BROWSE
Detail
Conference Paper
Contrastive Learning for Reducing False Negatives with Global and local views in Augmented Data
Cited 0 time in 
Share 
Share
- Authors
- Joonsun Auh, Changsik Cho, Seon-Tae Kim
- Issue Date
- 2023-10
- Citation
- Innovations in Intelligent Systems and Applications Conference (ASYU), pp.1-5
- Language
- English
- Type
- Conference Paper
- Abstract
- Self-supervised learning is a technique that learns representations from large scaled unlabeled data. Among various learning algorithms in self-supervised learning, contrastive learning is a method that compares the similarity of dataset to learn their features and relationships. It has been widely studied and explored. However, contrastive learning currently faces the issue of false negatives. This false negative problem refers to the misjudgment of data, hindering the model's learning process. Therefore, in this paper, research was conducted to select false negatives and learn in the right way using the global and local views and cosine similarity of data, and it was confirmed that the performance was improved compared to the existing contrastive learning model.
- KSP Keywords
- Cosine similarity, False negative, Global and local, Learning model, Unlabeled data, learning algorithms, learning process, self-supervised learning