이홍락 교수(University of Michigan)
Title : Towards Learning AI Agents for Solving Real-world Tasks
Abs :
A key objective in advancing AI technology is to develop agents that can help human users by autonomously solving complex tasks across various domains, such as task-oriented conversational systems and online transactions. Achieving this requires integrating advanced learning methods, including sequential decision making, learning from feedback, multimodal understanding, and large language models (LLMs). In this talk, I will discuss our recent work on improving planning, decision-making, and learning in AI agents. For tasks like task-oriented dialog, we explore how to enhance language models by augmenting them with subtask graphs to model task dependencies, and we introduce a novel approach using code representations to extract and enforce action preconditions, thereby improving few-shot policy learning in both dialog and embodied environments. Furthermore, we enhance LLM-based agents for web tasks through unsupervised intent discovery and self-exploration, improving decision-making without direct fine-tuning. Additionally, to address knowledge gaps and learning inefficiency in LLM-based agents, we propose a framework that extracts state-aware guidelines from offline experiences, significantly boosting agent performance in various generalization scenarios. We hope these advancements pave the way towards more adaptable, efficient, and intelligent AI systems that can seamlessly interact with and assist human users in real-world environments.
Bio :
Honglak Lee is currently the Chief Scientist of Artificial Intelligence and Executive Vice President at LG AI Research, as well as a Professor of Computer Science at the University of Michigan, Ann Arbor. Prior to his role at LG AI Research, he worked as a Research Scientist on the Google Brain Team. He received his Ph.D. from the Computer Science Department at Stanford University in 2010, advised by Prof. Andrew Ng. His research focuses on deep learning and representation learning, covering areas such as learning from minimal supervision, transfer learning and generalization, and reinforcement learning and sequential decision making. His work has been recognized with awards at prestigious conferences, including ICML, RSS, and NAACL, among others. He has served as an associate editor for IEEE TPAMI and JMLR, and as an area chair for major conferences like ICML, NeurIPS, CVPR, ICCV, ECCV, AAAI, IJCAI, and ICLR. He received the Google Faculty Research Award (2011), NSF CAREER Award (2015), and was selected as one of AI's 10 to Watch by IEEE Intelligent Systems (2013). He is also a research fellow of the Alfred P. Sloan Foundation (2016) and a foreign member of the National Academy of Engineering, Korea (2022).
Prof. Yoshua Bengio(Université de Montreal)
Title : Mitigating Catastrophic AI Risks with Safe-by-Design AI
Abs :
We are on a path towards human-level AI, also called AGI, with uncertain timelines and uncertain risks, ranging from threats to democracy and national security to existential risks due to loss of control to a rogue AI. In spite of these risks, corporations are competing fiercely to build AGI and racing ahead. To make sure they do not cut dangerous corners and to make sure the power of future AIs is not abused or destabilizes the geo-political order, we will need significant effort in how to manage these projects, ranging from organization-level governance to national regulation and international treaties, including to avoid dangerous proliferation, harmonization of policies and a move towards AI for the public good, at a global level. Mitigating those risks should therefore be a global priority and one of the key missing scientific advancement is a methodology to achieve high levels of AI capabilities that also provide strong safety assurances. We propose a safe-by-design AI approach based on risk bounds that could be used to avoid dangerous AI actions and that could be estimated using amortized probabilistic inference machine learning methods.
Bio :
Yoshua Bengio is Full Professor in the Department of Computer Science and Operations Research at Université de Montreal, as well as the Founder and Scientific Director of Mila and the Scientific Director of IVADO. He also holds a Canada CIFAR AI Chair. Considered one of the world’s leaders in artificial intelligence and deep learning, he is the recipient of the 2018 A.M. Turing Award, considered like the "Nobel prize of computing". He is a Fellow of both the Royal Society of London and Canada, an Officer of the Order of Canada, Knight of the Legion of Honor of France and member of the UN’s Scientific Advisory Board for Independent Advice on Breakthroughs in Science and Technology.
Fatih Porikli(Qualcomm AI Research / Senior director, Global Lead of AI Systems at Qualcomm AI Research)
Title : AI on the Edge
Abs :
Generative AI emerges as a transformative force, capable of creating new multimodal content such as text, speech, images, video, and more, while handling complex dialogues and reasoning about problems. This disruptive technology is reshaping traditional approaches across various domains and redefining the user interface to computing devices. Its impact transcends industries, promising substantial advancements in utility, productivity, and efficiency. As the adoption of generative AI accelerates, its computational demands are surging, and thus on-device processing becomes more important than ever. In this talk, we will look into the pivotal role of on-device AI deployment and full-stack AI optimizations to make it possible.
Bio :
Fatih Porikli is an IEEE Fellow and a Senior Director of Technology at Qualcomm AI Research. Previously, he was a full Professor in the Research School of Engineering, Australian National University. He also served as the Vice President at Futurewei, the Chief Scientist of Autonomous Vehicles at Huawei, the Computer Vision Research Group Leader at NICTA, and a Distinguished Research Scientist at Mitsubishi Electric Research Laboratories. He was the recipient of the R&D100 Scientist of the Year Award in 2006. He has won 8 best paper awards at premier conferences and 7 other professional awards, co-authored more than 400 publications, and co-organized more than twenty workshops.
박태웅 의장(녹서포럼)
Title : 미디어의 관점에서 바라본 인공지능, 현황과 전망
Abs :
세계적인 미디어 학자 마셜 맥루한은 “미디어는 메시지다”라는 유명한 말로 새로운 미디어의 등장이 사회에 불러오는 전면적인 변화에 관해 얘기했다. AI를 새로운 미디어의 등장의 관점에서 해석한다. 새로운 미디어로서의 AI가 불러올 변화를 짚어보고, 인공지능이 우리에게 던지고 있는 과제를 얘기한다.
Bio :
- 학력사항
81~85년 서울대학교 경영학과 졸업
- 상훈
2021년 동탑산업훈장
- 저서
2023년 <박태웅의 AI강의>
- 주요 경력사항
2024.3.20~ 현재 녹서포럼 의장
2018.09.01~ 2024.3.19 한빛미디어 이사회 의장
2016.11~2018.06.30 (주)푸드테크 / 부사장
2014.03~2023.03.30 KST&Partners 대표
2009.04~2013.03 KTH COO/CTO 부사장
2007.07~2009.03 열린사이버대학교(www.ocu.ac.kr) / 부총장
2004.07~2006.08 ㈜엠파스(검색 포털) /COO 부사장
주재걸 교수(KAIST)
Title: 컴퓨터 비전 분야 생성AI 기술 동향 및 연구 사례
Abs:
최근 컴퓨터 비전 분야에서 이미지 및 비디오 합성, 그리고 3D 복원 및 합성 기술은 나날이 발전하고 있으며, 그러한 기술은 여러 실제 어플리케이션 등에서 활발히 적용되고 있다. 본 발표에서는 이러한 발전을 가져온 핵심 기술인 Diffusion model, 3D Gaussian splatting 등을 소개하고, 이들을 대상으로 다양한 사용자 요구 사항을 유연하게 반영할 수 있는 기술, 그리고, 이에 대한 구체적인 어플리케이션인 Virtual try-on, Neural talking head, pose retargeting 등의 다양한 연구 사례를 소개한다.
Bio:
- 학력
. 2013: 박사, Computational Science and Engineering, Georgia Tech
. 2009: 석사, Electrical and Computer Engineering, Georgia Tech
. 2001: 학사, 서울대학교 전기공학부
- 경력
. 2020.03 - 현재: KAIST 김재철AI대학원 부교수
. 2019.09 – 2020.02: 고려대학교 인공지능학과 부교수
. 2015.03 – 2019.08: 고려대학교 컴퓨터학과 조교수
. 2011.12 – 2015.02: Research Scientist, Georgia Tech
