liveBook · Manning

Locally Euclidean

As explained in Math and Architectures of Deep Learning

A property of a space where each point has a neighborhood that can be mapped to Euclidean space without distortion of local topology.

A space is locally Euclidean when every point has a small neighborhood that can be mapped one-to-one to a region of ℝᵈ without folding, twisting, or tearing. Locally, the space can therefore be treated as flat, even when its global shape is curved or difficult to visualize. section 12.1 section 12.4

Geometric examples make the property concrete:

  • A small arc of a circle can be straightened into a line segment, making the circle a 1-manifold. section 12.1
  • A small patch of a sphere can be flattened into a plane, making the sphere surface a 2-manifold. section 12.1 section 12.4

This local flatness supports calculus by allowing curves and surfaces to be approximated with linear pieces. In neural-network geometry, layers can be viewed as mappings between manifolds; successive transformations can produce a representation whose classes are easier to separate with a hyperplane. section 12.4

How Math and Architectures of Deep Learning covers it

How deep it goes
Defines local one-to-one Euclidean mapping, gives geometric examples and counterexamples, and stops short of formal topological conditions.
Examples
Uses circles, sphere surfaces, torus surfaces, cylindrical manifolds, 8-curves, and hourglass surfaces as geometric examples.
What you can do after
Evaluate whether a space admits local flat approximations and understand why manifold representations aid calculus and classification.
How it connects
Connects manifolds with Euclidean space, calculus, homeomorphism, neural networks, and linear separation.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Math and Architectures of Deep Learning. The section links point to the parts of the book it draws on.