KSC 2019
Current Result Document :
| 한글제목(Korean Title) |
Evaluation of the Importance of High Level Features in CNN |
| 영문제목(English Title) |
Evaluation of the Importance of High Level Features in CNN |
| 저자(Author) |
Md Imtiaz Hossain
Md Alamgir Hossain
Md Delowar Hossain
Ngo Thien Thu
Ga-won Lee
Eui-Nam Huh
|
| 원문수록처(Citation) |
VOL 46 NO. 02 PP. 0039 ~ 0041 (2019. 12) |
한글내용
(Korean Abstract) |
|
영문내용
(English Abstract) |
With the increasing depth of deep learning neural network thus complexity to increase it’s accuracy of different tasks(classification, recognition etc.) training time for deep neural networks is also increasing. Very high number of parameters make the deep learning heavy to run into mobile and stationary devices. Though having high performance computing machines, reducing the training time and complexity is also one of the core concerns among the researchers. In this paper, We propose a probabilistic technique to measure the importance of the high level features in CNN. We first compute the importance of each feature using probabilistic method and mute them before feeding in dense layer. We use ResNet CNN model to implement our proposed technique. We present comparison between results using all the features and partially using the less number of features. We use a pre-trained model which is trained using Kinetics400 dataset. |
| 키워드(Keyword) |
|
| 파일첨부 |
PDF 다운로드
|
