Learning safe multi-label prediction for weakly labeled data

Abstract

In this paper we study multi-label learning with weakly labeled data, i.e., labels of training examples are incomplete, which commonly occurs in real applications, e.g., image classification, document categorization. This setting includes, e.g., (i) semi-supervised multi-label learning where completely labeled examples are partially known; (ii) weak label learning where relevant labels of examples are partially known; (iii) extended weak label learning where relevant and irrelevant labels of examples are partially known. Previous studies often expect that the learning method with the use of weakly labeled data will improve the performance, as more data are employed. This, however, is not always the cases in reality, i.e., weakly labeled data may sometimes degenerate the learning performance. It is desirable to learn safe multi-label prediction that will not hurt performance when weakly labeled data is involved in the learning procedure. In this work we optimize multi-label evaluation metrics (F 1 score and Top-k precision) given that the ground-truth label assignment is realized by a convex combination of base multi-label learners. To cope with the infinite number of possible ground-truth label assignments, cutting-plane strategy is adopted to iteratively generate the most helpful label assignments. The whole optimization is cast as a series of simple linear programs in an efficient manner. Extensive experiments on three weakly labeled learning tasks, namely, (i) semi-supervised multi-label learning; (ii) weak label learning and (iii) extended weak label learning, clearly show that our proposal improves the safeness of using weakly labeled data compared with many state-of-the-art methods.

Weakly Supervised Learning Meets Ride-Sharing User Experience Enhancement

“…supervised multi-label learning (Wei et al 2018), semisupervised weak label learning (Dong, Li, and Zhou 2018) and semi-supervised partial label learning (Wang, Li, and Zhou 2019). It is worth noting that our paper is different from these works as we focus on the mix of severe label noise and biased label distribution.…”

IWE-Net: Instance Weight Network for Locating Negative Comments and its application to improve Traffic User Experience

“…Step size for w supervised multi-label learning (Wei et al 2018), semisupervised weak label learning (Dong, Li, and Zhou 2018) and semi-supervised partial label learning (Wang, Li, and Zhou 2019). It is worth noting that our paper is different from these works as we focus on the mix of severe label noise and biased label distribution.…”

Semi-Supervised Streaming Learning with Emerging New Labels

“…Semi-Supervised Learning (SSL) (Chapelle, Schölkopf, and Zien 2006;Zhou and Li 2010;Wei et al 2018;Li, Guo, and Zhou 2019) aims to make use of unlabeled data for training -typically a small set of labeled data together with a large collection of unlabeled data. Graph-based SSL algorithms (Zhu, Ghahramani, and Lafferty 2003;Zhou et al 2003;Li, Wang, and Zhou 2016) have a long history of work and propagate limited label information to unlabeled examples following clustering or manifold assumptions.…”