142 In recent years, data-driven approaches have been introduced to improve exploration efficiency. In particular, supervised machine learning techniques have been widely applied to mineral prospectivity mapping, where models are trained to distinguish between known deposits and non-deposit locations (Xiong et al., 2018; Sun et al., 2019). While these methods have demonstrated promising results, they rely heavily on the availability and quality of labeled datasets. This dependence introduces a fundamental limitation in underexplored regions, where the absence of known deposits does not necessarily imply the absence of mineralization. Consequently, areas labeled as “non-deposit” may in fact represent undiscovered targets, leading to the introduction of false negatives and bias in model training (Cracknell & Reading, 2014; He et al., 2024) In this context, unsupervised learning approaches offer an alternative framework that does not rely on explicit labeling. Among these, Variational Autoencoders (VAEs), introduced by Diederik P. Kingma and Max Welling (Kingma & Welling, 2014), provide a probabilistic framework for learning latent representations of complex, high-dimensional data without requiring labeled outputs. A key property of Variational Autoencoders (VAEs) is their ability to learn a continuous latent space in which similar input patterns are mapped to nearby regions (Kingma & Welling, 2014). As a result, inputs sharing similar characteristics tend to be organized smoothly in latent space. This behavior has been widely observed across multiple domains. In computer vision, for example, VAEs have been successfully applied to benchmark datasets such as handwritten digits (LeCun et al., 1998), where they learn structured latent representations that group visually similar digits (Zemouri, 2020), as illustrated in Figure 1. In anomaly detection tasks, this same property allows the model to distinguish normal patterns from outliers by identifying samples that deviate from the learned latent distribution (An & Cho, 2015). Similarly, in medical imaging, VAEs have been extensively used for anomaly detection and representation learning, effectively capturing complex anatomical patterns and variability across imaging modalities (Chen et al., 2018; Baur et al., 2020). In the context of mineral exploration, subcrops exhibiting geological features associated with porphyry systems are expected to occupy nearby regions in latent space, enabling the definition of a representative latent signature and the identification of analogous patterns across the study area. This study proposes an unsupervised mineral prospectivity mapping framework based on a
RkJQdWJsaXNoZXIy MTM0Mzk2