Using machine learning to provide understanding along with results
Using machine learning to provide understanding along with results lead image
Over the past few years, machine learning algorithms and artificial intelligence (AI) have been increasingly employed to thoroughly analyze scientific data and generate more accurate and efficient computational models. However, while AI-based approaches can be powerful, they often fail to provide deeper insight into the physical mechanisms at work.
Zhu et al. employed AI in a multistep approach to first uncover physical information about a system and then use that information to better design AI-based analysis. They demonstrated their method with a case study optimizing amorphous networks with extreme Poisson’s ratios.
“A pivotal strategy in our approach is the ‘cyclical route’: using machine learning to discover the physical mechanism, which then feeds back to enhance the machine learning efficacy,” said author Lei Xu. “This iterative approach not only advances scientific discovery but also enriches machine learning techniques, creating a mutually beneficial cycle of discovery and innovation.”
The team applied this method to generate amorphous networks with extreme Poisson’s ratios. Their method exhibited unique characteristics not found in established configurations, such as an absence of concave structures. In the process, it identified key characteristics of amorphous networks, information that can serve as general design principles in subsequent studies.
“Our future research aims to explore the intricate relationship between complex network structures and deep neural networks,” said Xu. “These networks exhibit analogous patterns, with complex configurations mirroring the intricate networks akin to those in the human brain. Applications encompass phenomena from allosteric responses to fluid and heat transport, showcasing the potential for innovative advancements in understanding and optimizing complex systems with AI.”
Source: “A cyclical route linking fundamental mechanism and AI algorithm: An example from tuning poisson’s ratio in amorphous networks,” by Changliang Zhu, Chenchao Fang, Zhipeng Jin, Baowen Li, Xiangying Shen, and Lei Xu, Applied Physics Reviews (2024). The article can be accessed at https://doi.org/10.1063/5.0199530 .
