Abstract | ||
---|---|---|
Imbalanced learning is a traditional problem in machine learning and widely occurs in many applications. Most of the methods apply simple geometric transformation for data augment to imbalanced datasets. Due to those methods learn from local information, they might generate noisy samples in the dataset with high dimension and special complexity. To solve the problem, we propose an improved Generative Adversarial Networks with modification function (GAN-MF) to approximate the true distribution of the minority class of the dataset. The model could generate data from an overall perspective to overcome the limitation of the simple geometric transformation. The performance of GAN-MF is compared against multiple standard oversampling algorithms on several imbalanced learning tasks. Experiments demonstrate that the model has an improvement in data augment for imbalanced learning. |
Year | DOI | Venue |
---|---|---|
2019 | 10.1007/978-3-030-36808-1_3 | NEURAL INFORMATION PROCESSING (ICONIP 2019), PT IV |
Keywords | DocType | Volume |
Imbalanced learning, Generative Adversarial Networks (GAN), Data augment, Modification function | Conference | 1142 |
ISSN | Citations | PageRank |
1865-0929 | 0 | 0.34 |
References | Authors | |
0 | 5 |
Name | Order | Citations | PageRank |
---|---|---|---|
Zhuocheng Zhou | 1 | 0 | 0.68 |
Bofeng Zhang | 2 | 10 | 3.86 |
Ying Lv | 3 | 0 | 1.69 |
Tian Shi | 4 | 0 | 0.34 |
Furong Chang | 5 | 0 | 0.68 |