A deeper look into natural sciences with physics-based and data-driven measures

Area of Science:

• Natural Sciences
• Data Science
• Physics

Background:

• Machine learning advancements enable deeper analysis of experimental data.
• Traditional models often rely on restrictive assumptions.
• Analyzing latent effects in data is crucial for scientific discovery.

Purpose of the Study:

• To discuss state-of-the-art data-driven tools for analyzing latent effects.
• To explain the applicability of these tools in natural sciences.
• To highlight novel, computationally cheap, physics-motivated methods: latent entropy and latent dimension.

Main Methods:

• Focus on two recently introduced tools: latent entropy and latent dimension.
• These methods relax restrictive assumptions common in machine learning.
• The tools incorporate aspects that best fit dynamical systems.

Main Results:

• Demonstrated applicability on several natural science examples.
• Revealed unobserved features, including a magnetic measurement gradient.
• Identified a latent network of glymphatic channels in mouse brain microscopy data.

Conclusions:

• Latent entropy and latent dimension are powerful, flexible tools for data analysis.
• These methods offer new perspectives in natural sciences by uncovering hidden data features.
• The physics-motivated approach provides computationally efficient and insightful analysis.