Juliacon 2024

Tsetlin.jl – Tsetlin Machines in Julia
07-10, 19:00–19:30 (Europe/Amsterdam), Struct (1.4)

We present Tsetlin.jl, a Julia implementation of Tsetlin Machines, which are interpretable machine learning models based on propositional logic. Tsetlin Machines have gained prominence by achieving high accuracy on a variety of problems, while also being efficient in terms of speed, memory use and energy consumption. While the original Tsetlin Machine works on binary data, later extensions can handle continuous and categorical data, and have found applications to images, text and speech. Tsetlin


In 2018, Prof. Ole Christoffer Granmo introduced the Tsetlin Machine and named it after Michael Lvovitch Tsetlin, who invented the Tsetlin automaton. Tsetlin automata are binary state machines that make decisions based on a reward and penalty scheme, and the Tsetlin Machine recognizes patterns by collaboratively organizing these automata into teams. The Tsetlin automata within a team vote whether a pattern is present or not in the input, and the collective decision determines the output.

Some advantages of Tsetlin Machines include their interpretability and inherent parallelizability. They have a low energy and memory footprint, yet they have equaled or outperformed state-of-the-art methods on various benchmarks.

The algorithms contained in Tsetlin.jl are mostly adapted from the Python versions in https://github.com/cair/TsetlinMachine/. As we get more experienced with Julia, we will try to make use of the strengths of the language to improve the efficiency and flexibility of the models.

Darshana completed his BSc in Mechatronics Engineering at AIT university, Thailand in 2015. Then he joined the Big Data research group at Thammasat University, Thailand for his MSc studies and graduated in 2017. Darshana completed his PhD in Computer Science, specializing in Machine Learning, from the University of Agder, Norway, in 2022. His research interests are in Tsetlin Machine, Artificial Neural Networks, Data Mining, Optimization, and Operations Research. Currently, at DNV, he is working as a senior researcher.