Datar, ArchitArchitDatarLyu, QiangQiangLyuLI-CHIANG LIN2023-07-312023-07-312023-01-012022042566https://scholars.lib.ntu.edu.tw/handle/123456789/634266The combination of population growth, continuing industrialization, and excessive energy consumption has resulted in numerous prominent challenges of our time, including but are not limited to, climate change and water scarcity. Developing more energy-efficient and cost-effective processes such as separations, through discovery of novel materials, therefore plays an important role in addressing these challenges. Recently, nanoporous materials such as metal–organic frameworks (MOFs) and zeolites have drawn considerable attention for their potential in energy- and environmental-related applications. These materials possess several desirable properties such as selective adsorption, large surface area, and selective permeation, thus making them promising candidates as adsorbents, membranes, or catalysts. Another key feature of these materials is their tunability. Using MOFs as an example, their chemistry and topology can be tailored by strategically selecting specific combination of metal nodes and organic linkers. Tens of thousands of MOFs have been reported, and orders of magnitude more have been predicted. While such large material space provides tremendous opportunities, it also leads to an unprecedented challenge in the search for optimal materials. Brute-force-based materials discovery, i.e., examining each possible candidate one-by-one either experimentally or through computational methods such as molecular simulations, could be difficult. Data-driven approaches by employing machine learning techniques can therefore be advantageous for such a purpose. This chapter describes the applications of machine learning techniques in the high-throughput discovery of nanoporous materials for energy- and environmental-related applications with a focus on adsorption separation and storage (e.g., carbon capture and hydrogen storage). Specifically, this chapter will discuss machine learning approaches based two distinct approaches: i) human-selected features; and ii) machine-engineered features. For the former, machine learning models will be trained and developed using the random forest regressor, with a variety of structural, chemical, and/or energetic input features decided and/or developed by researchers for property or performance predictions. For the latter, deep learning approaches will be discussed for their potential in the materials discovery, such as the use of convolutional neural network (CNN) to train machines to “see” structures and “make” predictions. The fundamental of these methods, as well as recent studies employing these methods in the screening of promising nanoporous materials, will be introduced in detail in this chapter.Adsorption | CO2 Capture | Gas Storage | High‐throughput Material Discovery | Machine Learning | Molecular Simulations | Nanoporous Materials | Separation[SDGs]SDG6[SDGs]SDG7[SDGs]SDG13book part10.1002/9781119819783.ch112-s2.0-85164862792https://api.elsevier.com/content/abstract/scopus_id/85164862792