Loading [a11y]/accessibility-menu.js
Logic-Based Machine Learning with Reproducible Decision Model Using the Tsetlin Machine | IEEE Conference Publication | IEEE Xplore

Logic-Based Machine Learning with Reproducible Decision Model Using the Tsetlin Machine


Abstract:

Tsetlin Machine (TM) is a recent automaton-based algorithm for reinforcement learning. It has demonstrated competitive accuracy on many popular benchmarks while providing...Show More

Abstract:

Tsetlin Machine (TM) is a recent automaton-based algorithm for reinforcement learning. It has demonstrated competitive accuracy on many popular benchmarks while providing a natural interpretability. Due to its logically underpinning it is amenable to hardware implementation with faster performance and higher energy efficiency than conventional Artificial Neural Networks (ANNs). This paper provides an overview of Tsetlin Machine architecture and its hyper-parameters as compared to ANN. Furthermore, it gives practical examples of TM application for patterns recognition using MNIST dataset as a case study. In this work we also prove reproducibility of TM learning process to confirm its trustworthiness and convergence in the light of the stochastic nature of TAs reinforcement.
Date of Conference: 07-09 September 2023
Date Added to IEEE Xplore: 21 December 2023
ISBN Information:

ISSN Information:

Conference Location: Dortmund, Germany

I. Introduction

Machine learning has become an essential tool for analyzing complex data and making predictions in various fields, including finance, healthcare, and technology. Artificial neural networks have been the dominant machine learning technique in recent years due to their high accuracy in predictive tasks. While ANNs (especially deep neural networks, DNNs) have been successful in many application domains, they have certain limitations, such as high complexity and lack of interpretability. These limitations have led researchers to explore alternative machine learning methods, including the Tsetlin machine (TM), a relatively new logic-based approach, proposed by Granmo in 2018 [1]. Recent studies have shown that TM provides a promising alternative to DNNs with several advantages. TM is an interpretable, low complexity algorithm and has a unique logic-based learning mechanism supporting parallelism that makes TM suitable for native hardware implementation, which promises much better performance than traditional DNNs. TM has been actively developed over the last few years and demonstrated competitive accuracy on a number of benchmarks [2]. In summary, higher energy efficiency, native hardware support and design productivity make TM attractive for embedded applications and hardware acceleration [3], [4].

Contact IEEE to Subscribe

References

References is not available for this document.