We consider a problem of learning kernels for use in SVM classification inthe multi-task and lifelong scenarios and provide generalization bounds on theerror of a large margin classifier. Our results show that, under mildconditions on the family of kernels used for learning, solving several relatedtasks simultaneously is beneficial over single task learning. In particular, asthe number of observed tasks grows, assuming that in the considered family ofkernels there exists one that yields low approximation error on all tasks, theoverhead associated with learning such a kernel vanishes and the complexityconverges to that of learning when this good kernel is given to the learner.
展开▼