Artificial Intelligence for Scientific Discoveries: Extracting Physical Concepts from Experimental Data Using Deep Learning

This book explores the use of machine learning to automate the discovery of physical concepts and models, focusing on the automation of parameter identification for deep learning architectures. It demonstrates how SciNet can model a simplified version of a physicist's reasoning process and extract conceptual information from experimental data.

Format: Hardback
Length: 170 pages
Publication date: 12 April 2023
Publisher: Springer International Publishing AG

Will artificial intelligence soon render human researchers obsolete? This comprehensive book delves into cutting-edge methodologies for uncovering physical concepts through machine learning, highlighting their strengths and limitations. It explores the automation of experimental setup creation, physical model development, and model testing. The central theme of the book revolves around automating a crucial step in model creation, specifically identifying the minimal number of natural parameters that possess sufficient information for making predictions about the system under consideration. The core concept behind this approach is to employ a deep learning architecture called SciNet, which models a simplified version of a physicist's reasoning process. SciNet identifies relevant physical parameters, such as the mass of a particle, from experimental data and generates predictions based on the discovered parameters. The author showcases how to extract conceptual information from these parameters, exemplifying how Copernicus derived his heliocentric conclusion from such insights. This book serves as a valuable resource for researchers and practitioners seeking to harness the power of machine learning in the pursuit of physical understanding.

Weight: 453g
Dimension: 235 x 155 (mm)
ISBN-13: 9783031270185
Edition number: 1st ed. 2023