Our research group explores industrial AI through the lens of response surface optimization problems, where machine learning algorithms are used to estimate responses. This talk will begin by presenting two motivating projects from real-world industries, followed by an exploration of a key challenge: the disruption of essential input-output relationships, such as monotonicity. To overcome this issue, we identified the need to incorporate monotonicity constraints during model training with real-world data. This talk will introduce two recent works that address this challenge by embedding monotonicity as an inductive bias in neural network training. The first is the development of a new neural network architecture called Scalable Monotonic Neural Networks (SMNN). The second focuses on combining SMNN with Neural Additive Models (NAM) to achieve both monotonicity and interpretability in neural network learning. The talk will conclude by outlining potential future research directions in this area.

 

#Bio :

Jong-Seok Lee received his Ph.D. in industrial engineering from Iowa State University, USA. Following his graduation, he joined the Research and Development Group at SAS Institute in the USA, where he worked on statistical software development. Subsequently, He held positions as an assistant and associate professor in the Department of Systems Management Engineering at Sungkyunkwan University, Republic of Korea. Currently, he is an associate professor in the Department of Industrial and Systems Engineering at KAIST, where he leads the AIME Laboratory. His research focuses on developing learning algorithms specifically for manufacturing data with applications ranging from autonomous equipment control to design optimization.

 

-----------------------------------------------------------------------------------------------------------------------------------------------

 

 

# Location: E2 1501

# Presenter: prof. JongSeok Lee, KAIST

Our research group explores industrial AI through the lens of response surface optimization problems, where machine learning algorithms are used to estimate responses. This talk will begin by presenting two motivating projects from real-world industries, followed by an exploration of a key challenge: the disruption of essential input-output relationships, such as monotonicity. To overcome this issue, we identified the need to incorporate monotonicity constraints during model training with real-world data. This talk will introduce two recent works that address this challenge by embedding monotonicity as an inductive bias in neural network training. The first is the development of a new neural network architecture called Scalable Monotonic Neural Networks (SMNN). The second focuses on combining SMNN with Neural Additive Models (NAM) to achieve both monotonicity and interpretability in neural network learning. The talk will conclude by outlining potential future research directions in this area.

#Bio :

Jong-Seok Lee received his Ph.D. in industrial engineering from Iowa State University, USA. Following his graduation, he joined the Research and Development Group at SAS Institute in the USA, where he worked on statistical software development. Subsequently, He held positions as an assistant and associate professor in the Department of Systems Management Engineering at Sungkyunkwan University, Republic of Korea. Currently, he is an associate professor in the Department of Industrial and Systems Engineering at KAIST, where he leads the AIME Laboratory. His research focuses on developing learning algorithms specifically for manufacturing data with applications ranging from autonomous equipment control to design optimization.