Loading...

학술행사

Korean AI Association

학술행사

  >   학술행사   >   국내학술대회

국내학술대회

연사 및 초록 (21일)
 
 
김광호 교수(고려대학교)
 
Title:  Introduction to Causal and Counterfactual Learning
 
Abs: 
Causal inference is all about learning counterfactual parameters, i.e., about what would happen to some response when a “cause” of interest is changed or intervened upon. Many problems in modern causal inference often involve non-trivial structures including time-varying confounding, non-overlapping covariate distributions, high degree of heterogeneity, often with very complex, large datasets. These inherent complexities are not properly addressed in the standard approaches for causal Inference. In the first part of my talk, I will give a gentle introduction to statistical causal learning. The second part will offer a brief overview of recent efforts to  integrate classical causal inference with modern machine learning, highlighting a novel framework of counterfactual prediction that facilitates efficient generalization to unseen data and unanticipated scenarios.
 
Bio: 
Kwangho Kim is an assistant professor in the department of statistics at Korea University. Before joining Korea University, he was a Marshall J. Seidman Fellow in the Department of Health Care Policy at Harvard Medical School. He received a Ph.D. in Statistics and Machine Learning and an M.S in Machine Learning and at Carnegie Mellon University, and an M.S. in Statistics from Stanford University. Prior to that, he earned a B.S. in mathematics and a B.S. in electrical engineering from KAIST. Dr. Kim’s research spans causal inference, statistical machine learning, non- and semi-parametric statistics, and topological data analysis. Dr. Kim has received multiple awards, including the ASA, IMS, and WNAR graduate student awards.
 
 
 
신승재 박사(Qualcomm AI Research)
 
Title: Robust Machine Learning under Distribution Shifts
 
Abs: 
In this talk, we will explore robust machine learning strategies for handling distribution shifts, focusing on three complementary approaches: optimization-based methods, data-centric techniques, and weight-averaging strategies. On the optimization front, we will discuss Sharpness-Aware Minimization (SAM), along with methods leveraging gradient norms and Hessian information to achieve flatter minima. From a data-centric perspective, we will highlight how data augmentation contributes to robustness. Finally, we introduce weight-averaging approaches as a distinct yet complementary method to further stabilize and improve model performance. We will conclude by connecting these ideas to the challenges faced by large-scale models, outlining recent trends and open questions in the field.
 
Bio: 
Seungjae Shin is a senior machine learning engineer at Qualcomm AI Research, where he works on model quantization for on-device AI as part of the Quantization System Team. Prior to joining Qualcomm in 2024, he obtained his Bachelor's (2018), Master’s (2020) and Doctoral (2024) degrees in Industrial and Systems Engineering from KAIST.
 
 
 
김동우 교수(POSTECH)
 
Title: Probabilistic Graphical Model I <Latent Variable Models and Variational Inference>
 
Abs: 
본 강의는 PGM의 첫번째 강의로서 기계학습에서 널리 사용되는 확률모델에 대하여 소개하는 것을 목표로 한다. 특히 이러한 확률모델의 파라미터 추정을 위하여 널리 사용되고 있는 변분 추론(variational inference) 및 VAE (Variational Auto-Encoder)에 대하여 강의한다.
 
Bio: 
Dongwoo Kim is an associate professor at POSTECH. He leads the machine learning group (https://ml.postech.ac.kr) along with three other faculty members. His research focuses on the geometric structure of datasets and models and how they influence machine learning algorithms. Previously, he worked as a lecturer (assistant professor) and a post-doctoral researcher at the Australian National University. He holds a Ph.D. and M.S. degrees from the Korea Advanced Institute of Science and Technology (KAIST), as well as a B.E. from Sungkyunkwan University.
 
 
 
 
문일철 교수(KAIST)
 
Title: Probabilistic Graphical Model II <Structured Latent Variable in Diffusion>
 
Abs: 
본 강의는 PGM의 두번째 강의로서, VAE 이후 연구에서 잠재변수 모델링의 방법들을 소개한다. 특히 Diffusion과 같이 시계열 관점에서 잠재변수를 어떻게 모델링해왔으며, 이런 잠재변수의 Diffusion Framework에서의 추론과정을 강의한다. 또한 잠재변수 모델의 다양한 Manipulation (예, Disentanglement등)의 방법에 대한 부분도 강의한다.
 
Bio: 
2011 - 2017 : KAIST 산업 및 시스템 공학과 교수
2008 - 2011 : KAIST 박사후 연구원
2005 - 2008 : Carnegie Mellon University School of Computer Science 박사