Digital Library[ Search Result ]
Autoencoder-based Learning Contribution Measurement Method for Training Data Selection
Yuna Jeong, Myunggwon Hwang, Wonkyung Sung
http://doi.org/10.5626/JOK.2021.48.2.195
Despite recent significant performance improvements, the iterative process of machine-learning algorithms makes development and utilization difficult and time-consuming. In this paper, we present a data-selection method that reduces the time required by providing an approximate solution . First, data are mapped to a feature vector in latent space based on an Autoencoder, with high weight given to data with high learning contribution that are relatively difficult to learn. Finally, data are ranked and selected based on weight and used for training. Experimental results showed that the proposed method selected data that achieve higher performance than random sampling.
A Method for Training Data Selection based on LSTRf
Myunggwon Hwang, Yuna Jeong, Wonkyung Sung
http://doi.org/10.5626/JOK.2020.47.12.1192
This paper presents a data selection method that has a positive effect on learning for an efficient human-in-the-loop (HITL) process required for automated and intelligent artificial intelligence (AI) development. Our method first maps the training data onto a 2D distribution based on similarity, and then grids are laid out with a fixed ratio. By applying Least Slack Time Rate first (LSTRf) techniques, the data are selected based on the distribution consistency of the same class data within each grid. The finally selected data are used as convolutional neural network (CNN)-based classifiers to evaluate the performance. We carried out experiments on the CIFAR-10 dataset, and evaluated the effect of grid size and the number of data selected in one operation. The selected training data were compared to randomly selected data of the same size. The results verified that the smaller the grid size (0.008 and 0.005) and the greater the number selected in the single operation, the better the learning performance.
Search
Journal of KIISE
- ISSN : 2383-630X(Print)
- ISSN : 2383-6296(Electronic)
- KCI Accredited Journal
Editorial Office
- Tel. +82-2-588-9240
- Fax. +82-2-521-1352
- E-mail. chwoo@kiise.or.kr