The paper introduces *Masked Particle Modeling* (MPM), a self-supervised method for learning generic, transferable, and reusable representations on unordered sets of inputs for high energy physics (HEP) scientific data. MPM aims to address the challenges of continuous particle features and unordered sets by adapting masked modeling strategies from natural language processing (NLP) and computer vision (CV). The method involves masking particles in a jet and training the model to recover their identities using a pre-trained vector quantized variational autoencoder. The paper explores the efficacy of MPM in high energy jet data, including the impact of discretization, permutation invariance, and ordering. It also evaluates the fine-tuning capability of the model on various downstream tasks, such as supervised and weakly supervised jet classification, and its ability to transfer efficiently to new classes and data domains with small fine-tuning datasets. The results show that MPM can learn generic representations that are useful for a range of downstream tasks, even with limited labeled data, and can adapt to new data sets and weakly supervised settings. The work suggests that self-supervised learning strategies for HEP data, combined with larger datasets and model sizes, could be a promising direction for future developments in machine learning for high energy physics.The paper introduces *Masked Particle Modeling* (MPM), a self-supervised method for learning generic, transferable, and reusable representations on unordered sets of inputs for high energy physics (HEP) scientific data. MPM aims to address the challenges of continuous particle features and unordered sets by adapting masked modeling strategies from natural language processing (NLP) and computer vision (CV). The method involves masking particles in a jet and training the model to recover their identities using a pre-trained vector quantized variational autoencoder. The paper explores the efficacy of MPM in high energy jet data, including the impact of discretization, permutation invariance, and ordering. It also evaluates the fine-tuning capability of the model on various downstream tasks, such as supervised and weakly supervised jet classification, and its ability to transfer efficiently to new classes and data domains with small fine-tuning datasets. The results show that MPM can learn generic representations that are useful for a range of downstream tasks, even with limited labeled data, and can adapt to new data sets and weakly supervised settings. The work suggests that self-supervised learning strategies for HEP data, combined with larger datasets and model sizes, could be a promising direction for future developments in machine learning for high energy physics.