DocumentCode :
674
Title :
Hashing Hyperplane Queries to Near Points with Applications to Large-Scale Active Learning
Author :
Vijayanarasimhan, Sudheendra ; Jain, Paril ; Grauman, Kristen
Author_Institution :
Univ. of Texas at Austin, Austin, TX, USA
Volume :
36
Issue :
2
fYear :
2014
fDate :
Feb. 2014
Firstpage :
276
Lastpage :
288
Abstract :
We consider the problem of retrieving the database points nearest to a given hyperplane query without exhaustively scanning the entire database. For this problem, we propose two hashing-based solutions. Our first approach maps the data to 2-bit binary keys that are locality sensitive for the angle between the hyperplane normal and a database point. Our second approach embeds the data into a vector space where the euclidean norm reflects the desired distance between the original points and hyperplane query. Both use hashing to retrieve near points in sublinear time. Our first method´s preprocessing stage is more efficient, while the second has stronger accuracy guarantees. We apply both to pool-based active learning: Taking the current hyperplane classifier as a query, our algorithm identifies those points (approximately) satisfying the well-known minimal distance-to-hyperplane selection criterion. We empirically demonstrate our methods´ tradeoffs and show that they make it practical to perform active selection with millions of unlabeled points.
Keywords :
data analysis; file organisation; learning (artificial intelligence); pattern classification; query processing; binary key; data embedding; data mapping; database nearest point retrieval; euclidean norm; hyperplane classifier; hyperplane query hashing; minimal distance-to-hyperplane selection criterion; pool-based active learning; vector space; Algorithm design and analysis; Approximation algorithms; Approximation methods; Databases; Euclidean distance; Search problems; Vectors; Hashing; active learning; approximate nearest neighbors; large-scale search;
fLanguage :
English
Journal_Title :
Pattern Analysis and Machine Intelligence, IEEE Transactions on
Publisher :
ieee
ISSN :
0162-8828
Type :
jour
DOI :
10.1109/TPAMI.2013.121
Filename :
6544184
Link To Document :
بازگشت