A5100757399- Parallel Computing and Optimization Techniques
- Advanced Data Storage Technologies
- Digital Media Forensic Detection
- Embedded Systems Design Techniques
- Generative Adversarial Networks and Image Synthesis
- Multimodal Machine Learning Applications
- Advanced Neural Network Applications
- Human Pose and Action Recognition
- AI in cancer detection
- Image Retrieval and Classification Techniques
- Video Analysis and Summarization
- Semiconductor Lasers and Optical Devices
- Advanced MEMS and NEMS Technologies
- Cell Image Analysis Techniques
- Data Stream Mining Techniques
- Distributed systems and fault tolerance
- Education and Learning Interventions
- Advanced Image Processing Techniques
- Asian Culture and Media Studies
- Anomaly Detection Techniques and Applications
- Robotic Path Planning Algorithms
- Embedded Systems and FPGA Design
- Face recognition and analysis
- Algorithms and Data Compression
- Robot Manipulation and Learning
Korea University
2020-2021
Samsung (South Korea)
2003-2013
University of Ottawa
2005
Software (Spain)
2004
Abstract Despite decades of intensive search for compounds that modulate the activity particular protein targets, a large proportion human kinome remains as yet undrugged. Effective approaches are therefore required to map massive space unexplored compound–kinase interactions novel and potent activities. Here, we carry out crowdsourced benchmarking predictive algorithms kinase inhibitor potencies across multiple families tested on unpublished bioactivity data. We find top-performing...
Human-Object Interaction (HOI) detection is the task of identifying a set (human, object, interaction) triplets from an image. Recent work proposed transformer encoder-decoder architectures that successfully eliminated need for many hand-designed components in HOI through end-to-end training. However, they are limited to single-scale feature resolution, providing suboptimal performance scenes containing humans, objects, and their interactions with vastly different scales distances. To tackle...
Vision Transformers (ViTs) have achieved remarkable success in various computer vision tasks. However, ViTs a huge computational cost due to their inherent reliance on multi-head self-attention (MHSA), prompting efforts accelerate for practical applications. To this end, recent works aim reduce the number of tokens, mainly focusing how effectively prune or merge them. Nevertheless, since ViT tokens are generated from non-overlapping grid patches, they usually do not convey sufficient...
NAND flash memory has become an indispensable component in mobile embedded systems because of its versatile features such as non-volatility, solid-state reliability, low cost and high density. Even though is gaining popularity data storage, it can be also exploited code for XIP (execute-in-place). In this paper, we present a new architecture which incorporates into existing hierarchy execution. The usefulness the proposed approach demonstrated with real workloads on prototyping board.
NAND flash memory has become an indispensable component in mobile embedded systems because of its versatile features such as non-volatility, solid-state reliability, low cost and high density. Even though is gaining popularity data storage, it can be also exploited code for XIP (execute-in-place). In this paper, we present a new architecture which incorporates into existing hierarchy execution. The usefulness the proposed approach demonstrated with real workloads on prototyping board.
Recent advances in vision language pretraining (VLP) have been largely attributed to the large-scale data collected from web. However, uncurated dataset contains weakly correlated image-text pairs, causing inefficiency. To address issue, knowledge distillation explored at expense of extra image and text momentum encoders generate teaching signals for misaligned pairs. In this paper, our goal is resolve misalignment problem with an efficient framework. end, we propose ECLIPSE: Expediting...
As the value of Bitcoin increases, difficulty level mining keeps increasing. This is generally addressed with application-specific integrated circuits (ASIC), but block candidates are still created by software. The overhead candidate generation relatively growing because hash computation boosted ASIC. Additionally, it getting harder to find target nonce; If not found for a candidate, new must be generated. A can generated reduce modifying coinbase without selecting and verifying transactions...
Abstract In this paper, it is presented a capacitive touch sensor IC with noise‐based hybrid sensing scheme, which provides two modes noise detector. noiseless environment, fast power‐efficient peak‐detection mode enabled and if detected, the switched to high‐SNR demodulation mode. Therefore, adaptively offers both speed high immunity without large power consumption. The proposed shows higher than 50‐dB SNR in over 240‐Hz reporting rate 2.8‐mW analog consumption, evaluated 4‐inch AMOLED...
In this paper, we first present the character texture generation system \textit{Minecraft-ify}, specified to Minecraft video game toward in-game application. Ours can generate face-focused image for mapping tailored 3D virtual having cube manifold. While existing projects or works only texture, proposed inverse user-provided real image, average/random appearance from learned distribution. Moreover, it be manipulated with text-guidance using StyleGAN and StyleCLIP. These features provide a...
Large-scale Text-to-Image (TTI) models have become a common approach for generating training data in various generative fields. However, visual hallucinations, which contain perceptually critical defects, remain concern, especially non-photorealistic styles like cartoon characters. We propose novel hallucination detection system character images generated by TTI models. Our leverages pose-aware in-context learning (PA-ICVL) with Vision-Language Models (VLMs), utilizing both RGB and pose...
Vision Transformers (ViTs) have achieved remarkable success in various computer vision tasks. However, ViTs a huge computational cost due to their inherent reliance on multi-head self-attention (MHSA), prompting efforts accelerate for practical applications. To this end, recent works aim reduce the number of tokens, mainly focusing how effectively prune or merge them. Nevertheless, since ViT tokens are generated from non-overlapping grid patches, they usually do not convey sufficient...
NAND flash memory has become an indispensable component in mobile embedded systems because of its versatile features such as non-volatility, solid-state reliability, low cost and high density. Even though is gaining popularity data storage, it can be also exploited code for XIP (execute-in-place). In this paper, we present a new architecture which incorporates into existing hierarchy execution. The usefulness the proposed approach demonstrated with real workloads on prototyping board.
전 세계 암 발병의 큰 비중을 차지하는 폐암을 조기에 예방하기 위해서는 폐 결절을 찾아내 악성 여부를 검사해야 한다. 본 연구에서는 삼차원 시층 콘볼루션 신경망을 이용해 결절의 판단하는 모델을 제안한다. 숏컷 연결을 이용한 사용했고, 분류 성능 향상을 위해 앙상블 기법을 이용한다. LUng Nodule Analysis 2016 대회 데이터에 적용하여 모델의 성능을 측정하고 정확도를 검증한다. 모델은 대회의 평가 지표인 Competition Performance Metric 기준 0.899를 기록하였고, 이는 기존 참가자들의 성능과 비교하였을 때 우수한 결과이다.
본 논문은 1961년 5 · 16 쿠데타 이후 1972년 10월 유신 이전까지 기간에 초점을 맞춰 신문 기사 분석을 중심으로 한국 사회가 재일교포를 어떻게 표상하고 인식해 왔는지 살펴보는 것을 목적으로 한다. 구체적으로는 5월 17일부터 9월 30일까지『경향신문』,『 동아일보』,『 조선일보』기사 박정희 정권 시기 한편으로는 대한민국 “국민”으로 전제하고 혈연민족주의 프레임을 통해“같은 핏줄을 나눈 우리 교포”로 포섭하는 동시에 다른 문화민족주의, 경제개발주의, 반공주의 통해 문화적으로 “혼혈아” 또는 “일본인”이 되어버린 2세와 3세 재일교포, “한국 경제를 일본에 예속시키는” “매판” “조국을 배반” 하고 북한을 지지하는 조총련계 타자화하고 “우리” 의 범주에서 배제하는 과정을 추적하여 보여주고자 이를 “국민”의 경계 설정이 민족주의, 등 다양한 이념적 요인의 상호작용 속에 포섭하고 방식으로 이루어졌음을
Video face re-aging deals with altering the apparent age of a person to target in videos. This problem is challenging due lack paired video datasets maintaining temporal consistency identity and age. Most methods process each image individually without considering While some existing works address issue coherence through facial attribute manipulation latent space, they often fail deliver satisfactory performance transformation. To tackle issues, we propose (1) novel synthetic dataset that...
Recent advances in vision language pretraining (VLP) have been largely attributed to the large-scale data collected from web. However, uncurated dataset contains weakly correlated image-text pairs, causing inefficiency. To address issue, knowledge distillation explored at expense of extra image and text momentum encoders generate teaching signals for misaligned pairs. In this paper, our goal is resolve misalignment problem with an efficient framework. end, we propose ECLIPSE: Expediting...