Distributed Tsetlin Machine-Based Learning for Energy-Efficient Wireless Sensor Networks*

Abstract

This work-in-progress paper proposes a two-level intelligence framework for improving energy efficiency and service continuity in large-scale wireless sensor networks (WSN) operating under QoS constraints and asynchronous node behavior. At the edge level, each sensor node embeds lightweight, interpretable decision logic based on Tsetlin Machine clauses to regulate sensing and communication actions. At the global level, a centralized base station supervises learning indirectly by reconstructing the underlying spatio-temporal sensed field from sparse transmissions using a deep learning model, and by using reconstruction quality as a performance signal for network-level optimization through reinforcement learning. As an initial concrete step toward this paradigm, we present a current implementation where a Tsetlin Machine is instantiated as a sampling policy that learns sparse acquisition masks under a quality reconstruction–sparsity trade-off. We also introduce a dedicated, controllable WSN simulator designed for high-throughput experimentation and report preliminary validation results, including an ns-3-aligned comparison showing comparable network-level statistics and a promising runtime speedup.