Machine learning predictions from unpredictable chaos.

Area of Science:

• Complex Systems Science
• Computational Topology
• Machine Learning

Background:

• Chaos theory describes systems highly sensitive to initial conditions, exhibiting unpredictable behavior.
• Traditional understanding views chaotic systems as inherently unpredictable, limiting their practical applications.
• Understanding chaos offers significant social and economic benefits, driving research into its predictability.

Purpose of the Study:

• To introduce chaotic learning, a novel multiscale topological paradigm for accurate predictions from chaotic systems.
• To demonstrate that chaotic dynamics can yield unprecedented quantitative predictions.
• To bridge the fields of topology, chaos, and learning.

Main Methods:

• Development of multiscale topological Laplacians.
• Embedding real-world data into interactive chaotic dynamical systems.
• Modulation of dynamical behaviors for accurate data prediction.

Main Results:

• Successful prediction of physical properties from chaotic systems across diverse datasets.
• Demonstration of accurate predictions for brain waves, protein data, single-cell RNA sequencing, and image datasets.
• Validation using Lorenz and Rossler chaotic attractors.

Conclusions:

• Chaotic learning provides a paradigm shift in understanding and predicting chaotic systems.
• The method enables accurate quantitative predictions from seemingly random dynamics.
• This work integrates topology, chaos, and machine learning for novel insights.