Name
Abstract | ||
|---|---|---|
Incremental methods for structure learning of pairwise Markov random fields (MRFs), such as grafting, improve scalability to large systems by avoiding inference over the entire feature space in each optimization step. Instead, inference is performed over an incrementally grown active set of features. In this paper, we address the computational bottlenecks that current techniques still suffer by introducing online edge grafting, an incremental, structured method that activates edges as groups of features in a streaming setting. The framework is based on reservoir sampling of edges that satisfy a necessary activation condition, approximating the search for the optimal edge to activate. Online edge grafting performs an informed edge search set reorganization using search history and structure heuristics. Experiments show a significant computational speedup for structure learning and a controllable trade-off between the speed and the quality of learning. |
Year | Venue | Field |
|---|---|---|
2017 | arXiv: Learning | Pairwise comparison,Feature vector,Inference,Markov chain,Reservoir sampling,Heuristics,Artificial intelligence,Mathematics,Machine learning,Scalability,Speedup |
DocType | Volume | Citations |
Journal | abs/1705.09026 | 0 |
PageRank | References | Authors |
0.34 | 8 | 2 |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Walid Chaabene | 1 | 0 | 0.34 |
| Bert Huang | 2 | 563 | 39.09 |