Oleksandr Gordieiev

This paper investigates the performance of the Tsetlin Machine (TM) on the German Traffic Sign Recognition Benchmark (GTSRB), a real-world dataset collected under diverse weather and lighting conditions. We examined the efficiency of different image Booleanisation techniques and ensembling methods, evaluating their impact on TM performance. Leveraging the regular structure of traffic signs, the TM demonstrated strong pattern recognition capabilities in noisy environments. Using adaptive mean thresholding for Booleanisation, a TM with only 100 clauses per class achieved 91.89% test accuracy and over 34,000 predictions per second, without GPU acceleration. Further improvements were achieved through ensemble learning across three data modalities: thresholded images, HOG descriptors, and Haar features, boosting accuracy to 96.34%, all without data augmentation. While the primary focus of the paper was not to achieve maximum accuracy, the results highlight TM’s practical viability for efficient, interpretable, and scalable visual classification and pattern recognition.

Human activity recognition (HAR) based on smartphone and wearable sensor data is commonly addressed using statistical learning methods and deep neural networks that often provide strong predictive performance, but at the expense of limited interpretability and substantial computational and energy requirements. Such limitations reduce their suitability for deployment in practical sensing environments where model decisions must be transparent, verifiable and executable on resource-constrained devices. In this work, we investigate the Convolutional Tsetlin Machine (CTM) for multimodal HAR using only the raw inertial signals (9 × 128) of the UCI-HAR dataset, rather than its pre-computed 561-feature representation. The Tsetlin Machine is a novel neuro-symbolic machine learning approach that offers two important advantages over many conventional machine learning methods: (i) it learns logic-based decision rules that support human inspection and provide a transparent basis for analyzing model decisions, and (ii) it operates with comparatively low computational complexity, making it well suited to efficient and low-power on-device learning. The proposed study systematically analyses the contribution of different feature modalities by decomposing the inertial signals space into semantically defined subsets according to: (i) sensor source: accelerometer and gyroscope; (ii) signal group: gyroscope angular velocity, body and total acceleration (including gravity); (iii) coordinate axis: x, y and z. A separate CTM classifier was trained for each modality and its combinations in order to determine the relative discriminative value of each modality group for activity classification. In addition to predictive performance, the study emphasizes the interpretability of the CTM model ensured by expressing each decision in the form of propositional clauses, thereby enabling visualization and direct inspection of the modality-specific patterns supporting each activity class. Owing to its symbolic structure and modest computational demands, the CTM provides a principled framework for the design of explainable, resource-efficient and deployable HAR systems. The proposed work therefore contributes toward trustworthy multimodal sensing by jointly addressing predictive performance, interpretability and suitability for embedded and mobile platforms.

This paper addresses the growing need for reliable methods to evaluate the quality of artificial intelligence (AI) systems as they become widely used in both critical domains and everyday applications. The study aims to develop a metric-based approach to assessing AI system quality by harmonising product quality and quality in use models in line with updated international standards. To achieve this, the authors analyse existing ISO/IEC 25000 series standards, identifying inconsistencies between older and newer versions, and propose an updated quality model that incorporates both perspectives. Building on guidance documents, international standards, and contemporary research, the study introduces a set of metrics designed to measure new subcharacteristics of AI system quality, particularly where standardised metrics have not yet been developed. The proposed approach bridges the gap between established quality models (ISO/IEC 25010:2023, ISO/IEC 25019:2023, ISO/IEC 25059:2023) and standardised measurement practices (ISO/IEC 25023:2016, ISO/IEC 25022:2016), enabling more consistent and practical evaluation of AI systems. These metrics can be applied by researchers and practitioners to improve the quality of AI systems, enhance their reliability, and reduce risks associated with insufficient quality. Future work will focus on empirical validation of the proposed approach to confirm its applicability and usefulness across diverse AI applications.

This paper investigates the performance of the Tsetlin Machine (TM) on the German Traffic Sign Recognition Benchmark (GTSRB), a real-world dataset collected under diverse weather and lighting conditions. We examined the efficiency of different image Booleanisation techniques and ensembling methods, evaluating their impact on TM performance. Leveraging the regular structure of traffic signs, the TM demonstrated strong pattern recognition capabilities in noisy environments. Using adaptive mean thresholding for Booleanisation, a TM with only 100 clauses per class achieved 91.89% test accuracy and over 34,000 predictions per second, without GPU acceleration. Further improvements were achieved through ensemble learning across three data modalities: thresholded images, HOG descriptors, and Haar features, boosting accuracy to 96.34%, all without data augmentation. While the primary focus of the paper was not to achieve maximum accuracy, the results highlight TM's practical viability for efficient, interpretable, and scalable visual classification and pattern recognition.

Human activity recognition (HAR) based on smartphone and wearable sensor data is commonly addressed using statistical learning methods and deep neural networks that often provide strong predictive performance, but at the expense of limited interpretability and substantial computational and energy requirements. Such limitations reduce their suitability for deployment in practical sensing environments where model decisions must be transparent, verifiable and executable on resource-constrained devices. In this work, we investigate the Convolutional Tsetlin Machine (CTM) for multimodal HAR using the UCI-HAR dataset. The Tsetlin Machine is a novel neuro-symbolic machine learning approach that offers two important advantages over many conventional machine learning methods: (i) it learns logic-based decision rules that are human-readable and formally verifiable, and (ii) it operates with comparatively low computational complexity, making it well suited to efficient and low-power on-device learning. The proposed study systematically analyses the contribution of different feature modalities by decomposing the inertial signals space into semantically defined subsets according to: (i) sensor source: accelerometer or gyroscope; (ii) physical component: body or gravity; (iii) coordinate: x, y or z. A separate CTM classifier was trained for each modality and their combination in order to determine the relative discriminative value of each modality group for activity classification. In addition to predictive performance the study emphasizes the interpretability of the CTM model ensured by expressing each decision in the form of propositional clauses, thereby enabling visualization and direct inspection of the modality-specific patterns supporting each activity class. Owing to its symbolic structure and modest computational demands, the CTM provides a principled framework for the design of explainable, resource-efficient and deployable HAR systems. The proposed work therefore contributes toward trustworthy multimodal sensing by jointly addressing predictive performance, interpretability and suitability for embedded and mobile platforms.