A5070655070- Privacy-Preserving Technologies in Data
- Domain Adaptation and Few-Shot Learning
- Image Processing Techniques and Applications
- Advanced Fiber Optic Sensors
- Topic Modeling
- Advanced Sensor and Energy Harvesting Materials
- Cell Image Analysis Techniques
- Multimodal Machine Learning Applications
- Natural Language Processing Techniques
- Electrospun Nanofibers in Biomedical Applications
- Wound Healing and Treatments
- Photoacoustic and Ultrasonic Imaging
- Face recognition and analysis
- Image Enhancement Techniques
- Single-cell and spatial transcriptomics
- Machine Learning and ELM
- Non-Invasive Vital Sign Monitoring
- Silk-based biomaterials and applications
- Machine Learning in Healthcare
- Generative Adversarial Networks and Image Synthesis
- Balance, Gait, and Falls Prevention
- Effects of Vibration on Health
- Emotion and Mood Recognition
- Anomaly Detection Techniques and Applications
- Software Testing and Debugging Techniques
Korea Advanced Institute of Science and Technology
2019-2024
Kootenay Association for Science & Technology
2023
Hankyong National University
2021
Kyung Hee University Medical Center
2011
Abstract Individuals who are unable to walk independently spend most of the day in a wheelchair. This population is at high risk for developing pressure injuries caused by sitting. However, early diagnosis and prevention these still remain challenging. Herein, we introduce battery-free, wireless, multimodal sensors movable system continuous measurement pressure, temperature, hydration skin interfaces. The device design includes crack-activated sensor with nanoscale encapsulations enhanced...
Contrastive learning has shown remarkable results in recent self-supervised approaches for visual representation. By to contrast positive pairs' representation from the corresponding negatives pairs, one can train good representations without human annotations. This paper proposes Mix-up Contrast (MixCo), which extends contrastive concept semi-positives encoded mix-up of and negative images. MixCo aims learn relative similarity representations, reflecting how much mixed images have original...
In federated learning, a strong global model is collaboratively learned by aggregating clients' locally trained models. Although this precludes the need to access data directly, model's convergence often suffers from heterogeneity. This study starts an analogy continual learning and suggests that forgetting could be bottleneck of learning. We observe forgets knowledge previous rounds, local training induces outside distribution. Based on our findings, we hypothesize tackling down will...
Abstract Considerable efforts have been devoted to developing wound dressings with various functions, including rapid cell proliferation, protection against infection, and state monitoring minimize severe pain the risks of wound‐caused secondary infections. However, it remains challenging diagnose conditions achieve integration above functions without specialized equipment expertise in care. This study describes an electrospun composite micro/nanofiber‐based bilayer‐dressing patch comprising...
Recently, artificial intelligence has been successfully used in fields, such as computer vision, voice, and big data analysis. However, various problems, security, privacy, ethics, also occur owing to the development of intelligence. One problem are deepfakes. Deepfake is a compound word for deep learning fake. It refers fake video created using technology or production process itself. Deepfakes can be exploited political abuse, pornography, information. This paper proposes method determine...
Cell segmentation is a critical step for quantitative single-cell analysis in microscopy images. Existing cell methods are often tailored to specific modalities or require manual interventions specify hyper-parameters different experimental settings. Here, we present multi-modality benchmark, comprising over 1500 labeled images derived from more than 50 diverse biological experiments. The top participants developed Transformer-based deep-learning algorithm that not only exceeds existing but...
Contrastive loss has significantly improved performance in supervised classification tasks by using a multi-viewed framework that leverages augmentation and label information. The enables contrast with another view of single image but enlarges training time memory usage. To exploit the strength multi-views while avoiding high computation cost, we introduce multi-exit architecture outputs multiple features single-viewed framework. this end, propose Self-Contrastive (SelfCon) learning, which...
Cell segmentation is a fundamental task for computational biology analysis. Identifying the cell instances often first step in various downstream biomedical studies. However, many algorithms, including recently emerging deep learning-based methods, still show limited generality under multi-modality environment. Weakly Supervised Segmentation Multi-modality High-Resolution Microscopy Images was hosted at NeurIPS 2022 to tackle this problem. We propose MEDIAR, holistic pipeline instance...
As the damage caused by recent climate crisis increases, efforts are being made to develop low-power and high-efficiency technologies reduce pollution for energy production worldwide. Among them, research on mechano-responsive optical transmittance modulation technology is actively conducted as it can be applied various application fields reducing consumption: sensors smart windows. The piezo-transmittance structure, which one of structures, has fewer constraints installation environment;...
Over the years, open education in online environments, such as Massive Online Open Courses, has grown rapidly. While trend is expected to bridge educational gap among students, new environment also created challenges lack of feedback and difficulties interaction. The authors propose an automated engagement recognition system alleviate this problem, driven by recent developments computer vision artificial neural networks. authors' proposed extracts deep features from a facial image employs...
While learning to align Large Language Models (LLMs) with human preferences has shown remarkable success, aligning these models meet the diverse user presents further challenges in preserving previous knowledge. This paper examines impact of personalized preference optimization on LLMs, revealing that extent knowledge loss varies significantly heterogeneity. Although approaches have utilized KL constraint between reference model and policy model, we observe they fail maintain general...
In federated learning, the local devices train model with their data, independently; and server gathers locally trained to aggregate them into a shared global model. Therefore, learning is an approach decouple training from directly assessing data. However, requirement of periodic communications on parameters results in primary bottleneck for efficiency learning. This work proposes novel algorithm, Federated Weight Recovery(FEWER), which enables sparsely pruned phase. FEWER starts initial...
Purpose : This study examines the effect of vibration exercise grafting PNF patterns for 6 weeks on upper body stability and equilibrium seniors having fifteen or over MMSE-K. Method A total 10 senior citizens participated in this study. Each participant performed patterned exercises using sports equipment 30 minutes, once static a week, six weeks. We measured trunk balance degree before after six-week program. Motor Assessment Scale (MAS) was used to measure stability, while Functional...
With the success of deep learning in various fields and advent numerous Internet Things (IoT) devices, it is essential to lighten models suitable for low-power devices. In keeping with this trend, MicroNet Challenge, which challenge build efficient from view both storage computation, was hosted at NeurIPS 2019. To develop through challenge, we propose a framework, coined as SIPA, consisting four stages: Searching, Improving, Pruning, Accelerating. proposed our team, OSI AI, compressed 334x...
Abstract Allergic diseases, including asthma, are defined as chronic inflammatory disorders originating from an aberrant immune response to innocuous antigens. Current therapies effective in crosssections of patients, but remain mostly palliative. Given the capacity regulatory T cells (Tregs) suppress disease-promoting responses, enhancing Treg function represents attractive therapeutic strategy for treatment allergic disease. Among complementary approaches acupuncture has been widely used...
인간의 지각은 청각-시각 정보를 연관 지어 청각 정보로부터 시각 연상할 수 있고 그 역도 가능한다. 이러한 능력은 정보가 관련되어 있는 상황을 경험하며 자연스럽게 획득할 있지만, 두 유형의 충분히 결합된 영상 데이터는 각 장면의 맥락에 따라 가지 레이블을 동시에 만들어주어야 하므로 데이터셋을 만들기 어렵다. 본 논문에서는 같은 카테고리에 대해 한 유형에 대한 임베딩에서 다른 유형으로 변환(mapping)할 Contrastive Embedding Mapper (CoEM)을 제안한다. 쌍으로 짝지을 필요 없이 CoEM은 변환된 임베딩을 대조하는 방식으로 학습한다. 우리는 청각과 데이터셋에 CoEM의 효력을 확인하기 위해 20가지의 실험했다. 실험에서 CoEM에 의해 변환되어 연결된 임베딩들은 도메인에서의 검색 성능의 경우 이웃하는 기준점이 충분한 경우(20개) 약 90%의 성능을 보였다. 또한, 데이터 재 생성이 가능함을 확인했다.
Multifunctional Wound Dressings In article number 2201765, Ji-Hwan Ha, Jun-Ho Jeong, Inkyu Park, and colleagues present a multifunctional micro/nanofiber-based dressing patch containing natural materials such as hyaluronic acid, gelatin, dexpanthenol, curcumin for advanced wound care. The fabricated by an electrospinning process enables rapid healing, protection against external microbial, status monitoring colorimetric pH sensing.
Federated Learning (FL) aggregates locally trained models from individual clients to construct a global model. While FL enables learning model with data privacy, it often suffers significant performance degradation when have heterogeneous distributions. This heterogeneity causes the forget knowledge acquired previously sampled after being on local datasets. Although introduction of proximal objectives in updates helps preserve knowledge, can also hinder by interfering objectives. To address...