Department of Computer, Control and Management Engineering

The emerging paradigm of Tsetlin machines provides a fundamental shift from arithmetic-based to logic-based machine learning. At the core, finite state machines, so-called Tsetlin automata, learn patterns using logical clauses, and these constitute a global description of the task learnt. In this way, Tsetlin machines introduce the concept of logical interpretable learning, where both the learned model and the process of learning are easy to follow and explain. The combination of finite state machines and logical rules is particularly suited for cutting-edge hardware solutions, enabling nano-scale intelligence, ultra-low energy consumption, energy harvesting, and unrivaled inference speed. Tsetlin machines offer competitive accuracy, scalability, speed, and energy efficiency across diverse tasks, including classification, convolution, regression, natural language processing, and speech understanding. In this short course, I cover the basics and recent advances of Tsetlin machines.