Intrinsic Metrics and Mutual Information
April 18, 2023

Data gains meaning when a system can compare states from within the phenomenon itself. Intrinsic metrics define distance through internal geometry, while mutual information measures how strongly variables share structure.

Brian Greenforest develops those ideas into a seven-part path from differential geometry and information theory to nonlinear tensor methods, emergent features, and machine-learning applications.

Build Geometry From Information

The opening chapters connect intrinsic metrics with differential geometry and nonlinear tensor algebra, then develop mutual information for nonlinear systems, feature selection, and clustering.

Kernel methods extend the comparison into nonlinear spaces. A kernel can measure dependence between feature vectors even when ordinary Euclidean distance hides their relationship.

Let Abstract Bits Produce New Concepts

The next stage treats mutual information as a foundation for emergent features. Machine-learning systems can derive comparison, distance, order, and other concepts from measurements that encode the changing state of a system.

Case studies can compare this approach with traditional methods and reveal applications across science and engineering. Researchers can develop the mathematics, run the studies, and turn the full outline into a book and working research program.

Run Intrinsic Metrics and Mutual Information in a Four-Layer Transformer

The complete training run connects this mathematical argument to executable code, data flow, and a working small model.

Four-Layer Tiny Transformer Training Run

Originally posted on LinkedIn

Brian Greenforest · (2023-04-18 05:09:39 UTC)

Open the original LinkedIn post · LinkedIn activity 7053957730402803712

LinkedIn status when archived: Visible to anyone on or off LinkedIn.

"Intrinsic Metrics and Mutual Information: From Abstract Measurements to Machine Learning" Chapter 1: Introduction Motivation for the book Overview of the key concepts: intrinsic metrics, mutual information, nonlinear tensor algebra, etc. Chapter 2: Intrinsic Metrics and Differential Geometry Review of differential geometry and intrinsic metrics The relationship between intrinsic metrics and nonlinear tensor algebra Applications of intrinsic metrics in machine learning Chapter 3: Mutual Information and Information Theory Overview of information theory and mutual information Calculation of mutual information for nonlinear systems Applications of mutual information in feature selection and clustering Chapter 4: Nonlinear Tensor Algebra and Kernel Methods Overview of nonlinear tensor algebra Application of kernel methods to nonlinear tensor algebra Use of kernel functions to define the measure of linear dependency between feature vectors in a nonlinear space Chapter 5: From Abstract Measurements to Emergent Features Use of mutual information as an abstract foundation for emergent features Development of machine learning experiments that rely on abstract bits containing information about system states Analysis of the resulting emergent concepts of comparison, distance, order, etc. Chapter 6: Applications of Intrinsic Metrics and Mutual Information in Machine Learning Case studies of using intrinsic metrics and mutual information in machine learning tasks Comparison with traditional machine learning techniques Future directions and potential limitations of the approach Chapter 7: Conclusion Summary of key points Implications of the approach for the field of machine learning and beyond Suggestions for further reading and research