Skip to content

Implemented Locally Linear Embedding algorithm and some variants of LLE

Notifications You must be signed in to change notification settings

JAVI897/LLE-and-its-variants

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

27 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

LLE (Locally linear embedding) and its variants

Locally Linear Embedding is a method for non-linear dimensionality reduction. It was proposed by Roweis and Saul in 2000. At that time LLE was a novel techinque. Earlier techniques for non linear dimensionality reduction were based on MDS to preserve euclidean distances or more sophisticated distances such as geodesic distances (ISOMAP). LLE takes a different approach, LLE won't preserve distances, but local geometry of the data.

In this project I'll be publisihng some implementations for new variants of LLE. In addition, I'll reproduce results of the paper on which the algorithm is based. However, implementations will not be computationally optimal, this repository is primarily for research purposes.

Implementations

  • ISOLLE:

    • In LLE, each data point is reconstructed from a linear combination of its n nearest neighbors, which are typically found using the Euclidean distance. In ISOLLE the search for the neighbors is performed with respect to the geodesic distance. This leads to a more accurate preservation of the data structure [2]

    Example; ISOLLE - Swiss Roll

References

[1] S. T. Roweis and L. K. Saul. Think Globally, Fit Locally: Unsupervised Learning of Low Dimensional Manifolds. Journal of Machine Learning Research 4 (2003) 119-155

[2] Varini, Claudio & Degenhard, Andreas & Nattkemper, Tim. (2005). ISOLLE: Locally linear embedding with geodesic distance. 331-342. 10.1007/11564126_34.

About

Implemented Locally Linear Embedding algorithm and some variants of LLE

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published