동국대학교 인공지능학과

커뮤니티

공지사항

인공지능학과 연사 초청 온라인 세미나 안내

페이지 정보

작성자 관리자 조회184회 작성일 21-09-06 09:56

본문

2021학년도 2학기
동국대학교 일반대학원 인공지능학과 연사 초청 온라인 세미나 일정

 

9/10(금) 10:00

 - 연사 : Prof. Jean Oh

                  https://www.cs.cmu.edu/~./jeanoh/

 - 연사 소속 Carnegie Mellon University

 - 주제 Core Challenges of Embodied Vision-Language Planning

 - 내용 Service dogs or police dogs work on real jobs in human environments. Are robots ready to perform service tasks in a real, physical space? Embodied AI or robot intelligence is generally considered as one of the ultimate AI problems that would require complex, integrated intelligence combining multiple subfields of AI including natural language understanding, visual understanding, planning, reasoning, inferencing, and prediction. While steep progresses have been witnessed in several subfields in recent years, the field of embodied AI remains extremely challenging. In this talk, we will focus on the Embodied Vision-Language Planning (EVLP) problem to understand the unique technical challenges imposed at the intersection of computer vision, natural language understanding, and planning problems. We will review several examples of the EVLP problem to discuss the current approaches, training environments, and evaluation methodologies. Through in-depth investigation of the current progress on the EVLP problem, this talk aims to assess where we are in term of making progress in EVLP and facilitate future interdisciplinary research to tackle core challenges that have not been fully addressed.

- Webex 정보

- 미팅 링크 : https://dongguk.webex.com/dongguk/j.php?MTID=meaab6584bc741187b3c5c1dfa9e94b3b

- 비밀번호 : aixx

 9/24(금) 10:00

 연사 : Prof. Raymond J. Monney

     https://www.cs.utexas.edu/~mooney/

 - 연사 소속 The University of Texas at Austin

 - 주제 Dialog with Robots: Perceptually Grounded Communication with Lifelong Learning
 - 내용 : 
Developing robots that can accept instructions from and collaborate with human users is greatly enhanced by an ability to engage in natural language dialog.  Unlike most other dialog scenarios, this requires grounding the semantic analysis of language in perception and action in the world.  Although deep-learning has greatly enhanced methods for such grounded language understanding, it is difficult to ensure that the data used to train such models covers all of the concepts that a robot might encounter in practice. Therefore, we have developed methods that can continue to learn from dialog with users during ordinary use by acquiring additional targeted training data from the responses to intentionally designed clarification and active learning queries.  These methods use reinforcement learning to automatically acquire dialog strategies that support both effective immediate task completion as well as learning that improves future performance. Using both experiments in simulation and with real robots, we have demonstrated that these methods exhibit life-long learning that improves long-term performance.

 

 
- Webex 정보

- 미팅 링크 : https://dongguk.webex.com/dongguk/j.php?MTID=m5381ee3385c09d471b2d296eb5fe5819
비밀번호 : aixx

 

  10/29(금) 10:00

 연사 : Prof. Sang Wan Lee

     https://bioeng.kaist.ac.kr/index.php?mid=bio_03_01&document_srl=6844

 - 연사 소속 : KAIST

 - 주제 : 뇌모사 강화학습 기술
 - 내용 : 본 세미나에서는 계산신경과학 연구와 기계학습 기술을 융합하여 뇌가 가진 고위수준의 학습 능력을 탐구하는 본 연구팀의 3단계 기술을 소개할 예정입니다. 먼저 과소적합(underfitting)과 과적합(overfitting) 없이 인간의 강화학습 과정을 모델로 이식하는 학습 파이프라인 연구를 소개하고 (레벨 1기술), 도출된 뇌의 모델 기반 강화학습(model-based reinforcement learning)이 어떻게 기계학습의 근본적인 문제인 편향-분산 오류 최소화(bias-variance tradeoff)나 성능-계산량-속도 균형(performance-efficiency-speed tradeoff)을 유지하는지에 대해 논의할 예정입니다 (레벨2 기술). 끝으로 뇌의 강화학습 과정을 제어하는 새로운 기계학습 패러다임(Neural task control)에 대해서 소개할 예정입니다 (레벨3 기술).

이러한 기술개발의 궁극적인 목표는 문제해결 관점에서 형식화된 인공지능이라는 테두리 안에서 인간의 지능을 이해하는 것입니다. 관련 연구의 활성화를 위해 2019년 설립된 KAIST 신경과학-인공지능 융합연구센터에서는 인간의 고위수준 기능들을 기계학습 알고리즘으로 구현하는 연구를 수행하고 있습니다.

 
- Webex 정보

- 미팅 링크 : https://dongguk.webex.com/dongguk/j.php?MTID=mdd06a08a47c3569d1c2552463c3a0199
비밀번호 : aixx
 

   11/12(금) 10:00

 연사 : Prof. Yoonsuck Choi

     https://engineering.tamu.edu/cse/profiles/cyoonsuck.html

 - 연사 소속 : Texas A&M University

 - 주제 : Overcoming Limitations of Deep Learning
 - 내용 : Deep learning has revolutionized the field of artificial intelligence. However, there are several limitations, some that are practical and some that are more fundamental. In this seminar, I will briefly discuss practical limitations of deep learning such as catastrophic forgetting, need for massive amounts of data, etc., and move on to talk about more fundamental issues that are inspired by brain science such as the source of meaning in the brain, the concept of extended mind, and prediction and consciousness.

 
- Webex 정보

- 미팅 링크 : https://dongguk.webex.com/dongguk/j.php?MTID=m5501b040cbc0d13cf6ee904063fb1e25
비밀번호 : aixx
 

    11/26(금) 10:00

  연사 : Prof. Wonseok Jeon

 - 연사 소속 : Qualcomm AI Research

 - 주제 : Learning from Demonstration: Imitation Learning, Inverse Reinforcement Learning, and Offline Reinforcement Learning
 - 내용 : The availability of prior information from datasets has been a crucial factor contributing to the success of machine learning algorithms. Particularly in control problems, learning from demonstration (LfD) has addressed several issues on reinforcement learning from scratch, i.e., overcoming exploration problems, avoiding complicated reward specification, and so on. In this talk, I will introduce general concepts and algorithms for imitation learning, inverse reinforcement learning, and offline reinforcement learning. I will also briefly talk about my research efforts on each topic.

 
- Webex 정보

- 미팅 링크 : 미정
비밀번호 : aixx
 

12/03(금) : 10:00

 - 연사 Prof. Minjun Kim

 - 연사 소속 : KAIST

 - 주제 : 복잡한 환경에서의 비행매니퓰레이션
 - 내용 : Aerial manipulation is an emerging research field after success stories of multi-rotor studies which have impacted not only research community but also markets. There are several branches of aerial manipulation studies, but I am particularly interested in performing manipulation tasks in a complex industrial site. Risk management, however, is a keyword that hinders us from performing manipulation tasks in such a complex site, because the crash of an aerial system will essentially cause a big trouble. To deal with this difficulty, we have developed a new system called SAM (stands for cable-Suspended Aerial Manipulator) that inherently minimizes the risk. In this talk, I will introduce design features of SAM, and will introduce how we integrated different functionalities such as self-stabilization of SAM, vision-based object localization, and telemanipulation, in order to accomplish an industrial scenario. This scenario, which is designed under an EU project AEROARMS, includes deployment and retrieval of a robotic crawler for pipe inspection. I will conclude the talk by sharing our ideas on the future directions.


 - Webex 정보

- 미팅 링크 : https://dongguk.webex.com/dongguk/j.php?MTID=m471ccb0ea36fc9de8aa6927d52e6264f
비밀번호 : aixx

 12/17(금) : 10:00

 - 연사 Daegil Park, Ph.D.

 - 연사 소속 : 선박해양플랜트연구소

 - 주제 : Perception, cognition and control for autonomous marine robots
 - 내용 : Marine robot is one of most challenging area because of the extreme underwater environment. Marine robot can not use the conventional robot sensor, actuator and algorithm due to hydraulic lossy medium condition, and it must cosider the water pressure and waterproof along to the depth. In this saminar, I will speak about the characteristics of marine(underwater) condition, and our group's research experiences to operate robots in underwater fields. In particular, I will talk about perception, cognition, and control approaches that researched to solve the problems of autonomous navigation and control system of marine robots. And I would like to talk with audience to come up with ways to overcome marine robots while talking about the need for AI technology in the ocean, which is still an undeveloped field.

Webex 정보

- 미팅 링크 : https://dongguk.webex.com/dongguk/j.php?MTID=m00a3d030b3c15e5781a458c6556cee29
비밀번호 : aixx


04620 서울시 중구 필동로 1길 30 동국대학교 신공학관 10152호 전화: 02-2290-1442
Copyright(c) 2016 DONGGUK UNIVERSITY. ALL RIGHTS RESERVED.