Title: Unsupervised Cognition

Abstract:

Cognitive models have long served as a subtle inspiration for unsupervised learning techniques. To date, the most effective unsupervised methods have primarily focused on clustering data points within mathematical spaces. In this study, we introduce a novel, primitive-driven unsupervised learning framework for decision-making, grounded in a new cognitive paradigm. This representation-focused methodology constructively models the input space as a distributed, hierarchical structure, operating in an input-agnostic manner.

We benchmark our approach against leading unsupervised learning classification methods, as well as state-of-the-art techniques for classifying small and incomplete datasets, and current best practices in cancer type classification. Our results demonstrate that the proposed method surpasses existing state-of-the-art solutions. Furthermore, we assess several cognition-mimicking attributes of our model. The analysis reveals that our proposal not only outperforms the compared algorithms—including supervised learning models—but also exhibits distinct behavioral patterns that align more closely with cognitive processes.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC