Decentralized Federated Learning with Non-IID Data: Challenges, Trends, and Future Opportunities

As artificial intelligence and machine learning advance, increasing privacy concerns and regulatory constraints have limited cross-border data sharing for traditional model training. Federated Learning (FL) offers a privacy-preserving approach by enabling distributed training without exposing raw data. However, FL faces significant challenges, particularly when dealing with Non-Independent and Identically Distributed (Non-IID) data, which results in inconsistent model performance. Moreover, relying on a central server also raises reliability and scalability issues. Decentralized Federated Learning (DFL) eliminates the central server, thereby fostering more robust and scalable collaboration. Despite the growing interest in DFL, a comprehensive review focusing on Non-IID challenges remains scarce. This article presents a Systematic Literature Review (SLR) of existing research on DFL under Non-IID settings. Studies were retrieved from six major academic publishers and categorized into four pillars: architecture, topology, optimization, and security. The SLR review provides insights into current trends and systematically summarizes real-world applications, commonly used datasets, and neural network models. This article also examines prevalent methods for conducting Non-IID experiments and evaluating performance metrics. By providing a structured analysis of the literature, experimental setups, and evaluation practices, this survey highlights key trends, uncovers research gaps, and proposes future directions for advancing DFL in Non-IID environments.