IP Library › Granted Patent US 12,651,160
Granted Patent B2
US 12,651,160 · App. 17/795,597 · Granted Jun 9, 2026

Method for training neural network by using de-identified image and server providing same

Inventor: Ho Yeol Choi (Seongnam-si, KR)
Assignee: KakaoBank Corp.
G06N3/08G06V10/761G06V10/7715G06V10/82
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,651,160
App. No.
17/795,597
Granted
Jun 9, 2026
Kind
B2
Abstract

The present invention relates to a neural network training method. The neural network training method using a de-identified image according to the present invention comprises the steps of: encoding a first image represented by a n-th dimensional vector into a predetermined p-th dimensional second image; decoding the second image into a q-th dimensional third image; inputting the third image to a neural network and extracting object information included in the third image; and training at least one parameter information used for computation in the neural network by using the extracted object information. According to the present invention, de-identified images are used for neural network training such that neural network training is made possible without using personal information included in images.

Claims (22)

1 . An iterative neural network training method using a de-identified image, performed by a neural network training server, the method comprising, during each iteration of the iterative neural network training method:

encoding a first image represented by a vector of n-th dimensions into a second image of predetermined pth-dimensions;

decoding the second image into a third image of q-th dimensions, wherein the third image corresponds to the de-identified image of the first image;

inputting the third image to the neural network and extracting object information included in the third image; and

training at least one parameter information of the neural network and de-identification parameter information by using an error of the extracted object information compared with a ground truth label associated with the first image, and a backpropagated error obtained by first updating the at least one parameter information of the neural network, and then continuing to backpropagate the error of the extracted object information through the decoder and the encoder,

wherein the error of the extracted object information updates the at least one parameter information in the neural network, and the backpropagated error updates the de-identification parameter information to determine a compression rate of the encoding computation, and

wherein, during service operation, the first image is deleted, and only the second image and the ground truth label associated with the first image are stored in a database, and a plurality of the stored second images accumulated over a plurality of service operations are bulk-decoded in a batch to generate the third image for further training parameter information of the neural network using the third image.

2 . The method of claim 1 , wherein a size of the p-th dimensions is determined based on the compression rate of the encoding computation, and

wherein the compression rate of the encoding computation is associated with a degree of a de-identification of the de-identified image.

3 . The method of claim 2 , wherein the n-th dimensions and the q-th dimensions have the same size.

4 . The method of claim 1 , further comprising storing the second image encoded in the predetermined p-th dimensions,

wherein the decoding comprises decoding the stored second image into the third image when the neural network performs the training.

5 . The method of claim 1 , wherein the decoding comprises decoding the second image to have a data value different from that of the first image when decoding the second image in the q-th dimensions.

6 . One or more computers and a program stored in a non-transitory computer-readable recording medium to allow the one or more computers to perform operations of each method of claim 1 when executed by the one or more computers.

7 . An iterative neural network training server using a de-identified image, the server comprising during each iteration of the iterative neural network training method:

a de-identification unit configured to encode a first image represented by a vector of n-th dimensions into a second image of predetermined p-th dimensions, and then decode the second image into a third image of q-th dimensions, wherein the third image corresponds to the de-identified image of the first image; and

a training unit configured to input the third image to the neural network and extract an object information included in the third image to train at least one parameter information of the neural network and de-identification parameter information by using an error of the extracted object information compared with a ground truth label associated with the first image, and a backpropagated error obtained by first updating the at least one parameter information of the neural network, and then continuing to backpropagate the error of the extracted object information through the decoder and the encoder,

wherein the error of the extracted object information updates the at least one parameter information in the neural network, and the backpropagated error updates the de-identification parameter information to determine a compression rate of the encoding computation, and

wherein, during service operation, the first image is deleted, and only the second image and the ground truth label associated with the first image are stored in a database, and a plurality of the stored second images accumulated over a plurality of service operations are bulk-decoded in a batch to generate the third image for further training of the neural network using the third image.

8 . The server of claim 7 , further comprising a database storing the second image encoded in predetermined p-th dimensions,

wherein the de-identification unit is configured to decode the stored second image into the third image when the neural network performs the training.

9 . The server of claim 7 , wherein the de-identification unit is configured to decode the second image to have a data value different from that of the first image when decoding the second image in the q-th dimensions.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Jul 27, 2022
From: CHOI, HO YEOL
To: KAKAOBANK CORP.
Reel/Frame 060639/0620 →
Priority Claims (1)
KR 10-2020-0010271 · Jan 29, 2020 · national
Continuity (1)
Related Publication 20230076017A1 · Mar 9, 2023
References Cited (28)
US 10284432B1 · Kuo et al. · 2019 [cited by applicant]
US 10984560B1 · Appalaraju · 2021 [cited by examiner]
US 20100074322A1 · Terashima · 2010 [cited by examiner]
US 20180052971A1 · Hanina · 2018 [cited by examiner]
US 20180129900A1 · Kiraly et al. · 2018 [cited by applicant]
US 20180247193A1 · Holtham · 2018 [cited by examiner]
US 20180374105A1 · Azout et al. · 2018 [cited by applicant]
US 20190138748A1 · Long · 2019 [cited by examiner]
US 20190273948A1 · Yin · 2019 [cited by examiner]
US 20200034520A1 · Kim · 2020 [cited by examiner]
US 20200167966A1 · Kim · 2020 [cited by examiner]
US 20200177470A1 · Kuo · 2020 [cited by examiner]
US 20200372180A1 · Venkataraman · 2020 [cited by examiner]
KR 1020110070735A · 2011 [cited by applicant]
KR 101713089B1 · 2017 [cited by applicant]
KR 101793510B1 · 2017 [cited by applicant]
KR 101861520B1 · 2018 [cited by applicant]
WO 2018017467A1 · 2018 [cited by applicant]
Evaluating the Effectiveness of Automated Identity Masking (AIM) Methods with Human Perception and a Deep Convolutional Neural Network (CNN) Hooge, 2019, arXiv (Year: 2019). [cited by examiner]
Al-Allaf, “Improving the Performance of Backpropagation Neural Network Algorithm for Image Compression/Decompression System,” 2010, Journal of Computer Science 6 (11): 1347-1354 (Year: 2010). [cited by examiner]
Akyazi et al., “Learning-Based Image Compression using Convolutional Autoencoder and Wavelet Decomposition,” 2019, Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops (Year… [cited by examiner]
Baluja et al., “Task-specific color spaces and compression for machine-based object recognition”, 2019, Technical Disclosure Commons, https://www.tdcommons.org/dpubs_series/2067 (Year: 2019). [cited by examiner]
Dimililer, “Backpropagation Neural Network Implementation for Medical Image Compression,” 2013, http://dx.doi.org/10.1155/2013/453098 (Year: 2013). [cited by examiner]
Li et al., “DeepObfuscator: Adversarial Training Framework for Privacy-Preserving Image Classification,” 2019, arXiv:1909.04126v1 (Year: 2019). [cited by examiner]
International Search Report dated Apr. 19, 2021, issued in counterpart International Application No. PCT/KR2021/001023 (3 pages). [cited by applicant]
Notice of Preliminary Examination Results dated May 4, 2020, issued in counterpart KR Patent Application No. 10-2020-0010271, w/English translation (8 pages). [cited by applicant]
Written Decision on Registration dated Jun. 16, 2020, issued in counterpart KR Patent Application No. 10-2020-0010271, w/English translation (4 pages). [cited by applicant]
Office Action dated Mar. 8, 2024, issued in counterpart Singapore Patent Application No. 11202251590R (7 pages). [cited by applicant]