You are the Lead Data Architect for a healthcare consortium. The goal is to train a predictive model for patient readmission rates using data from 10 different hospitals, but strict regulations (HIPAA/GDPR) prohibit raw patient data from leaving the local hospital servers. Develop a detailed implementation plan using Federated Learning. Your plan should cover: 1) The model aggregation strategy (e.g., FedAvg). 2) Handling of non-IID (Independent and Identically Distributed) data across hospitals. 3) Mechanisms to detect and mitigate adversarial attacks or poisoning from compromised nodes. 4) Communication overhead and bandwidth optimization techniques.