蔡明剛Tsai, Ming-Kang林仰徹Lin, Yang-Che2025-12-092025-07-302025https://etds.lib.ntnu.edu.tw/thesis/detail/c9dd6f70b996c0bb38ae489a79ec2acb/http://rportal.lib.ntnu.edu.tw/handle/20.500.12235/125442近幾年,機器學習(ML)正迅速改變化學領域的研究方式,藉由使用這種數學工具,我們能夠在大量的資料中,對化學材料的性質進行預測、探索,甚至以此進行逆向設計等各種化學空間的應用。在本篇論文中,我們簡單分析了目前現有的機器學習方法,包括監督式學習(Supervised Learning)、非監督式學習(Unsupervised Learning)、半監督學習(Semi-supervised Learning, SSL)、強化學習(Reinforced Learning),及自監督學習(Self-supervised Learning),並探討上述方法在目前化學材料領域中的挑戰扮演之重要角色。此外,由此發展之深度學習(Deep Learning)與主動學習(Active Learning),進一步擴展了機器學習處理複雜結構的能力,有效降低資料取得的成本,相對於機器學習,亦展現出不凡的效果。接著在實務方面,在後續章節我們會介紹機器學習在預測材料性質(如能隙、吸附能、熱導率等)、材料分類、催化劑設計及高熵材料探索之應用,以及在應用方法前,對原始資料的前處理,與針對不同情境所使用的任務類型,如迴歸、分類、聚類與生成建模。本篇分析中也深入分析了機器學習在化學材料領域中之挑戰,如資料稀缺與品質不一、模型的解釋性及泛化性不足、結果的可合成性等。最後,我們概述了新興趨勢與未來發展方向,包括基礎模型、自動化實驗系統,以及以資料為核心的研究策略。此分析論文期以提供基本概念及有結構的入門指引,協助推動機器學習驅動之化學材料科學研究之參考資源。In recent years, machine learning (ML) has rapidly reshaped research in chemistry by enabling predictions of material properties, exploration of chemical spaces, and inverse design through large-scale data analysis. This review briefly outlines major ML approaches—including supervised, unsupervised, semi-supervised, reinforcement, and self-supervised learning—and their critical roles in tackling challenges in chemical materials science.Advances such as deep learning and active learning further enhance ML’s ability to handle complex structures while reducing data acquisition costs, often outperforming conventional methods.We also highlight practical applications of ML in property prediction (e.g., band gap, adsorption energy, thermal conductivity), material classification, catalyst design, and the discovery of high-entropy materials. Key tasks such as regression, classification, clustering, and generative modeling are discussed alongside essential preprocessing strategies.Finally, we examine current challenges—such as data scarcity, inconsistent quality, limited interpretability, and generalizability—and introduce emerging trends including foundation models, autonomous experimentation, and data-centric approaches. This review serves as both an entrypoint and a reference for researchers leveraging ML in chemical materials science.材料資訊學機器學習資料驅動設計Materials informaticsMachine LearningData-driven design分析機器學習對化學材料的應用Analyzing the machine learning methods for chemical materials application學術論文