Logic Gate Networks (LGNs) are an appealing family of artificial intelligence models that efficiently perform inference, by replacing floating-point multiply-accumulate operations with Boolean ones. Despite their performance and efficiency, training still represents a problem: the selection and wiring of logic gates are discrete design choices, making the direct use of gradient-based optimization difficult, at least in its native form. This paper provides an initial investigation on whether Cartesian Genetic Programming (CGP) can be turned into a practical, gradient-free framework for evolving layered LGNs for image classification. To this purpose, we extend a CGP implementation with data handling functions specific for the modified National Institute of Standards and Technology (MNIST) dataset, redundant multi-bit output decoding, custom loss functions, batch-based evaluation, layered connectivity constraints, and adaptive mutation control. Experiments on binarized MNIST show that these design choices substantially improve the search process. In the final configuration, the proposed method reaches more than 80% test accuracy with a fully discrete evolutionary pipeline executed on a single processor core (CPU), providing a concrete indication that non-trivial logical classifiers can be obtained without back-propagation, Graphics Processing Units (GPUs), or continuous relaxations.
Francesco Lubrano, Emanuel Messetti, A. Scionti et al.· Proceedings of the Genetic a...· 0 citations
Search Trajectory Networks (STNs) are a graph-based tool for visualizing and characterizing the behavior of optimization algorithms. STNs'reliance on discretization of the search space has largely confined them to low-dimensional or combinatorial settings. We introduce a methodology for constructing STNs in semantic spaces, defined as the space of a model's predictions on a fixed sample set. Our approach discretizes semantic vectors and aggregates them into network nodes via agglomerative clustering with complete linkage under a normalized Hamming distance. Since any predictor can be summarized by its semantic vector, this method enables comparison of learning dynamics across otherwise incomparable algorithm families. We apply semantic space STNs to classification and regression tasks solved using different machine learning algorithms, recovering known qualitative differences between them. Additionally, we use semantic space STNs to study neural network generalization by contrasting standard training with the label randomization regime of Zhang et al. (2017). The resulting STNs exhibit consistent structural differences, training on real labels produces denser, more efficient and more centralized graphs than training on shuffled labels. Together, our results show that semantic space STNs capture functional training dynamics arising from the interaction between learning algorithms and data, providing a tool for analyzing and comparing learning dynamics across machine learning models and training regimes.
J. Agudelo, A. Tonda, Gabriela Ochoa et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
