Large margin transductive transfer learning

Abstract

Recently there has been increasing interest in the problem of transfer learning, in which the typical assumption that training and testing data are drawn from identical distributions is relaxed. We specifically address the problem of transductive transfer learning in which we have access to labeled training data and unlabeled testing data potentially drawn from different, yet related distributions, and the goal is to leverage the labeled training data to learn a classifier to correctly predict data from the testing distribution. We have derived efficient algorithms for transductive transfer learning based on a novel viewpoint and the Support Vector Machine (SVM) paradigm, of a large margin hyperplane classifier in a feature space. We show that our method can out-perform some recent state-of-the-art approaches for transfer learning on several data sets, with the added benefits of model and data separation and the potential to leverage existing work on support vector machines.

A survey of transfer learning

“…There are three baseline methods tested where different classifiers are trained with the labeled source data. There are five transfer learning methods tested against, which include methods by Ling [66], Pan [83], Pan [87], Quanz [94], and Xiao [133]. The order of performance from best to worst is ARTL, Xiao [133], Pan [87], Pan [83], Quanz [94] and Ling [66] (tie), and the baseline approaches.…”

Auxiliary Training Information Assisted Visual Recognition

“…With the encouraging successes in applying metric learning techniques to various problems (e.g., machine translation [9], [10], multimedia information retrieval [13], and visual recognition [16], [28]), the metric learning methods have also found its way into the visual recognition tasks, especially in the form of domain adaptation and transfer learning [26], [28].…”

“…Unlike conventional learning assumptions where the training data and the testing data follow the identical probability distribution [26], this problem requires techniques that can incorporate the beneficial information from the auxiliary view into the training of a classification model that works only on the main view. We shall emphasize that this problem is different from the domain adaptation or the transfer learning problems [9], [16], [28].…”

Learning with Minimum Supervision: A General Framework for Transductive Transfer Learning

“…Large-margin transductive transfer learning (LMPROJ) [28] is most related to our model, but with two major difference: (1) it uses the same mapping function for the source and target domains; (2) it utilizes projected mean discrepancy distance instead of squared loss as the reconstruction loss. The basic assumption is that in latent space, the centroid of examples in the source domain should be as close to the one in the target domain as possible.…”

“…This learning problem, i.e., abundant labeled example in source domain and unlabeled examples in the target domain, is also know as transductive transfer learning. There have been several studies to address the problem [2], [28], but it is far from being solved for real applications and is therefore the main focus of this paper.…”