lle-single-hyperparameter-no-out-of-sample

IN premise — summaries/2026/08/24/wiki-Nonlinear_dimensionality_reduction-chunk-1.md

Created 2026-08-24T17:11:20+00:00

Locally Linear Embedding (LLE) has exactly one hyperparameter (K, the number of nearest neighbors) and cannot embed out-of-sample points without additional regularization (e.g., RKHS).

Summary

LLE is simple to tune because there is only one knob to turn, the neighborhood size K, but it comes with a hard deployment constraint: it has no built-in way to place a brand-new point into the low-dimensional space, so any system that needs to embed new data online must bolt on extra machinery like kernel regularization. This makes LLE a good fit for offline batch analysis but a poor default choice for streaming or interactive settings unless that extension is already part of the pipeline.