Loading...

학술행사

Korean AI Association

학술행사

  >   학술행사   >   국내학술대회

국내학술대회

튜토리얼
 
백승렬 교수 (UNIST)
 
Title: Recent Advances Towards Efficient Transformers in Vision Domain
 
Abs:
Transformers have already replaced LSTMs in the NLP domain and they are also replacing CNNs in the vision domain. However, their huge computational cost may hinder the practical use in real-time industrial applications. This tutorial will first briefly introduce basics of Transformers and will cover recent techniques that tried to achieve the efficient Transformers in the context of vision applications.
 
Bio: 
Seungryul Baek is an associate professor at the Artificial Intelligence Graduate School (AIGS) and Department of Computer Science and Engineering (CSE) of UNIST. He obtained BS (2009) and MS (2011) degrees from Dept. of Electrical Engineering at KAIST and Ph.D. degree (2020) from Dept. of Electrical and Electronic Engineering at Imperial College London. Before joining Ph.D., he was an employee at DMC Research Center of Samsung Electronics for four years (2011.2.-2015.2.). 
 
 
 
김형훈 교수 (UNIST)
 
Title: Large Language Models: Ability and Alignment
 
Abs:
Large Language Models (LLMs) demonstrate remarkable abilities across a variety of tasks, highlighting significant advancements in AI technology. This presentation will begin by exploring the foundational concepts of LLMs, including language modeling and the attention mechanisms essential for their development and training. Building on this foundation, we will discuss the emergent abilities of LLMs, particularly their innate reasoning capabilities. We will then highlight how these abilities are not only intrinsic but can also be significantly amplified through innovative techniques such as advanced prompt engineering and instruction-tuning.
 
Bio: 
Hyounghun Kim is an assistant professor in the Artificial Intelligence Graduate School and the Department of Computer Science and Engineering at the Ulsan National Institute of Science and Technology (UNIST). He earned his Ph.D. from the Department of Computer Science at UNC-Chapel Hill, advised by Prof. Mohit Bansal. He received M.S. in Computer Science from USC and B.S. in Electrical and Electronic Engineering from Yonsei University. His research interests are in natural language processing and multimodal learning focusing on embodied AI, commonsense reasoning, VQA/VideoQA, visual dialogue, as well as AI assistant applications. He has been serving as an action editor (area chair), reviewer, and program committee member for ACL Rolling Review, top conferences, and workshops. He also worked for Adobe Research and Amazon Alexa AI as an intern, and Samsung Electronics as a software engineer.
 
 
 
박노성 교수 (KAIST)
 
Title: Physics-inspired Deep Learning
 
Abs:
Scientific knowledge, written in the form of differential equations, plays a vital role in various deep learning fields. In this talk, I will present a graph neural network (GNN) design based on reaction-diffusion equations, which addresses the notorious oversmoothing problem of GNNs. Since the self-attention of Transformers can also be viewed as a special case of graph processing, I will present how we can enhance Transformers in a similar way. I will also introduce a spatiotemporal forecasting model based on neural controlled differential equations (NCDEs). NCDEs were designed to process irregular time series in a continuous manner and for spatiotemporal processing, it needs to be combined with a spatial processing module, i.e., GNN. I will show how this can be done.
 
Bio:
Noseong Park is currently a tenured associate professor of the School of Computing and the director of the Big Data Analytics and Learning Laboratory at Korea Advanced Institute of Science and Technology (KAIST) in South Korea. Before joining KAIST, he was a tenured associate professor at Yonsei University from 2020 to 2024 and was an assistant professor at George Mason University from 2018 to 2019 and the University of North Carolina at Charlotte from 2016 to 2018. He received his computer science Ph.D. in 2016 from the University of Maryland, College Park under the supervision of Prof. V. S. Subrahmanian. He received his Master's degree from KAIST and Bachelor's degree from Soongsil University. He has much experience on the core fields of data mining and machine learning, and their applications to scientific machine learning, deep generative learning, spatiotemporal processing, time series processing, recommender systems, etc. He has published more than 50 papers in top-tier venues, such as NeurIPS, ICLR, ICML, KDD, WWW, VLDB, ICDM, WSDM, SIGIR, IJCAI, AAAI and so on. In particular, his VLDB paper, which proposed a tabular data synthesis method called TableGAN, is now the most cited among all VLDB papers published within the last 5 years. He received the best student paper award, as an advisor, from IEEE BigData 2022, the best demonstration runner-up award from IJCAI 2019, and the best paper runner-up award from ASONAM 2016. He also received a couple of medals in Samsung Humantech Paper Award, the most prestigious award in Korea, at 2022 and 2023. He has also secured several research grants that amount to, in total, $530K as a PI or co-PI from NSF, the Office of Naval Research, and so forth when he was in the US. Including the grants that he has received in Korea after 2020, the total research fund amounts to $2M. He also has served (or is now serving) as a PC member in many top-tier venues, such as ICML, ICLR, NeurIPS, KDD, WWW, AAAI, ICDM, ICWSM, and so on.
 
 
임성빈 교수 (고려대학교) 
 
Title: Diffusion Models: Foundation and Algorithm
 
Abs:
Diffusion models have recently gained significant attention in probabilistic machine learning due to their theoretical properties and impressive applications in generative AI, including Stable Diffusion and DALL-E. This tutorial provides an introduction to the theory and algorithm for understanding the foundation of diffusion models.
 
Bio: 
Sungbin Lim is currently an Assistant Professor at the Department of Statistics at Korea University. He earned a Ph.D. in Mathematics from Korea University in 2016, under the supervision of Prof. Kyeong-Hun Kim, and transitioned to the private sector as a senior AI Scientist until 2019 and returned to academia in 2020. During his doctoral studies, he focused primarily on research in stochastic partial differential equations and diffusion theory. Since then, he has developed a keen interest in probabilistic machine learning including deep generative modeling, and is also collaborating with LG AI Research on AI for Science projects.
 
- 2023 - Current: Assistant Professor, Dept. of Statistics, Korea University.
- 2020 - 2023: Assistant Professor, Dept. of IE & AIGS, UNIST.
- 2018 - 2019: Research Scientist, Kakao Brain.
- 2016 - 2017: Data Scientist, Samsung Fire & Marine Insurance.
- 2016: Ph.D. in Mathematics, Korea University.
 
 
 
송환준 교수 (KAIST)
 
Title: Innovation in text summarization: the role and prospects of large language models
        텍스트 요약의 혁신: 초거대 언어 모델의 역할과 전망
 
Abs:
본 강연에서는 텍스트 요약에 관한 세 가지 핵심 주제를 다룹니다. 첫째로, 텍스트 요약 방법에 대한 소개와 함께, 기존 생성형 언어 모델을 활용한 문서 요약과 초거대 언어 모델을 이용한 문서 요약에 대해 살펴봅니다. 둘째로, 거대 언어 모델의 추론 한계인 환각, 완전성, 간결성 문제들을 분석하고, 전통적인 문서 요약 평가 방법의 한계를 논의합니다. 그리고 거대 언어 모델을 활용한 자동 평가 방법에 대해 알아봅니다. 셋째로, 벤치마킹 데이터의 구축과정부터 초거대 언어 모델의 벤치마킹 및 분석, 그리고 이에 대한 항후 연구 과제에 대해 다룹니다.
 
Bio: 

- (현) 한국과학기술원 산업및시스템 공학과 조교수

- (현) Research Fellow at Asian Trustworthy Machine Learning (ATML) Group 

- Amazon Web Services Inc. 근무 (미국/정규직) 

- NAVER AI Lab 근무 (한국/정규직) 

- Google Research 근무 (미국/인턴) 

- 한국과학기술원 데이터사이언스 대학원 박사

 
 
최종현 교수 (서울대학교)
 
Title: Practical set-ups and methods for continual learning​
 
Abs:
Prevalent setups in continual learning often fall short of practical applicability, posing unrealistic assumptions and constraints. First, I will try to address these gaps by presenting various continual learning setups (mostly for class incremental learning) that are more realistic and feasible, on which we would need to evaluate our algorithms. Then, I will discuss one of our approaches to address a continual learning (e.g., class incremental learning) in a bit realistic set-up, online continuous data stream and beyond.
 
Bio: 

I am an associate professor in Seoul National University and was an associate professor in Yonsei University (2022-2024), an assistant professor in GIST (2018-2022) and a researcher in Allen Institute for Artificial Intelligence (AI2) (2016-2018) and Comcast Labs, DC (2015). During my PhD, I was fortunate to work as a research intern in Microsoft Research, Redmond (2014 Summer), Disney Research, Pittsburgh (2014 Spring), Adobe Research, San Jose (2013 Summer) and U.S. Army Research Lab. (Adelphi, MD, Summer 2011).

I received a Ph.D. degree from University of Maryland, College Park, under the supervision of Prof. Larry S. Davis, and a B.S. and a M.S. degree from Seoul National University, under the supervision of Prof. Kyoung-Mu Lee in SNU Computer Vision Lab.