| Position | Professor |
|---|---|
| Organization | School of Electrical Engineering, KAIST |
| iskweon@kaist.ac.kr |
| BA | Seoul National University |
|---|---|
| MS | Seoul National University |
| PHD | Carnegie Mellon University |
| 2010-present | Head, Future Vehicle Division, School of Electrical Engineering, KAIST |
|---|---|
| 2015-2018 | KEPCO Endowed Chair Professor, KAIST |
| 2017-2020 | Advisory Professor, Samsung Research, Samsung |
| 2010-2017 | Director, The P3 DigiCar Center, KAIST. |
| 1992-present | Professor, School of Electrical Engineering, KAIST |
| 1991-1992 | Researcher, Toshiba R&D Center, Toshiba |
| Deep Learning AI: Challenges and Limitations |
| In recent years, the performance of Deep Learning AI models for many computer vision tasks has surpassed human performance in some benchmarking datasets. However, there seems to be a large gap in terms of performance for real-world applications, mainly due to the data problem and the robustness of the deep learning models. In this talk, we first present two practical applications of Deep Learning AI models to show the feasibility and limitations of the cutting-edge deep learning technology. The first use case is the application of a deep learning model to a reverse vending machine (SuperBin Nephron) for recycling bottles and cans. We then present another use case of a deep learning model (Lunit INSIGHT) for detecting breast cancer in a mammogram. We show some fundamental limitations, such as the data problem and the robustness, of those deep learning models for real-world deployment. We then present some recent deep learning models by KAIST EE for overcoming these fundamental limitations. |
