A Methodology for Transparent Logic-Based Classification Using a Multi-Task Convolutional Tsetlin Machine

TL;DR

This work addresses the need for transparent, high-performing image classifiers by extending Tsetlin Machines with convolutional and coalesced architectures. It introduces a methodology to derive local explanations and global class representations from convolutional clauses, enabling direct mapping of predictions to input features and highlighting class-specific patterns. Empirically, Convolutional CoTM achieves competitive accuracy on MNIST (98.5%) and CelebA (86.56% F1) while maintaining interpretability on large-scale, multi-channel data. The approach offers a hardware-friendly, interpretable alternative to deep nets, with practical value for bias detection, model validation, and applications requiring explainable AI.

Abstract

The Tsetlin Machine (TM) is a novel machine learning paradigm that employs finite-state automata for learning and utilizes propositional logic to represent patterns. Due to its simplistic approach, TMs are inherently more interpretable than learning algorithms based on Neural Networks. The Convolutional TM has shown comparable performance on various datasets such as MNIST, K-MNIST, F-MNIST and CIFAR-2. In this paper, we explore the applicability of the TM architecture for large-scale multi-channel (RGB) image classification. We propose a methodology to generate both local interpretations and global class representations. The local interpretations can be used to explain the model predictions while the global class representations aggregate important patterns for each class. These interpretations summarize the knowledge captured by the convolutional clauses, which can be visualized as images. We evaluate our methods on MNIST and CelebA datasets, using models that achieve 98.5\% accuracy on MNIST and 86.56\% F1-score on CelebA (compared to 88.07\% for ResNet50) respectively. We show that the TM performs competitively to this deep learning model while maintaining its interpretability, even in large-scale complex training environments. This contributes to a better understanding of TM clauses and provides insights into how these models can be applied to more complex and diverse datasets.

Figures (11)

• Figure 1: A TA with $2N$ states and two actions
• Figure 2: Structure of a convolutional clause
• Figure 3: Combining convolutional clauses for interpretation
• Figure 4: Random samples from the CelebA dataset.
• Figure 5: The distribution of the classes in the CelebA dataset. The orange bars indicate the classes selected for the experiment.