US 12,488,236 B2
Server for refining model in terminal and operation method thereof
Sangha Kim, Suwon-si (KR); Seokchan Ahn, Suwon-si (KR); Yoonjung Choi, Suwon-si (KR); Hakjung Kim, Suwon-si (KR); and Yoonjin Yoon, Suwon-si (KR)
Assigned to Samsung Electronics Co., Ltd., Suwon-si (KR)
Appl. No. 17/432,631
Filed by Samsung Electronics Co., Ltd., Suwon-si (KR)
PCT Filed Aug. 18, 2021, PCT No. PCT/KR2021/010951
§ 371(c)(1), (2) Date Aug. 20, 2021,
PCT Pub. No. WO2022/039494, PCT Pub. Date Feb. 24, 2022.
Claims priority of application No. 10-2020-0104810 (KR), filed on Aug. 20, 2020.
Prior Publication US 2022/0343156 A1, Oct. 27, 2022
Int. Cl. G06N 3/08 (2023.01); G06F 18/21 (2023.01); G06N 3/045 (2023.01); G06N 3/063 (2023.01)
CPC G06N 3/08 (2013.01) [G06F 18/217 (2023.01); G06N 3/045 (2023.01); G06N 3/063 (2013.01)] 15 Claims
OG exemplary drawing
 
1. A method, performed by a server, of generating information for updating a model in a terminal, the method comprising:
in response to input information being received by a first terminal-side model in the terminal, receiving, from the terminal, intermediate data output from a first layer included in a shared portion of the first terminal-side model and identification information of the first layer, wherein the shared portion includes a plurality of layers from an input layer within the first terminal-side model;
obtaining a first server-side model and a second model which correspond to the first terminal-side model, wherein the first server-side model is the same artificial intelligence model as the first terminal-side model, the second model includes the shared portion which includes an input layer of the second model, the second model has more layers than the first server-side model, and the second model and the first server-side model are pre-trained using the same training data;
based on the intermediate data and the identification information of the first layer, obtaining output data of the second model as correct answer data for the input information by inputting the intermediate data received from the terminal as an input to a layer following the first layer included in the shared portion of the second model;
based on the intermediate data and the identification information of the first layer, updating the first server-side model by updating a plurality of weight values and bias values of the first server-side model so that the correct answer data is output from the first server-side model when the intermediate data received from the terminal is input to a layer following the first layer included in the shared portion of the first server-side model; and
transmitting difference information between the updated first server-side model and the first server-side model prior to the update to the terminal so that the first terminal-side model outputs the correct answer data when the input information is input to the first terminal-side model by updating the first terminal-side model based on the difference information.