Multi-label optimal margin distribution machine

Abstract

Multi-label support vector machine (Rank-SVM) is a classic and effective algorithm for multi-label classification. The pivotal idea is to maximize the minimum margin of label pairs, which is extended from SVM. However, recent studies disclosed that maximizing the minimum margin does not necessarily lead to better generalization performance, and instead, it is more crucial to optimize the margin distribution. Inspired by this idea, in this paper, we first introduce margin distribution to multi-label learning and propose multi-label Optimal margin Distribution Machine (mlODM), which optimizes the margin mean and variance of all label pairs efficiently. Extensive experiments in multiple multi-label evaluation metrics illustrate that mlODM outperforms SVM-style multi-label methods. Moreover, empirical study presents the best margin distribution and verifies the fast convergence of our method.

Self-supervised clustering of mass spectrometry imaging data using contrastive learning

“…We used EfficientNet-B0 trained on ImageNet as a baseline CNN. EfficientNet 35 has been demonstrated to achieve high accuracy on ImageNet and provide an order of magnitude higher efficiency than previous models, such as ResNet and Xception. In SimCLR framework (see ESI, † Fig.…”

Rethinking and Reweighting the Univariate Losses for Multi-Label Ranking: Consistency and Generalization

“…Technically, focusing on the widely used kernel-based algorithms [14,15,13], we present the generalization analyses based on Rademacher complexity [16] and the vector-contraction inequality [17], following recent work [13]. For the Fisher consistency, we consider more general reweighted univariate losses, which naturally extends the results in prior work [9,8].…”

“…In the following, we consider the kernel-based learning algorithms which have been widely used in practice [3,28,29,14,15] and in theory [13] in MLC. Besides, our following analyses can be extended to other forms of hypothesis class, such as neural networks [30].…”

Multi-label Classification Based on Adaptive Resonance Theory

“…Various types of algorithm adaptation approach have been introduced based on k-NN [7,33], decision tree [34], regression [35], Support Vector Machine (SVM) [36], and feed-forward neural networks [37,38]. In order to achieve high classification performance, recent studies consider label distributions and their correlations in a label learning process [39,40]. These algorithms, however, cannot cope with the situation where new label information is sequentially provided.…”