High Performance and Lightweight Single Semi-Lattice Algebraic Machine Learning

Abstract

Algebraic machine learning is a novel parameter-free model that has demonstrated impressive accuracy in challenging tasks such as the MNIST dataset and N-Queens completion. However, its utilization of two semi-lattices can lead to significant computational demands. To tackle this issue, a solution has been proposed that employs a single semi-lattice model, resulting in reduced memory requirements and improved efficiency. This research endeavors to bridge the gap between algebraic concepts and programming concepts, thereby making them more accessible to a broader range of researchers. This paper presents the development of a lightweight single semi-lattice algebraic model. The implementation achieved remarkable accuracy on the MNIST dataset of handwritten digits, with an error rate of 2.7%, a False Positive Rate (FPR) of 2.5%, and a False Negative Rate (FNR) of 4.4%. This work holds great significance due to its ability to explain the algebraic model in a simpler manner compared to the original work, while also serving as a proof of concept. Additionally, it studies the memory and time performance of the novel model with an average of 383 MB and 3 hours per digit.

Keywords

• Algebraic machine learning
• generalization
• distributed artificial intelligence
• data conceptualisation
• error function minimization