Blockchain-based Asynchronous Federated Learning for Internet of Things

摘要: As an emerging distributed machine learning (ML) technology, federated can protect data privacy through collaborative AI models across a large number of IoT devices. However, inefficiency and vulnerability to poisoning attacks have slowed performance. To solve the above problems, blockchain-based asynchronous framework (BAFL) is proposed pursue both security efficiency. Blockchain ensures that cannot be tampered with secured while asynchrony speeds up global aggregation. In further, we propose concept device's score use entropy weight method measure quality model update. The design directly determines proportion in aggregation allowed local update delay. By analyzing optimal block generation rate, paper also balances equipment energy consumption delay by adjusting training communication extensive evaluation results show BAFL has performs better aspects efficiency anti-poisoning than other ML methods.