For Intel® Geti™, an active learning AI platform desiring high-interactivity, I hold a technical leadership driving the development of the data (Datumaro) and training backends (OpenVINO™ Training Extensions). With a strong focus on few-shot transfer, I spearhead efforts to study domain discrepancy analysis and self-supervised domain adaptation. Additionally, I am deeply interested in the application of supervised contrastive learning techniques. By leveraging these methodologies, I have achieved state-of-the-art retrieval performance through binary hashing (refer to CVIU journal).

While conducting research in the field of localization and 3D reconstruction, I specifically focused on the study of monocular camera-based depth estimation. A particular area of interest was self-supervised learning, which offers intriguing possibilities for enhancing performance without relying on external cost-expensive supervision.

As one of seed members of A-Project team in System LSI, where is the department of silicon design in Samsung Electronics, I had responsibilities for automatic camera calibration and simultaneously localization and mapping (SLAM) algorithms. My research interests are multi-view geometry and 3D reconstruction through the graph optimization.

Contributing to the "Real-time Scene Understanding" project, I engaged in tackling challenges such as object detection, visual tracking, and heterogeneous sensor fusion for collision avoidance capabilities for self-driving systems.