KSC 2020
Current Result Document :
| 한글제목(Korean Title) |
Edge-Coordinated On-Device Learning for Personalized Intelligent Digits Recognition |
| 영문제목(English Title) |
Edge-Coordinated On-Device Learning for Personalized Intelligent Digits Recognition |
| 저자(Author) |
Phoo Pyae Sone
Subina Khanal
Eui-Nam Huh
|
| 원문수록처(Citation) |
VOL 47 NO. 02 PP. 0813 ~ 0815 (2020. 12) |
한글내용
(Korean Abstract) |
|
영문내용
(English Abstract) |
Neural networks (NNs) and machine learning models are gaining popularity because of their ability to learn the latent features in an optical image, and then classifying an unseen image based on their latent properties. In doing so, the inherent challenges of character extraction and features extraction in the conventional handwritten digit recognition tools are addressed. However, expansive number of heterogeneous datasets are required for getting the best NN model that performs efficient classification of input handwritten image dataset. Besides, the available local dataset is small, privacysensitive, and in practice, mobile users are mostly reluctant to share their raw data for training an efficient learning model. To support collaborative learning, we leverage a decentralized model training approach, to be specific federated learning (FL), and propose a FL-based intelligent digit classification framework for improving personalization. To that end, the users download a pre-trained global model and participate in the FL process without sharing their raw data. In doing so, the proposed approach not only minimizes the generalization error, but also improves the performance of the personalized model for each user.
|
| 키워드(Keyword) |
|
| 파일첨부 |
PDF 다운로드
|
