Course contentsShow
← Machine Learning and Deep Learning
Lesson 404 of 3,5389. Dimensionality ReductionFree lesson

Isomap: Geodesic Distance Preservation

How Isomap estimates manifold distances through nearest-neighbor graphs and applies MDS to geodesic distances.

Isomap: Geodesic Distance Preservation

What you'll learn: How Isomap preserves the true "walking distance" along a curved data manifold by building a neighbor graph and then flattening it intelligently.

The Core Idea

Imagine your data lives on a curved surface, like a rolled-up sheet of paper. PCA would measure straight-line distances through the paper, but the true distances along the surface are longer. Isomap solves this by first estimating the geodesic distance—the actual path you'd walk along the manifold's surface—then using that to create a low-dimensional map.

How Isomap Works

Step 1: Build a Nearest-Neighbor Graph

Isomap connects each data point to its k-nearest neighbors (or all neighbors within radius ε), creating a network that traces the manifold's structure.

Step 2: Compute Geodesic Distances

Instead of straight-line distances, Isomap calculates the shortest path between all point pairs through this graph—like finding driving directions on a road map. This approximates the true manifold distance.

Step 3: Apply MDS

With geodesic distances in hand, Isomap uses Multidimensional Scaling (MDS)—a technique that places points in low-dimensional space so their pairwise distances match the geodesic distances as closely as possible.

The Analogy

Think of a subway map: stations are connected by tracks (nearest neighbors), and travel time between distant stations follows the rail network (geodesic distance), not straight-line "as the crow flies" distance. Isomap preserves these network distances when creating a simplified map.

Key Takeaway: Isomap estimates true manifold distances by computing shortest paths through a nearest-neighbor graph, then applies MDS to preserve those geodesic distances in a lower-dimensional embedding.