https://pubs.kist.re.kr/handle/123456789/75400 Wed, 15 Apr 2026 17:36:38 GMT 2026-04-15T17:36:38Z https://pubs.kist.re.kr/handle/201004/148624 Title: Organic Molecule Treatment on Interfaces of Semiconducting TMD-Based Field Effect Transistors Authors: Cho, Kyungjune Tue, 09 Nov 2021 00:00:00 GMT https://pubs.kist.re.kr/handle/201004/148624 2021-11-09T00:00:00Z https://pubs.kist.re.kr/handle/201004/113583 Title: Federated Learning for Face Recognition Authors: Kim, Jaehyeok; Kim, Suhyun; Kim, Hyorin; Park, Taehyeong Abstract: With the rapid development of deep learning, the accuracy of face recognition has significantly increased. However, training a face recognition model requires the collection of private data to a centralized server to obtain high performance in the desired domain. Since federated learning is a technique to train a model without collecting data to a server, it is a suitable architecture to train a face recognition model with privacy-sensitive face images held in personal smartphones. This study proposes strategies to apply federated learning to face recognition model training. Fri, 01 Jan 2021 00:00:00 GMT https://pubs.kist.re.kr/handle/201004/113583 2021-01-01T00:00:00Z https://pubs.kist.re.kr/handle/201004/113582 Title: ASAP: Auto-generating Storyboard And Previz with Virtual Humans Authors: Kim, Hanseob; Ali, Ghazanfar; Hwang, Jae-In Abstract: We present a tool for Auto-generating Storyboard And Previz for screenwriters and filmmakers, called ASAP. Our system allows users to easily simulate their stories in the form of 3D animated/visual scenes with virtual humans in a virtual environment. We only ask users to write their script using Final Draft, an exclusive screen-writing tool, and upload them to our system. The uploaded script is parsed into paragraphs of the action, character, and dialogue. From those paragraphs (i.e., text data), our system uses a combination of deep learning, data-driven, and rule-based approaches to instantly generate virtual human's physical motions and co-speech gestures, presenting natural behavior/dialogue scenes. Thus, users can observe automatically generated pre-visualized animations (i.e., previz) from the script and can create the storyboard by capturing scenes being played. Our ASAP can minimize time-and-money consuming and labor-intensive work in the early stages of filmmaking, and do it as soon as possible. We believe that our tool and approach have a good potential for wide dissemination in the film industry. Fri, 01 Oct 2021 00:00:00 GMT https://pubs.kist.re.kr/handle/201004/113582 2021-10-01T00:00:00Z https://pubs.kist.re.kr/handle/201004/113581 Title: An approach for assessing arousal characteristics of ASMR using electroencephalographic power Authors: Koo, Bonseok; Kim, Laehyun; Ha, Jihyeon; Kim, Da-Hye Abstract: Autonomous sensory meridian response (ASMR) is a theory that the brain feels a sense of psychological stability by stimulating the five senses. Existing ASMR studies mainly use subjective evaluation such as questionnaires having potential biases. Therefore, objective evaluation needs to be made. To increase objectivity, we determined a biomarker based on electroencephalography. This study found that the characteristics of ASMR depend on arousal. We detected differences in delta waves in the frontal and central areas and gamma waves in the occipital region. This result means that arousal characteristics can be applied to the ASMR biomarker. We anticipate that our study will improve our understanding of objective ASMR characteristics. Mon, 01 Feb 2021 00:00:00 GMT https://pubs.kist.re.kr/handle/201004/113581 2021-02-01T00:00:00Z