Semi-supervised learning for the study of structural formation in colloidal systems via image recognition.

Area of Science:

• Colloidal science
• Materials science
• Computational physics

Background:

• Machine learning is increasingly used for analyzing colloidal system structures.
• Local bond-order parameters (LBOPs) are effective for crystal structures but limited for random or unknown phases.
• Image-based convolutional neural networks (CNNs) can identify various structures but struggle with novel phases in supervised learning.

Purpose of the Study:

• To develop a robust machine learning framework for analyzing colloidal system structures, capable of identifying both known and unknown phases.
• To overcome the limitations of purely supervised learning methods when encountering novel structural configurations.

Main Methods:

• A hybrid machine learning scheme integrating supervised and unsupervised learning techniques was proposed.
• The approach utilizes both image-based convolutional neural networks (CNNs) and generalized local bond-order parameters (LBOPs).
• The method was applied to analyze two-dimensional colloidal systems.

Main Results:

• The proposed hybrid scheme demonstrated significant efficiency in analyzing colloidal system structures.
• The combined approach successfully identified various structural phases, including potentially unknown ones.
• The method offers a more comprehensive analysis compared to traditional techniques.

Conclusions:

• The hybrid supervised and unsupervised learning approach provides an effective solution for analyzing colloidal system structures.
• This method enhances the ability to detect and classify diverse structural phases, advancing the field of machine learning in materials science.