Learning with Augmented Class by Exploiting Unlabeled Data

By Qing Da, Yang Yu, Zhi-Hua Zhou. 2014

111 citations recorded by the source corpus.

In many real-world applications of learning, the environment is open and changes gradually, which requires the learning system to have the ability of detecting and adapting to the changes. Class-incremental learning (C-IL) is an important and practical problem where data from unseen augmented classes are fed, but has not been studied well in the past. In C-IL, the system should beware of predicting instances from augmented classes as a seen class, and thus faces the challenge that no such instances were observed during training stage. In this paper, we tackle the challenge by using unlabeled data, which can be cheaply collected in many real-world applications. We propose the LACU framework as well as the LACU-SVM approach to learn the concept of seen classes while incorporating the structure presented in the unlabeled data, so that the misclassification risks among the seen classes as well as between the augmented and the seen classes are minimized simultaneously. Experiments on diverse datasets show the effectiveness of the proposed approach.

Subfields: Artificial Intelligence

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

Towards Open Set Deep Networks (2015, background): “Altough a range of algorithms has been developed to address this issue [4, 20, 21, 25, 2], performing open set recognition with deep networks has remained an unsolved problem.”
A survey of data fusion in smart city applications (2019, methodology): “Research works such as Zhou et al. [261], [262] have attempted to ﬁx unlabeled data by transforming them into useful features to achieve certain objectives.”
Transductive Multi-View Zero-Shot Learning (2015, background): “Additionally, zero-shot learning can be taken as the generalisation of class-incremental learning (C-IL) [4, 35] or life-long learning [26].”
Recent Advances in Open Set Recognition: A Survey (2018, background): “In fact, as the KUCs’ data is often available at hand [25], [114], [115], we can fully leverage them to reduce such a risk and further improve the robustness of the OSR models for UUCs.”
Open Category Detection with PAC Guarantees (2018, background): “Additional work has addressed open category detection by tuning the decision boundary based on unlabeled data which contains data from novel categories (Da et al., 2014).”
Learnware: on the future of machine learning (2016, background): “, data distribution may change [3], new classes may occur [4,5], features may change [6], etc.”
Deep Class-Incremental Learning: A Survey (2023, methodology): “, the model is only updated with a single new stage [16], [17], [18].”
Class-Incremental Learning: A Survey (2023, background): “However, most of them address the incremental learning within two tasks, i.e. , the model is only updated with a single new stage [12], [13].”
Adversarial Reciprocal Points Learning for Open Set Recognition (2021, methodology): “Inspired by a classifier with rejection option [9], [10], [11], Scheirer et al.”
Multi-Instance Learning With Emerging Novel Class (2021, methodology): “Moreover, MIEN-metric is signiﬁcantly better than the one based on one-class MIL, and also outperforms OVR-MIL (which is a strong baseline method in augmented class recognition, cf. [ 26]) by a large margin, i.e., 5 (cid:11) 6 percent F1 score.”
Learning to Classify With Incremental New Class (2021, unknown): “3) LACU-SVM [35]: By adopting the one-versus-rest approach, it picks a classiﬁcation boundary among all low-density separators that minimizes the empirical risk and structure risk as well as the augment risk simultaneously.”