Domestic research teams successfully concluded a paper competition using accumulated data for developing artificial intelligence (AI) that understands human behavior, emotions, and experiences.

The Korea Institute of Electronics and Telecommunications (ETRI) announced it held the "2nd ETRI Human-Understanding Artificial Intelligence Paper Competition" following last year's inaugural event.

The competition, which ran for approximately four months starting in February this year, selected creative AI papers through presentation evaluation on the 19th, awarding one grand prize team, two teams with excellence awards, four teams with encouragement awards, and two honorable mention teams.

Notably, this year expanded participation to "any general adult," which is evaluated as contributing to the spread of AI technology research for understanding humans beyond a simple paper competition.

ETRI research teams have accumulated lifelog and multimodal emotion data over several years for developing AI technology that understands humans.

The accumulated data was disclosed on the 'ETRI AI Sharing' platform in early 2022 to allow researchers studying AI to freely utilize it, with continuous updates being made.

This competition was conducted to discover creative and challenging AI research utilizing the data disclosed by ETRI.

The competition, open to the general public, attracted a total of 144 teams comprising 318 participants, with approximately 60 papers submitted.

Participants selected one of two data fields—lifelog or multimodal emotion data—accumulated by ETRI to write their papers.

In the field utilizing the lifelog dataset, participants were required to propose daily information extraction and artificial intelligence service application technologies utilizing various multimodal sensor data extracted from ETRI's lifelog data.

In the field utilizing the multimodal emotion dataset, participants were asked to propose research papers that recognize or predict emotions of participants during conversations involving two or more participants.

The grand prize of this competition was awarded to Korea University's Mixergi team (Jo Dae-hyun, Ryu Hoe-sung).

The Mixergi team proposed a method using modality interactions to effectively combine emotion analysis models through speech and text, earning the distinction of winning the grand prize.

Additionally, Kyung Hee University's HappyEnJoy team (Cho Hyun-gi, Seo Yu-ri, Noh Seol) proposed a methodology for transferring knowledge from a teacher model to a student model using text and speech data from a Korean-based multimodal emotion dataset, and Kyung Hee University's Han Hye-sung team (Han Joo-hyuk, Cho Hye-jung, Kim Tae-sung) presented a multimodal fusion learning method model for emotion recognition, each receiving excellence awards.

Additionally, four teams—Bigbase, ROND4, CutMix-SupCon, and AIFFEL—received encouragement awards, while the SKEZ@901 team and Milab team were selected as honorable mentions through poster presentation evaluation.

The one grand prize-winning team and two excellence award-winning teams received awards from the Minister of Science and ICT along with prize money, while the four encouragement award-winning teams and two honorable mention teams received awards from the ETRI President along with prize money.

ETRI President Bang Seung-chan stated that this competition is expected to "increase social interest in warm AI technology that understands humans and serve as a foundation for more researchers to actively exchange opinions and achieve advances in AI technology."

ETRI was able to hold this competition while conducting years of research on 'human-understanding artificial intelligence' technology that recognizes emotions and behaviors of people to understand their experiences.

The competition was conducted as a cooperative workshop during the 2023 Korea Computer Conference (KCC2023) in cooperation with the Korean Information Scientists Society to encourage participation of domestic research teams and secure academic completeness.