- Conference Article
2
- 10.1109/icdmw.2016.0124
Structural Nonparallel Support Vector Machine Based on LSH for Large-Scale Prediction
- Dec 01, 2016
- Dandan Chen
Considering the fact that the underlying structural information in the training data within classes is vital for a good classifier in real-world classification problems, Structural Nonparallel Support Vector Machine (or SNPSVM, for short) has been proposed. By combining the structural information with nonparallel support vector machine (NPSVM), SNPSVM can fully exploit prior knowledge to directly improve the algorithm’s generalization capacity, each model of which considers not only the compactness in both classes by the structural information but also the separability between classes. In recent decades, discussions on how to train effective classifiers on huge amount of data has attracted more and more research works across several communities. However, the counterpart scalability issue how to apply big trained models efficiently on huge non annotated media collections has been reported rarely. In this paper, we address the problem of speeding-up the prediction phase of linear SNPSVM via Locality Sensitive Hashing (LSH). We propose building efficient hash-based classifiers that are applied in a first stage in order to approximate the exact results and filter the hypothesis space. Furthermore, SNPSVM is solved efficiently by an alternating optimization method, called the improved alternating direction method of multipliers (ADMM). The proposed SNPSVM based on LSH and the solving algorithm can guarantee that we can deal with large-scale classification problems with a huge number of instances as well as features. Experiments performed with millions of one-against-one classifiers show that the proposed hash-based classifier can be more than two orders of magnitude faster than the exact classifier with minor losses in quality.
Read more