The Problem It Solves

Federated learning lets multiple parties train a machine learning model together without sharing their data. However, it faces issues

Malicious Participants: Some users might submit bad updates to the
Weak Security: Security is spread out, making it easier to
Lack of Trust: No unified way to ensure everyone is
Data Privacy Risks: Ensuring that the training process does not compromise the privacy of the underlying data, especially when using homomorphic encryption.

EigenSecure wants to address these challenges

Pooling Security: Combines the security efforts of all participants, enhancing overall network robustness.
Validating Updates: Ensures only legitimate and beneficial updates are incorporated into the model.
Economic Incentives: Penalizes bad actors by slashing their stake, encouraging honest participation and deterring malicious
Secure Training Pipeline: Implements an automated pipeline that automatically accepts model trainings performed on encrypted data and ensures that the execution code adheres to privacy constraints by preventing any attempts to decrypt homomorphic components. This guarantees that data remains secure and private throughout the training process.