Back to Search
Start Over
Submodularity in Batch Active Learning and Survey Problems on Gaussian Random Fields
- Publication Year :
- 2012
- Publisher :
- arXiv, 2012.
-
Abstract
- Many real-world datasets can be represented in the form of a graph whose edge weights designate similarities between instances. A discrete Gaussian random field (GRF) model is a finite-dimensional Gaussian process (GP) whose prior covariance is the inverse of a graph Laplacian. Minimizing the trace of the predictive covariance Sigma (V-optimality) on GRFs has proven successful in batch active learning classification problems with budget constraints. However, its worst-case bound has been missing. We show that the V-optimality on GRFs as a function of the batch query set is submodular and hence its greedy selection algorithm guarantees an (1-1/e) approximation ratio. Moreover, GRF models have the absence-of-suppressor (AofS) condition. For active survey problems, we propose a similar survey criterion which minimizes 1'(Sigma)1. In practice, V-optimality criterion performs better than GPs with mutual information gain criteria and allows nonuniform costs for different nodes.
Details
- Database :
- OpenAIRE
- Accession number :
- edsair.doi.dedup.....37197172e87de0c9f046fe5cf4fe896d
- Full Text :
- https://doi.org/10.48550/arxiv.1209.3694