Advancing nanomaterials design using novel machine learning methods
The rise of machine learning (ML) has revolutionized the usage of data. Researchers continue to develop new ways to use ML and find new targets to apply ML on. One of these areas of application is found in nanoscience. Nanoscience is a constantly expanding field with applications in almost every part of life, such as medicine, materials design, and consumer products. The experimental research of nanoscience is expensive, augmented by computational research. Computational research is, however, also resource-intensive and time-consuming due to the complexity of the simulation models. Machine learning promises to alleviate that strain. This work and the articles presented focus on a family of distance-based machine learning algorithms, Minimal Learning Machine (MLM), and Extreme Minimal Learning Machine (EMLM), in the context of computational nanoscience. Specifically in the context of monolayer protected nanoclusters (MPC).
The distance-based ML methods are studied as surrogates in feature selection and knowledge discovery. A set of benchmark, generated, and molecular dynamics-based datasets were used in the included articles. The performance of MLM was studied by using it as a surrogate, comparing it to other methods, and inspecting the effect of a solver on its function. EMLM was used as the ML model in feature selection and knowledge discovery. A set of scaling-focused benchmark datasets were developed based on the simulation data of Au38(SCH3)24 MPC and a set of synthetic benchmark & development datasets were created to test the performance of a feature selection algorithm. A Mean Absolute Sensitivity (MAS) utilizing distance-based feature selection algorithm, Distance-based one-shot wrapper, was developed and then extended to Feature Importance Detector. An umbrella review was made to contextualize the one-shot wrapper to feature selection literature. The results prove the viability of distance-based ML methods in the context of computational nanoscience.
Keywords: Machine Learning, Distance–Based Regression, Nanoscience, MLM, EMLM, Hybrid Nanoparticles, Feature Selection, Knowledge discovery
Publisher
Jyväskylän yliopistoISBN
978-951-39-9517-1ISSN Search the Publication Forum
2489-9003Contains publications
- Artikkeli I: Pihlajamäki, A., Hämäläinen, J., Linja, J., Nieminen, P., Malola, S., Kärkkäinen, T., & Häkkinen, H. (2020). Monte Carlo Simulations of Au38(SCH3)24 Nanocluster Using Distance-Based Machine Learning Methods. Journal of Physical Chemistry A, 124(23), 4827-4836. DOI: 10.1021/acs.jpca.0c01512. JYX: jyx.jyu.fi/handle/123456789/69062
- Artikkeli II: Linja, J., Hämäläinen, J., Nieminen, P., & Kärkkäinen, T. (2020). Do Randomized Algorithms Improve the Efficiency of Minimal Learning Machine?. Machine Learning and Knowledge Extraction, 2(4), 533-557. DOI: 10.3390/make2040029
- Artikkeli III: Linja, J., Hämäläinen, J., Nieminen, P., & Kärkkäinen, T. (2023). Feature selection for distance-based regression : An umbrella review and a one-shot wrapper. Neurocomputing, 518, 344-359. DOI: 10.1016/j.neucom.2022.11.023
- Artikkeli IV: Linja, J., Hämäläinen, J., Pihlajamäki, A., Nieminen, P., Malola, S., Häkkinen, H., Kärkkäinen, T. Knowledge Discovery from Atomic Structures using Feature Importances. Manuscript.
Metadata
Show full item recordCollections
- JYU Dissertations [836]
- Väitöskirjat [3535]
License
Related items
Showing items with similar title or keywords.
-
Problem Transformation Methods with Distance-Based Learning for Multi-Target Regression
Hämäläinen, Joonas; Kärkkäinen, Tommi (ESANN, 2020)Multi-target regression is a special subset of supervised machine learning problems. Problem transformation methods are used in the field to improve the performance of basic methods. The purpose of this article is to test ... -
Autonomous maritime ecosystem : digital concepts and business case : results from the JYU TJTSM54 course on advanced topics on systems development
Impiö, Johannes; Risku, Juhani; Kollanus, Sami; Vakkuri, Ville; Kemell, Kai-Kristian; Kultanen, Joni; Himmanen, Joonas; Abrahamsson, Pekka (2019) -
Integrating Advanced Visual Information with Ball Projection Technology Constrains Dynamic Interceptive Actions
Stone, J. A.; Panchuk, D.; Davids, Keith; North, J. S.; Maynard, I. (Elsevier BV, 2014)The role of advanced visual information in ball catching was investigated by integrating video images of action and ball projection technology in four different conditions: Integrated video and ball projection (VBP), ... -
Advancing service design research with design science research
Teixeira, Jorge Grenha; Patrício, Lia; Tuunanen, Tuure (Emerald, 2019)Purpose – Service design is a multidisciplinary approach that is key to service innovation, as it brings new service ideas to life. In this context, the development of new service design methods and models for creating new ... -
Advancing Design Science Research with Solution-based Probing
Briggs, Robert O.; Böhmann, Tilo; Schwabe, Gerhard; Tuunanen, Tuure (University of Hawai'i at Manoa, 2019)We propose solution-based probing as an extension of action design research. The core idea is that researchers bring a prototype solution (probe) into one or more fields and explore to synthesize robust and generalizable ...