Current Result Document :
ÇѱÛÁ¦¸ñ(Korean Title) |
Plant Disease Classification with Model-Contrastive Loss Based Federated Learning |
¿µ¹®Á¦¸ñ(English Title) |
Plant Disease Classification with Model-Contrastive Loss Based Federated Learning |
ÀúÀÚ(Author) |
Ye Lin Tun
Chu Myaet Thwal
Choong Seon Hong
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 49 NO. 01 PP. 0299 ~ 0301 (2022. 06) |
Çѱ۳»¿ë (Korean Abstract) |
|
¿µ¹®³»¿ë (English Abstract) |
Farms and plantations are a primary source of global food supply as well as the main source of income for the local populace around the world. A decline in crop productivity and quality due to plant diseases is a major threat to many agricultural businesses. Deep neural network models can be trained to assist in the diagnosis of plant diseases to prevent such losses. Federated learning (FL) is a distributed model training approach that can leverage the private data of different agricultural organizations for the plant disease classification task. However, non-IID (Independent and Identically Distributed) data in a typical FL environment decreases the resulting global model performance. MOON is a promising approach to tackle the non-IID data challenge by using model-contrastive loss in the local training step of FL. In the same way, we use model-contrastive loss to handle the heterogeneous data distributions of different agricultural organizations, while training our plant disease classification model with FL. Our experiments in the non-IID FL settings show that using model-contrastive loss can train a better plant disease classification model compared to vanilla FL. |
Å°¿öµå(Keyword) |
federated learning
contrastive learning
plant disease classification
non-IID data
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|