Phi_{ji} \] - **Weighted aggregation based on AUC**: \[ \Phi'_j=\sum_{i = 1}^{n}\frac{\text{AUC}_{ji}}{\sum_{k = 1}^{n}\text{AUC}_{jk}}\cdot\Phi_{ji} \] - **Weighted aggregation based on precision**: \[ \Phi'_j=\sum_{i = 1}^{n}\frac{\text{prec}_{ji}}{\sum_{k = 1}^{n}\text{prec}_{jk}}\cdot\Phi_{ji} \] Through these methods, the paper aims to achieve a federated learning framework that can both protect data privacy and effectively handle non - IID data.

Concurrent vertical and horizontal federated learning with fuzzy cognitive maps

Blind Federated Learning without initial model

Federated learning for preserving data privacy in collaborative healthcare research

Federated Learning Privacy: Attacks, Defenses, Applications, and Policy Landscape - A Survey

A privacy-preserving, distributed and cooperative FCM-based learning approach for Cancer Research

A Study of Federated Learning with Internet of Things for Data Privacy and Security using Privacy Preserving Techniques

Privacy-Preserving Federated Learning with Differentially Private Hyperdimensional Computing

Privacy-preserving federated machine learning on FAIR health data: A real-world application

Differential Privacy-enabled Federated Learning for Sensitive Health Data

Privacy-preserving Data Selection for Horizontal and Vertical Federated Learning

PrivacyFL: A simulator for privacy-preserving and secure federated learning

Towards Fair and Privacy Preserving Federated Learning for the Healthcare Domain

From distributed machine learning to federated learning: In the view of data privacy and security

Federated Learning: Navigating the Landscape of Collaborative Intelligence

Federated Learning: Balancing the Thin Line Between Data Intelligence and Privacy

MultiConfederated Learning: Inclusive Non-IID Data handling with Decentralized Federated Learning

Privacy-Preserving Heterogeneous Federated Learning for Sensitive Healthcare Data

Asymmetrical Vertical Federated Learning

PrivMaskFL: A private masking approach for heterogeneous federated learning in IoT

Privacy-preserving Heterogeneous Federated Transfer Learning.

An overview of implementing security and privacy in federated learning