US 12,332,869 B2
Method and apparatus for labeling data in machine to machine system
Jae Seung Song, Seoul (KR)
Assigned to Hyundai Motor Company, Seoul (KR); Kia Corporation, Seoul (KR); and Industry Academy Cooperation Foundation of Sejong University, Seoul (KR)
Filed by Hyundai Motor Company, Seoul (KR); Kia Corporation, Seoul (KR); and Industry Academy Cooperation Foundation of Sejong University, Seoul (KR)
Filed on Sep. 12, 2022, as Appl. No. 17/942,737.
Claims priority of provisional application 63/252,311, filed on Oct. 5, 2021.
Prior Publication US 2023/0120195 A1, Apr. 20, 2023
Int. Cl. G06F 16/22 (2019.01)
CPC G06F 16/2272 (2019.01) 19 Claims
OG exemplary drawing
 
1. A method for operating a first device in a machine-to-machine (M2M) system, the method comprising:
receiving, from a second device having a labeling tool, a first message for discovering resources for data labeling;
transmitting, to the second device, a second message including a resource for unlabeled raw data stored in the first device, the resource for the unlabeled raw data including the unlabeled raw data;
receiving, from the second device, a third message requesting to perform data labeling, the third message including information related to a label for the unlabeled raw data;
performing the data labeling of the unlabeled raw data for training an artificial intelligence (AI) model and updating the resource for the unlabeled raw data based on the information related to the label using a reference ontology;
storing the updated resource as training data based on a structure of the updated resource;
in response to a fourth message from a third device for retrieving labeled data, transmitting the updated resource to the third device, wherein the updated resource is used as learning data for training the artificial intelligence (AI) model in the third device; and
training the AI model to perform a predictive operation using the labeled data, determine a loss value based on the labeled data, and update the AI model based on the loss value,
wherein the structure of the updated resource includes a first attribute indicating that the data is for training, a second attribute indicating a type of labeling for the unlabeled raw data, a third attribute indicating a format of labeling for the unlabeled raw data, a fourth attribute containing annotation contents for the unlabeled raw data based on the format of labeling, and a fifth attribute indicating the reference ontology used to label the unlabeled raw data, and
wherein the information related to the label for the unlabeled raw data comprises annotations obtained through the labeling tool of the second device.