SNU College of Engineering Professor Gunehee Kim’s Team has their AI System Recognized by the World 

- Won the AI challenge in 3 competitions
- Exemplary results with RippleAI, a startup within the lab



▲ Examples of the Fashion IQ Challenge and the LID Challenge

The team led by Professor Gunhee Kim of the SNU Department of Computer Science and Engineering and the startup RippleAI, which was founded in the laboratory, recently won three world-class artificial intelligence challenges and was recognized for its outstanding research achievements.
 
RippleAI is a company that develops an artificial intelligence system that understands and elicits customer data in all data formats (natural language, images, video, emojis, etc.) used online.
 
High accuracy of the natural language feedback algorithm
In the Fashion IQ Challenge of CVPR2020 (Conference on Computer Vision and Pattern Registration 2020), the world's largest academic society in the field of computer vision that was the second competition organized by IBM Research, the RippleAI-SNU team (Professor and Chief Executive Officer of RippleAI - Gunhee Kim,  Graduate students Youngjae Yu and Jongseok Kim, and RippleAI researcher Seunghwan Lee) became first place out of 44 teams. They took the place of runner-up at the competition that was hosted by ICCV2019 for the first time.

 

▲ Startup RippleAI

 
Fashion IQ is a competition that recommends the most appropriate or suitable clothing according to the user's natural language search (the language people normally use), and when presented an image of a product, the extent to which how accurately the user's modification requests that is inputted through natural language are reflected to recommend new product images is measured. For example, if an image of a white dress is provided, when the user enters "add a belt to the waist in red", an image that reflects this request must be newly and accurately presented. 
 
This problem of devising algorithms that well reflect users' repetitive natural language feedback is a key technology for developing interactive search/recommendation systems. The presentation and awards ceremony of the competition was held through the CVPR Workshop on June 19.
 
Reference link: https://sites.google.com/view/cvcreative2020/fashion-iq
 
Showing the importance of research on Weakly Supervised Learning
Professor Gunhee Kim's team (Postdoctoral researcher Junhyug Noh, researchers Wonho Bae and Jinhwan Seo) participated in another competition LID (Learning from Impact Data), with the support from Baidu, participated in the Learning from Imperfect Data Challenge of the LID competition at the CVPR2020 and won the Weakly-supervised Object Localization part while took the position of runner-up of the Weakly-supervised Semantic Segmentation part, being entitled to a prize money of worth 2,000 USD. The process of Weakly Supervised Learning is aimed to achieve the level of functionality that is possible when complete information is given even when limited amount of information is provided.
 
Regarding the significance of participating in the competition, Professor Gunhee Kim said, "It is highly expensive and oftentimes impossible to label the information needed for tens or millions of large-capacity images for a particular problem. As a result, the importance of research on "weakly supervised learning" aimed at solving high-level problems with simple, low-cost information is becoming increasingly important, and is receiving great attention in industries such as video search, self-driving, and defect testing, which use videoa to deal with real-life problems."
 
Reference link: https://lidchallenge.github.io/
 
Demonstration of ability to naturally engage in conversations
The RippleAI-SNU (Professor Gunhee Kim, RippleAI researcher Hankyol Lee, and graduate student Youngjae Yu) also won the first prize among 39 teams at the Sarcasm-recognition Artificial Intelligence Competition of the world's most prestigious natural language processing institute ACL2020 (Annual Meeting of the Association for Computational Linguistics).
 
The competition is meaningful in exchanging and presenting skills to understand and distinguish satire or irony in online communities such as Twitter and Reddit. RippleAI and SNU teams have achieved overwhelming result of sweeping the position of first place in both competition categories, proving their smart interactive artificial intelligence technology that allows machines to comprehensively understand satire and metaphors that are difficult for machines understand. For a machine's AI to understand the conversations or postings that it has been exposed to is one of the most basic elements needed to understand and proceed with the conversation as naturally as a human being would.
 
The event was hosted by the FigLang2020 workshop that was organized by ETS and Columbia University of the U.S., where the RippleA-SNU team will receive the overall first prize at the ACL conference on July 9.
 
Other than the three competitions, the RippleAI-SNU team has a world-class technological prowess that has won several competitions for years, including the search/recommendation/question challenge for movies and videos based on natural language. "The technologies that won first place will be unveiled as a demo and core solution for Ripple AI," said Professor Gunhee Kim.
 
Reference link: https://competitions.codalab.org/competitions/22247