Machine Intelligence and Information Learning Laboratory

기계지능 및 정보학습 연구실

관련기사 바로가기
기계지능 및 정보학습 연구실

기계지능 및 정보학습 연구실에서는 현재의 심층학습(deep-learning) 기술을 뛰어넘는 차세대 머신러닝 알고리즘을 연구합니다. 현재 심층학습 알고리즘은 다음과 같은 한계를 가지고 있습니다: i) 막대한 양의 학습 데이터에 의존하는 한계점, ii) 학습이 되지 않은 상황에 대해 성능이 크게 저하되는 한계점, iii) 한번 학습이 완료되면 학습된 지식이 확장이 어려운 한계점. 이러한 한계를 극복하기 위해 우리 기계지능 및 정보학습 연구실에서는 적은 수의 학습 데이터로 학습 성능을 달성할 수 있는 소수샷 학습(few-shot learning), 새로운 학습 태스크/상황에 대해서도 높은 성능을 보일 수 있는 메타학습(meta-learning), 기존 지식을 보존하면서 새로운 태스크를 점진적으로 학습할 수 있는 점진적 학습(incremental learning) 및 연속 학습(continual learning) 알고리즘 개발을 연구합니다. 또한 향후 연구 방향으로 메타 강화학습(meta reinforcement learning) 및 인과관계 학습(causal learning) 연구를 목표로 하고 있습니다.
다가올 4차 산업혁명에서는 인공지능에 기반한 알고리즘 개발 뿐만 아니라 인공지능 알고리즘을 위한 통신, 분산시스템(distributed system)의 혁신 또한 필요합니다. 우리 기계지능 및 정보학습 연구실은 정보이론(information theory) 및 통신이론(communication theory) 연구 경험을 인공지능 알고리즘 개발과 접목하여 차세대 6G 지능형 통신 기술 그리고 분산 학습/통신/저장이 자동적으로 이루어지는 분산 시스템의 구성 알고리즘 개발을 연구하고 있습니다.
In our Machine Intelligence and Information Learning laboratory (MIIL lab), we focus on developing novel machine learning algorithms which overcome challenges for future intelligence systems beyond deep-learning algorithms. The current deep learning algorithms are i) dependent on a massive amount of training data, ii) hard to generalize to unseen tasks and iii) difficult to learn a new concept on the learned knowledge. To tackle these problems, members of our lab are interested in following topics: few-shot learning, meta-learning, lifelong learning (continual and incremental learning), meta-reinforcement Learning and causal learning.
Also, our research interests include intelligence information/communication systems such as 6-Generation wireless communications. For the beyond 5/6G communication systems, resources such as storage/communication and computing capabilities should be deployed on highly-distributed and connected devices or servers for supporting intelligent services (autonomous driving, image recognition, natural language processing, etc.). We are researching about theoretical understanding of distributed learning systems while considering the tradeoffs of communications/storage and computing capabilities. Also, we are interested in developing advanced federated learning algorithms which are suitable for future communication systems.

Major research field

Meta-learning, Few-shot learning, Lifelong learning, distributed learning, federated learning, 6G communications

Desired field of research

Meta-learning, Few-shot learning, Lifelong learning, distributed learning, federated learning, 6G communications

Research Keywords and Topics

In a broad sense, my research direction is to develop novel learning algorithms which overcome remaining challenges of deep-learning techniques. The current deep learning algorithms are i) dependent on a massive amount of training data, ii) hard to generalize to unseen tasks and iii) difficult to learn a new concept on the learned knowledge. To tackle these problems, I`m interested in following topics: few-shot learning, meta-learning, lifelong learning (continual and incremental learning), meta-reinforcement Learning and causal learning.

Research Publications

Sung Whan Yoon*, Do-Yeon Kim*, Jun Seo and Jaekyun Moon "XtarNet: Learning to Extract Task-Adaptive Representation for Incremental Few-Shot Learning," Proceedings of the 37th International Conference on Machine Learning (ICML), Vienna, Austria, PMLR 119, 2020. *equal contribution.
Sung Whan Yoon, Jun Seo and Jaekyun Moon "TapNet: Neural Network Augmented with Task-Adaptive Projection for Few-Shot Learning," Proceedings of the 36th International Conference on Machine Learning (ICML), Long Beach, CA, USA, PMLR 97:7115-7123, 2019.
Jy-yong Sohn, Beongjun Choi, Sung Whan Yoon and Jaekyun Moon, “Capacity of Clustered Distributed Storage,” IEEE Transactions on Information Theory, vol. 65, no. 1, pp. 81-107, Jan. 2019

Patents

[US2] Jaekyun Moon, Soonyoung Kang and Sung Whan Yoon, “Controller and oper-
ating method thereof,” Notice of Allowance (NOA) Sep. 9, 2019, Application Number:US15601039.

[US1] Jaekyun Moon, Beongjun Choi and Sung Whan Yoon, “Controller, semi-
conductor memory system and operating method thereof,” Registration Number: US 10,439,647 Oct. 8, 2019.