JiRAIYA framework uses Web3 to make federated learning transparent
Reputation-based hierarchy and consensus prevent poisoning attacks in FL.
A new academic paper proposes JiRAIYA, a federated learning framework that leverages Web3 technologies to address transparency and trust deficits in traditional FL deployments. Authors Venkata Raghava Kurada and Pallav Kumar Baruah describe a hierarchical architecture where delegated managers oversee training within federations. Model updates are encoded and broadcast to all managers, who independently evaluate validity; only consensus-approved updates enter the global model. A reputation-based backup mechanism further ensures model generation, mitigating adversarial and poisoning attacks without relying on opaque aggregation or external validators.
Extensive real-world experiments demonstrate JiRAIYA's effectiveness and resilience compared to conventional FL systems. The framework's novelty lies in combining novelty detection with decentralized consensus—reducing computational overhead while boosting auditability. By making the entire training process transparent on Web3, JiRAIYA could extend federated learning beyond enterprise settings to open, collaborative environments where participants demand verifiable model integrity.
- Uses a hierarchical structure with delegated managers to orchestrate FL training within federations.
- Combines novelty detection and Web3 consensus to evaluate and approve only valid model updates.
- Reputation-based backup mechanism ensures model generation even under adversarial conditions.
Why It Matters
Brings blockchain-level transparency to federated learning, enabling trust for non-enterprise collaborations and public AI training.