Maurizio Petrelli 

The present manuscript reports on the state-of-the-art and future perspectives of Machine Learning (ML) in Petrology. To do that, it first introduces the basics of ML, including definitions, core concepts, and applications. Then, it starts reviewing the state-of-the-art of ML in petrology. Established applications mainly concern clustering, dimensionality reduction, classification, and regression. Among them, clustering and dimensionality reduction are particularly valuable for decoding the chemical record stored in igneous and metamorphic phases and to enhance data visualization, respectively. Classification and regression tasks find applications, for example, in petrotectonic discrimination and geothermobarometry, respectively. The main core of the manuscript consists of depicting the next future for ML in petrological applications. I propose a future scenario where ML methods will progressively integrate and support established petrological methods in boosting new findings, possibly providing a paradigm shift. In this framework, the use of multimodal data, data fusion, physics-informed neural networks, and ML-supported numerical simulations, will play a significant role. Also, the use of ML hypotheses formulation and symbolic regression could significantly boost new findings. In the proposed scenario, the main challenges are: a) progressively link machine learning algorithms with the physical and thermodynamic nature of the investigated petrologic processes, b) unblur deep learning algorithms that too often operate as black boxes, c) go ahead in exploring cutting edge tools that rise from researches in Artificial Intelligence, and overall, d) start a collaborative effort among researchers coming from different disciplines in research and teaching.