KAERI Ranked Runner-Up in International Video Story Understanding AI Contest

- KAERI is conducting a research on detecting abnormalities and defects of nuclear power plants using AI-based nuclear reactor data analysis

□ Early in this year, a Korean company which is specialized in artificial intelligence (AI) graphics drew people’s attention, as it gave an AI-painted portrait to Lee Se-dol, a professional Korean Go player of 9 dan rank. It is a prime example of the fact that , AI stands out even in the fields of literature and arts, which require the unique process of ‘creation’ and thus have been considered the exclusive domain of mankind.
ㅇ Earlier in April 2016, a project named ‘the Next Rembrandt’ was unveiled where AI learned 346 artworks created by Rembrandt and collected and analyzed the graphic data amounting to 150 gigabytes. The results received favorable review as ‘the resurrection of Rembrandt,’ except for a minor inconsistency in the brushstrokes. In 2018, an AI composer Emmy composed Chopin-style musical pieces, which touched the hearts of musicians and surprised the world. In the same year, the Botnik Studio in the United States employed their AI Predictive Text Algorithm to publish a new Harry Potter chapter, entitled ‘Harry Potter and the Portrait of What Looked Like a Large Pile of Ash,’ which fascinated Harry Potter fans. These examples show how AI is rapidly developing in literature and arts.

□ With the increase of AI research, contests have been launched  to compare which AI systems best understand videos, such as movies and dramas. One prime example is the international AI contest of Video Story Understanding, held at the European Conference on Computer Vision (ECCV) Workshop (hereinafter referred to as ‘Drama QA Challenge’). Interestingly, one team from Korea Atomic Energy Research Institute (KAERI) participated in the contest and was chosen as the challenge winner. KAERI announced on August 10 that the ‘HARD KAERI’ team from its Intelligent Computing Lab ranked runner-up in this year’s Drama QA Challenge held on August 1, demonstrating Korea’s technological prowess in intelligent image analysis.
ㅇ As one of the challenge winners, KAERI will be invited to the international ECCV conference on August 28 to receive the prize and present the results of the relevant research. KAERI participated in the contest through the National Information Society Agency’s AI Learning Data Establishment Project jointly conducted by EuclidSoft.

 □ The Drama QA Challenge, hosted by the Seoul National University since 2019 and held for the second time this year, is a video story understanding contest with participation by global companies and universities around the world. The participating teams compete with each other through their AI systems by evaluating how well they understand the contents of the videos and the conversations therein.
 ㅇ This year, 11 qualified teams competed with their technological prowess in a blind test. AI algorithms were developed based on the Korean TV drama "Another Miss Oh," by using composite data about the conversations between the speakers, the emotions of the speakers and others, in preparation for question-and-answer sessions. The accuracy was then evaluated by choosing the candidates giving the best answers to five-option multiple-choice questions. For example, when given a question such as “Why did Haeyeong’s Mom get angry with her?”, the AI system chooses the answer based on the contents of the TV drama.
ㅇ The video story understanding requires high-level AI technology to understand the drama contents and question and answer in natural language.

□ The Intelligent Computing Lab of KAERI has steadily accumulated AI technology by participating in various AI content in video security and video recognition. Winning the prize in the Drama QA Challenge, KAERI demonstrated its world-class intelligent video analysis technology. In addition to the AI-based intelligent video analysis technology, KAERI is working on the development of technologies for detecting the abnormality and defects of nuclear power plants  based on  nuclear reactor data analysis.

□ Dr. Lim Kyung-tae and Dr. Lee You-han of the Intelligent Computing Lab, who led the ‘HARD KAERI’ team in the contest, said, “We really enjoyed the time spent studying both image processing and natural language processing. We will get the first prize next time.” Dr. Yu Yong-gyun, head of the Intelligent Computing Lab, said, “Our intelligent image processing technology will contribute to the safe operation of nuclear power plants. We will make efforts to develop AI technologies that can protect those who are vulnerable in our society, including the visually impaired, children, and the elderly.