Multi-dimensional Classification via Selective Feature Augmentation

Abstract

In multi-dimensional classification (MDC), the semantics of objects are characterized by multiple class spaces from different dimensions. Most MDC approaches try to explicitly model the dependencies among class spaces in output space. In contrast, the recently proposed feature augmentation strategy, which aims at manipulating feature space, has also been shown to be an effective solution for MDC. However, existing feature augmentation approaches only focus on designing holistic augmented features to be appended with the original features, while better generalization performance could be achieved by exploiting multiple kinds of augmented features. In this paper, we propose the selective feature augmentation strategy that focuses on synergizing multiple kinds of augmented features. Specifically, by assuming that only part of the augmented features is pertinent and useful for each dimension’s model induction, we derive a classification model which can fully utilize the original features while conduct feature selection for the augmented features. To validate the effectiveness of the proposed strategy, we generate three kinds of simple augmented features based on standard kNN, weighted kNN, and maximum margin techniques, respectively. Comparative studies show that the proposed strategy achieves superior performance against both state-of-the-art MDC approaches and its degenerated versions with either kind of augmented features.

Modular Graph Transformer Networks for Multi-Label Image Classification

“…If one seeks for more accurate GBNCs, there should be no restriction on the use of ensemble learning methods, except the availability of computational resources. This flexibility of the framework is remarkably different from existing probabilistic MDC approaches [28,35]. Roughly speaking, so long as you train good local models p ′ Y,π : X c −→ Y k (for which you can use all toolsets available in the literature for "standard" single-class-variable classification), the framework in this paper does the rest to combine them optimally into an MDC solution.…”

“…Proof. Using the shorthand notation (35), a regularized variant of the CLL function can be rewritten as…”

“…An example of MDC is predicting subtypes/stages of diseases associated with each patient given his/her medical image and/or demographic information. Few other examples of MDC tasks are classification of biomedical text [60], vehicle classification [34] and beyond [28,35].…”

Multi-dimensional classification: paradigm, algorithms and beyond

“…all dimensions can be indeed brought into the feature space which will facilitate the subsequent MDC model induction. The following works LEFA [58] and SFAM [92] improve KRAM from (43)…”

Distilling Reliable Knowledge for Instance-Dependent Partial Label Learning

“…However, these methods are inefficient for large-scale datasets and high-dimensional features. With the emergence of deep learning, deep PLL methods have been explosively studied (Lv et al 2020;Feng et al 2020;Wen et al 2021;Wu, Wang, and Zhang 2022;Wang et al 2022b;Qiao, Xu, and Geng 2023;Xu et al 2023;Jia and Zhang 2022). For example, self-training techniques are utilized in (Lv et al 2020;Wen et al 2021) to progressively identify the ground-truth label during training.…”