A Field Theory for Multi-dimensional Scaling

An approach to multi-dimensional scaling is described which employs an analogy from the physics of conservative vector fields. This analogy allows the introduction of kinematic concepts into the data science problem in a natural way. Specific examples are presented. The method described here uses multi-dimensional scaling to introduce information redundantly into feature sets for classifier problems. This is empirically shown to have beneficial effects for certain difficult classification problems. This extends work done previously [1, 2] by using the posited physical analogy to make training more intuitive, efficient, and effective. A concept of super features is introduced and shown to improve classifier performance.