A stratified strategy for efficient Kernel-based learning

摘要: In Kernel-based Learning the targeted phenomenon is summarized by a set of explanatory examples derived from training set. When model size grows with complexity task, such approaches are so computationally demanding that adoption comprehensive models not always viable. this paper, general framework aimed at minimizing problem proposed: multiple classifiers stratified and dynamically invoked according to increasing levels corresponding incrementally more expressive representation spaces. Computationally expensive inferences thus adopted only when classification lower too uncertain over an individual instance. The application complex functions avoided where possible, significant reduction overall costs. proposed strategy has been integrated within two well-known algorithms: Support Vector Machines Passive-Aggressive Online classifier. A cost (up 90%), negligible performance drop, observed against Natural Language Processing tasks, i.e. Question Classification Sentiment Analysis in Twitter.