Explainable AI and TinyML for Trustworthy and Efficient IoT: A Survey

Abstract

The convergence of Explainable Artificial Intelligence (XAI) and Tiny Machine Learning (TinyML) is emerging as a key enabler for trustworthy and efficient Internet of Things (IoT) systems. TinyML allows real-time intelligence to be deployed directly on resource-constrained edge devices, while XAI enhances transparency by providing interpretable insights into model decisions. This survey presents a comprehensive review of recent advances at the intersection of XAI, TinyML, and IoT, examining how explainability can be achieved under strict computational and hardware limitations. A structured taxonomy of TinyML is presented, along with an overview of the end-to-end integration workflow of XAI-enabled TinyML within edge IoT architectures. In addition, state-of-the-art studies are comparatively analyzed in terms of learning models, explainability techniques, hardware platforms, optimization strategies, and application domains. The survey identifies major challenges, such as severe resource constraints, XAI computational overhead, accuracy degradation due to model compression, and the lack of standardized benchmarking practices, and underscores the need for lightweight, interpretable, and standardized solutions for safety-critical and large-scale IoT deployments.