Details
Poster
Presenter(s)
![Seung-Hwan Bae Headshot](https://confcats-catavault.s3.amazonaws.com/CATAVault/ieeecass/master/files/styles/cc_user_photo/s3/user-pictures/11681_0.jpg?h=4537b3c5&itok=GY3WxEKB)
Display Name
Seung-Hwan Bae
- Affiliation
-
AffiliationSeoul National University
- Country
Abstract
In order to reduce the memory bandwidth, this paper proposes a compression method for feature maps in CNN that adaptively exploits the varying correlation between feature map planes. For every feature map plane, the proposed method searches the most similar plane in the same layer, and compresses only the residual of the two planes. Experimental results show that the average bit length to store feature maps is reduced by 14.2% compared to the compression without correlation reduction, and the CNN accuracy does not change and additional training is also not required because the proposed method applies lossless compression.