Blessing of dimensionality: mathematical foundations of the statistical physics of data

TL;DR

This paper explores how measure concentration phenomena in high-dimensional spaces can be leveraged to transform the curse of dimensionality into a blessing, enabling simplified machine learning problems and robust correction of AI systems.

Contribution

It introduces new stochastic separation theorems showing that high-dimensional random points are linearly separable, facilitating non-iterative correction methods for AI systems.

Findings

High-dimensional measure concentration simplifies machine learning tasks.

Random points in high dimensions are linearly separable even in large sets.

Theorems enable one-shot learning and correction of AI errors.

Abstract

The concentration of measure phenomena were discovered as the mathematical background of statistical mechanics at the end of the XIX - beginning of the XX century and were then explored in mathematics of the XX-XXI centuries. At the beginning of the XXI century, it became clear that the proper utilisation of these phenomena in machine learning might transform the curse of dimensionality into the blessing of dimensionality. This paper summarises recently discovered phenomena of measure concentration which drastically simplify some machine learning problems in high dimension, and allow us to correct legacy artificial intelligence systems. The classical concentration of measure theorems state that i.i.d. random points are concentrated in a thin layer near a surface (a sphere or equators of a sphere, an average or median level set of energy or another Lipschitz function, etc.). The new stochastic separation theorems describe the thin structure of these thin layers: the random points are not only concentrated in a thin layer but are all linearly separable from the rest of the set, even for exponentially large random sets. The linear functionals for separation of points can be selected in the form of the linear Fisher's discriminant. All artificial intelligence systems make errors. Non-destructive correction requires separation of the situations (samples) with errors from the samples corresponding to correct behaviour by a simple and robust classifier. The stochastic separation theorems provide us by such classifiers and a non-iterative (one-shot) procedure for learning.