2019/03/02 by Giulia Luise, Luise, Giulia, Massimiliano Pontil +5 · 1 citation
Computer Science · #Machine Learning and Data Classification #Gaussian Processes and Bayesian Inference #Machine Learning and Algorithms
paper · pdf · doi:10.48550/arxiv.1903.00667
We study the interplay between surrogate methods for structured prediction\nand techniques from multitask learning designed to leverage relationships\nbetween surrogate outputs. We propose an efficient algorithm based on trace\nnorm regularization which, differently from previous methods, does not require\nexplicit knowledge of the coding/decoding functions of the surrogate framework.\nAs a result, our algorithm can be applied to the broad class of problems in\nwhich the surrogate space is large or even infinite dimensional. We study\nexcess risk bounds for trace norm regularized structured prediction, implying\nthe consistency and learning rates for our estimator. We also identify relevant\nregimes in which our approach can enjoy better generalization performance than\nprevious methods. Numerical experiments on ranking problems indicate that\nenforcing low-rank relations among surrogate outputs may indeed provide a\nsignificant advantage in practice.\n