Title:
METHOD FOR PROVIDING PRE-TRAINED INTEGRATED FRAMEWORK BASED ON TEXT-IMAGE COMPARISON AND ELECTRONIC DEVICE USING SAME
Document Type and Number:
WIPO Patent Application WO/2023/224344
Kind Code:
A1
Abstract:
An electronic device for providing a pre-trained integrated framework based on text-image comparison, according to an embodiment of the present disclosure, comprises a pre-training module, a loss application module, a score application module, and a processor for controlling operations of the pre-training module, the loss application module, and the score application module, wherein the processor may be configured to perform pre-training on a data set including at least one of text and an image, which correspond to a data set domain input through the pre-training module, apply, through the loss application module, a loss to a plurality of positive samples in the pre-trained data set, and apply, through the score application module, a score for embedding the pre-trained data set in the same space from a plurality of domains on the basis of similarity.
More Like This:
Inventors:
KIM JONGSUK (KR)
LEE JANGHYEON (KR)
SHON HYOUNGUK (KR)
KIM BUMSOO (KR)
LEE JANGHYEON (KR)
SHON HYOUNGUK (KR)
KIM BUMSOO (KR)
Application Number:
PCT/KR2023/006577
Publication Date:
November 23, 2023
Filing Date:
May 16, 2023
Export Citation:
Assignee:
LG MAN DEVELOPMENT INSTITUTE CO LTD (KR)
International Classes:
G06N20/00; G06F16/906
Foreign References:
| US10885111B2 | 2021-01-05 |
Other References:
YANGGUANG LI; FENG LIANG; LICHEN ZHAO; YUFENG CUI; WANLI OUYANG; JING SHAO; FENGWEI YU; JUNJIE YAN: "Supervision Exists Everywhere: A Data Efficient Contrastive Language-Image Pre-training Paradigm", ARXIV.ORG, 14 March 2022 (2022-03-14), XP091171303
YOSHIKAWA YUYA, IWATA TOMOHARU, SAWADA HIROSHI, YAMADA TAKESHI: "Cross-Domain Matching for Bag-of-Words Data via Kernel Embeddings of Latent Distributions", ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 28 (NIPS 2015), 1 January 2015 (2015-01-01), pages 1 - 9, XP093110530
YOU YUNING, CHEN TIANLONG, SHEN YANG, WANG ZHANGYANG: "Graph Contrastive Learning Automated", ARXIV.ORG, 28 June 2021 (2021-06-28), XP093110533, DOI: 10.48550/arxiv.2106.07594
LEWEI YAO; RUNHUI HUANG; LU HOU; GUANSONG LU; MINZHE NIU; HANG XU; XIAODAN LIANG; ZHENGUO LI; XIN JIANG; CHUNJING XU: "FILIP: Fine-grained Interactive Language-Image Pre-Training", ARXIV.ORG, 9 November 2021 (2021-11-09), XP091098985
LEE JANGHYEON, KIM JONGSUK, SHON HYOUNGUK, KIM BUMSOO, KIM SEUNG HWAN, LEE HONGLAK, KIM JUNMO: "UniCLIP: Unified Framework for Contrastive Language-Image Pre-training", 36TH CONFERENCE ON NEURAL INFORMATION PROCESSING SYSTEMS (NEURIPS 2022), ITHACA, 1 January 2022 (2022-01-01), Ithaca, pages 1 - 12, XP093110539
YOSHIKAWA YUYA, IWATA TOMOHARU, SAWADA HIROSHI, YAMADA TAKESHI: "Cross-Domain Matching for Bag-of-Words Data via Kernel Embeddings of Latent Distributions", ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 28 (NIPS 2015), 1 January 2015 (2015-01-01), pages 1 - 9, XP093110530
YOU YUNING, CHEN TIANLONG, SHEN YANG, WANG ZHANGYANG: "Graph Contrastive Learning Automated", ARXIV.ORG, 28 June 2021 (2021-06-28), XP093110533, DOI: 10.48550/arxiv.2106.07594
LEWEI YAO; RUNHUI HUANG; LU HOU; GUANSONG LU; MINZHE NIU; HANG XU; XIAODAN LIANG; ZHENGUO LI; XIN JIANG; CHUNJING XU: "FILIP: Fine-grained Interactive Language-Image Pre-Training", ARXIV.ORG, 9 November 2021 (2021-11-09), XP091098985
LEE JANGHYEON, KIM JONGSUK, SHON HYOUNGUK, KIM BUMSOO, KIM SEUNG HWAN, LEE HONGLAK, KIM JUNMO: "UniCLIP: Unified Framework for Contrastive Language-Image Pre-training", 36TH CONFERENCE ON NEURAL INFORMATION PROCESSING SYSTEMS (NEURIPS 2022), ITHACA, 1 January 2022 (2022-01-01), Ithaca, pages 1 - 12, XP093110539
Attorney, Agent or Firm:
LEE, Jung Hoon (KR)
Download PDF:
Previous Patent: CUTTING DEVICE FOR MANUFACTURING LEAD TABS OF SECONDARY BATTERY
Next Patent: PRIMARY FRYING METHOD USING COLLABORATIVE ROBOT
Next Patent: PRIMARY FRYING METHOD USING COLLABORATIVE ROBOT