Introduction

Official documentation for Diagnokare's Federated Learning platform

Our federated learning platform allows organisations to collaboratively train deep learning models while maintaining data privacy and security. Unlike traditional centralized machine learning, where all data is collected in a central location for model training, federated learning enables organizations to train models on decentralized data sources without exposing sensitive data to unauthorized parties.

The following guide explains how to use Diagnokare's platform to train or validate your own model with decentralized data sources. We verify the usability and authenticity of the data at each client site in our network using properietary algorithms which detect the uniqueness of data points, accuracy of annotations and image quality of scans which are necessary in a federated scenario. Following is a summary of tested capabilities and constrains of our platform's v1.0.0:

• Language: Python

• Domain: Computer Vision

• Integrations: AWS, Azure, On-Prem

• Deep learning framework: PyTorch

• Federated Learning Algorithm: FedAvg

• Federation Workflow: Scatter & Gather

• Evaluation: Global Model Evaluation, Cross-Site Model Evaluation

If you have any custom needs for your particular usecase, please feel free to contact us for enquiry.