Assessing the Sustainability and Trustworthiness of Federated Learning Models

Abstract: Artificial intelligence is widely used in various sectors and significantly impacts decision-making processes. Novel AI paradigms, such as Federated Learning (FL), focus on training AI models collaboratively while preserving data privacy. In such a context, the European Commission's AI-HLEG group has highlighted the importance of sustainable AI for trustworthy AI. While existing literature offers several solutions for assessing the trustworthiness of FL models, a significant gap exists in considering sustainability associated with FL. Thus, this work introduces the sustainability pillar to the trustworthy FL taxonomy, making this work the first to address all AI-HLEG requirements. The sustainability pillar assesses the FL system's environmental impact, incorporating notions and metrics for hardware efficiency, federation complexity, and energy grid carbon intensity. An algorithm is developed to evaluate the trustworthiness of FL models, incorporating sustainability considerations. Extensive evaluations with the FederatedScope framework and various scenarios demonstrate the effectiveness of the proposed solution.