giannhskp / Software-Development-for-Algorithmic-Problems_Project-1

Neighbor Search and Clustering for Vectors using Locality-sensitive hashing and Randomized Projection to Hypercube

Project 1

• Project Description
• Explanation of Implementation Details

Development Team

• Ioannis Kapetangeorgis (giannhskp)
• Dimitrios Sitaras (sitaras)

Part 1

Given a input dataset with vectors and a query dataset, for every query vector find:

• The true nearest neighbor of the input dataset.
• The aproximate nearest neighbor of the input dataset.
• The aproximate N nearest neighbors of the input dataset.
• All vectors inside a given range R. (approximate search)

In order to find the aproximate nearest neighbor(s) we use:

• Locality-sensitive hashing (LSH)
• Projection to Hypercube

Part 2

Vector Clustering.

The initialization of the clusters is done using kMeans++.

The assignment to each cluster can be performed by each of the following:

• Lloyds assignment.
• LSH reverse assignment using Range search.
• Hypercube reverse assignment using Range search.