Name
Abstract | ||
|---|---|---|
The use of nearest neighbors graphs has been leading to considerable gains over classic approaches (e.g., tree-indexing-based methods) in approximate nearest neighbor searches. Classical searches on these graphs are conduced in a greedy way by moving at each step to the neighbor (of current vertex) with the lowest distance to the query. In this work, we explore the combination of topological properties of graphs and the distance itself through a Genetic Programming framework to obtain a better indicator for selecting the next vertex in the search process. Our objective is to minimize the scan rate needed to reach the true nearest neighbors. Experimental results, conducted with three different graph-based methods over a large textual collection, show significant gains of the proposed approach over the classic search algorithm on graphs.
|
Year | DOI | Venue |
|---|---|---|
2019 | 10.1145/3323873.3325014 | ICMR '19: International Conference on Multimedia Retrieval
Ottawa ON
Canada
June, 2019 |
Keywords | Field | DocType |
Nearest neighbor search, nearest neighbors graph, graph topological features, genetic programming, distance learning | Graph,Computer science,Theoretical computer science,Genetic programming,Artificial intelligence,Machine learning | Conference |
ISBN | Citations | PageRank |
978-1-4503-6765-3 | 0 | 0.34 |
References | Authors | |
0 | 3 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Javier Alvaro Vargas Muñoz | 1 | 0 | 0.34 |
| Zanoni Dias | 2 | 262 | 44.40 |
| Ricardo da Silva Torres | 3 | 787 | 61.46 |