Linear Algebra and Duality of Neural Networks
share
Natural for Neural networks bases, mappings, projections and metrics are built. Graph-theoretical interpretation is offered. Non-Gaussianity naturally emerges, even in relatively simple datasets. Training statistics and hierarchies are discussed, from physics point of view. Relationship between exact and numerical solutions is looked into. Duality between observables and observations is established. Examples support all new concepts.
READ FULL TEXT