A Study on Synergy of Contrastive Learning-Based Pre-Training and Generative Model 


Vol. 46,  No. 12, pp. 2319-2326, Dec.  2021
10.7840/kics.2021.46.12.2319


PDF
  Abstract

In recent years, unsupervised learning has been studied in various fields and has attracted a lot of attention. Among various unsupervised learning methods, the pretraining method using contrastive learning, which learns the similarity and dissimilarity of the positive and negative samples, is being studied with a lot of interest. To overcome the difficulties of supervised learning due to insufficient data and accuracy drop, utilizing contrastive pretraining is an important strategy in computer vision areas. In this paper, we investigate the problem of combining the generative model with contrastive-learning-based pretraining, which has emerged as a major breakthrough resulting in an excellent performance in related studies. More specifically, we present a procedure for the combination which can generate low-dimensional latent trajectories characterizing the high dimensional sensor data features more effectively. We also conducted experiments about human motion using smartphone sensors, which shows a meaningful synergy of contrastive pretraining and generative models.

  Statistics
Cumulative Counts from November, 2022
Multiple requests among the same browser session are counted as one view. If you mouse over a chart, the values of data points will be shown.


  Cite this article

[IEEE Style]

C. Choi, T. Kim, J. Park, J. Lee, S. Nam, J. Park, "A Study on Synergy of Contrastive Learning-Based Pre-Training and Generative Model," The Journal of Korean Institute of Communications and Information Sciences, vol. 46, no. 12, pp. 2319-2326, 2021. DOI: 10.7840/kics.2021.46.12.2319.

[ACM Style]

ChangKyun Choi, Taehwan Kim, Jeongho Park, Juwon Lee, Sangwoo Nam, and Jooyoung Park. 2021. A Study on Synergy of Contrastive Learning-Based Pre-Training and Generative Model. The Journal of Korean Institute of Communications and Information Sciences, 46, 12, (2021), 2319-2326. DOI: 10.7840/kics.2021.46.12.2319.

[KICS Style]

ChangKyun Choi, Taehwan Kim, Jeongho Park, Juwon Lee, Sangwoo Nam, Jooyoung Park, "A Study on Synergy of Contrastive Learning-Based Pre-Training and Generative Model," The Journal of Korean Institute of Communications and Information Sciences, vol. 46, no. 12, pp. 2319-2326, 12. 2021. (https://doi.org/10.7840/kics.2021.46.12.2319)