A5100430150- Bioinformatics and Genomic Networks
- Gene expression and cancer classification
- Advanced Neural Network Applications
- Advanced Steganography and Watermarking Techniques
- Algorithms and Data Compression
- Video Surveillance and Tracking Methods
- Adversarial Robustness in Machine Learning
- Chaos-based Image/Signal Encryption
- Advanced Image and Video Retrieval Techniques
- Metabolomics and Mass Spectrometry Studies
- Auditing, Earnings Management, Governance
- Data Management and Algorithms
- Natural Language Processing Techniques
- Advanced Malware Detection Techniques
- Domain Adaptation and Few-Shot Learning
- User Authentication and Security Systems
- Fire Detection and Safety Systems
- Corporate Finance and Governance
- Molecular Biology Techniques and Applications
- COVID-19 diagnosis using AI
- Anomaly Detection Techniques and Applications
- Advanced Graph Neural Networks
- Genetic Associations and Epidemiology
- Graph Theory and Algorithms
- Image Retrieval and Classification Techniques
Konkuk University
2018-2025
Massachusetts General Hospital
2025
Harvard University
2025
Ca' Foscari University of Venice
2024
Agency for Defense Development
2023
Pusan National University
2011-2019
Pohang University of Science and Technology
2013-2018
Keimyung University
2017-2018
Korea Post
2013-2018
Korea University
2016-2017
Of late, high-throughput microarray and sequencing data have been extensively used to monitor biomarkers biological processes related many diseases. Under this circumstance, the support vector machine (SVM) has popularly successful for gene selection in applications. Despite surpassing benefits of SVMs, single analysis using small- mid-size inevitably runs into problem low reproducibility statistical power. To address problem, we propose a meta-analytic (Meta-SVM) that can accommodate...
Abstract Motivation With the prevalent usage of microarray and massively parallel sequencing, numerous high-throughput omics datasets have become available in public domain. Integrating abundant information among is critical to elucidate biological mechanisms. Due high-dimensional nature data, methods such as principal component analysis (PCA) been widely applied, aiming at effective dimension reduction exploratory visualization. Results In this article, we combine multiple identical or...
In this study, we propose an automated system for measuring the size of strawberries and predicting their weight using AI technology. The combines computer vision techniques with LiDAR sensor data to accurately estimate dimensions infer weight. By integrating deep learning models, such as HRNet keypoint detection, leveraging capabilities sensors, minimize human intervention achieve precise measurement. relative errors width height are 3.71% 5.42%, respectively, exhibiting a lower error rate....
Despite achieving impressive results in general-purpose semantic segmentation with strong generalization on natural images, the Segment Anything Model (SAM) has shown less precision and stability medical image segmentation. In particular, SAM architecture is designed for 2D images therefore not support to handle three-dimensional information, which particularly important imaging modalities that are often volumetric or video data. this paper, we introduce MediViSTA, a parameter-efficient...
In this study, we propose a solution for automatically measuring body circumferences by utilizing the built-in LiDAR sensor in mobile devices. Traditional measurement methods mainly rely on 2D images or manual measurements. This research, however, utilizes 3D depth information to enhance both accuracy and efficiency. By employing HRNet-based keypoint detection transfer learning through deep learning, precise locations of parts are identified combined with maps calculate circumferences....
Deep Neural Networks (DNNs)-based semantic segmentation models trained on a source domain often struggle to generalize unseen target domains, i.e., gap problem. Texture contributes the gap, making DNNs vulnerable shift because they are prone be texture-biased. Existing Domain Generalized Semantic Segmentation (DGSS) methods have alleviated problem by guiding prioritize shape over texture. On other hand, and texture two prominent complementary cues in segmentation. This paper argues that...
Abstract Motivation: Supervised machine learning is widely applied to transcriptomic data predict disease diagnosis, prognosis or survival. Robust and interpretable classifiers with high accuracy are usually favored for their clinical translational potential. The top scoring pair (TSP) algorithm an example that applies a simple rank-based identify rank-altered gene pairs classifier construction. Although many classification methods perform well in cross-validation of single expression...
With the rapid advances in technologies of microarray and massively parallel sequencing, data multiple omics sources from a large patient cohort are now frequently seen many consortium studies. Effective multi-level integration has brought new statistical challenges. One important biological objective such integrative analysis is to cluster patients order identify clinically relevant disease subtypes, which will form basis for tailored treatment personalized medicine. Several methods have...
To date, the support vector machine (SVM) has been widely applied to diverse bio-medical fields address disease subtype identification and pathogenicity of genetic variants. In this paper, I propose weighted K-means (wKM-SVM) (wSVM), for which allow SVM impose weights loss term. Besides, demonstrate numerical relations between objective function weights. Motivated by general ensemble techniques, are known improve accuracy, directly adopt boosting algorithm newly proposed KM-SVM (and wSVM)....
The rapid advances of omics technologies have generated abundant genomic data in public repositories and effective analytical approaches are critical to fully decipher biological knowledge inside these data. Meta-analysis combines multiple studies a related hypothesis improve statistical power, accuracy reproducibility beyond individual study analysis. To date, many transcriptomic meta-analysis methods been developed, yet few thoughtful guidelines exist. Here, we introduce comprehensive...
We initiate the study of sub-linear sketching and streaming techniques for estimating output size common dictionary compressors such as Lempel-Ziv '77, run-length Burrows-Wheeler transform, grammar compression. To this end, we focus on a measure that has recently gained much attention in information-theoretic community which approximates up to polylogarithmic multiplicative factor sizes those compressors: normalized substring complexity function δ. As matter fact, δ itself is very accurate...
In the recent era of AI, instance segmentation has significantly advanced boundary and object detection especially in diverse fields (e.g., biological environmental research). Despite its progress, edge amid adjacent objects organism cells) still remains intractable. This is because homogeneous heterogeneous are prone to being mingled a single image. To cope with this challenge, we propose weighted Mask R‐CNN designed effectively separate overlapped virtue extra weights boundaries. For...
In this paper, we propose a robust and reliable face recognition model that incorporates depth information such as data from point clouds maps into RGB image to avoid false facial verification caused by spoofing attacks while increasing the model’s performance. The proposed is driven spatially adaptive convolution (SAC) block of SqueezeSegv3; attention enables weight features according their importance spatial location. We also utilize large-margin loss instead softmax supervision signal for...
To compare and analyze the differences in sociodemographic clinical characteristics of suicide attempters who visited an emergency department (ED) before during coronavirus disease (COVID-19) pandemic.This single center, retrospective study was conducted by reviewing medical records patients "self-injury/suicide" category National Emergency Department Information System ED between January 2019 December 2020. We obtained information on baseline characteristics, attempt, disposition. Data were...
Face anti-spoofing (FAS) is a technology that protects face recognition systems from presentation attacks. The current challenge faced by FAS studies the difficulty in creating generalized light variation model. This because data are sensitive to domain. models using only red green blue (RGB) images suffer poor performance when training and test datasets have different variations. To overcome this problem, study focuses on detection ranging (LiDAR) sensors. LiDAR time-of-flight depth sensor...
Deep neural networks yield desirable performance in text, image, and speech classification. However, these are vulnerable to adversarial examples. An example is a sample generated by inserting small amount of noise into an original (with minimal distortion) such that it recognized incorrectly the targeted model. A typical method attack using examples must satisfy two conditions: distortion be kept minimum misrecognition induced deep network. Therefore, considerable time numerous iterations...
Predictive models on fire have been increasingly popular in computer image analysis. Due to late strides of deep learning techniques, we are now unprecedently benefited from its flexible applicability. In most cases, however, the conventional algorithms limited only single-framed images unlike sequence data that inevitably entails heavy computational time and memory. this paper, propose an effective algorithm exploiting combination CNNs (convolution neural networks) RNNs (recurrent a...
Maximum likelihood estimation is used widely in classical statistics. However, except a few cases, it does not have closed form. Furthermore, takes time to derive the maximum estimator (MLE) owing use of iterative methods such as Newton–Raphson. Nonetheless, this method has several advantages, chief among them being invariance property and asymptotic normality. Based on first approximation solution equation, we obtain an that same behavior MLE for multivariate gamma distribution. The newly...
As a large amount of genetic data are accumulated, an effective analytical method and significant interpretation required. Recently, various methods machine learning have emerged to process data. In addition, analysis tools using statistical models been proposed. this study, we propose adding integrated layer the deep structure, which would enable discovery biomarkers diseases. We conducted simulation study in order compare proposed with metalogistic regression meta-SVM methods. The...
This paper studies the optimization of list intersection, especially in context matching phase search engines. Given a user query, we intersect postings lists corresponding to query keywords generate documents all keywords. Since speed intersection depends algorithm, hardware, and lengths their correlations, none existing algorithms outperforms others every scenario. Therefore, develop cost-based approach which identify space, spanning combinations. We propose cost model estimate with...
The Convolutional Neural Network (CNN) model, often used for image classification, requires significant training time to obtain high accuracy. To this end, distributed is performed with the parameter server (PS) architecture using multiple servers. Unfortunately, scalability has been found be poor in existing architectures. We find that PS network bottleneck as it communicates a large number of gradients and parameters many workers. This because synchronization workers occur at every step...
Automatic garment size measurement approaches using computer vision algorithms have been attempted in various ways, but there are still many limitations to overcome. One limitation is that the process involves 2D images, which results constraints of determining actual distance between estimated points. To solve this problem, paper, we propose an automated method for measuring sizes deep learning models and point cloud data. In proposed method, a learning-based keypoint estimation model first...