A5060124284- Topic Modeling
- Adversarial Robustness in Machine Learning
- Natural Language Processing Techniques
- Advanced Neural Network Applications
- Cell Adhesion Molecules Research
- Epigenetics and DNA Methylation
- Occupational exposure and asthma
- Hate Speech and Cyberbullying Detection
- Allergic Rhinitis and Sensitization
- Melanoma and MAPK Pathways
- Autophagy in Disease and Therapy
- Cytokine Signaling Pathways and Interactions
- Sirtuins and Resveratrol in Medicine
- Adipose Tissue and Metabolism
- Bone Metabolism and Diseases
- Machine Learning in Healthcare
- Education and Learning Interventions
- Viral gastroenteritis research and epidemiology
- Calcium signaling and nucleotide metabolism
- Hippo pathway signaling and YAP/TAZ
- Cholangiocarcinoma and Gallbladder Cancer Studies
- Asthma and respiratory diseases
- Animal Disease Management and Epidemiology
- Immune cells in cancer
- Plant Molecular Biology Research
Seoul National University
2024
Yonsei University
2002-2023
Ajou University
2019
Inje University Ilsan Paik Hospital
2014
National Institutes of Health
2009-2011
National Heart Lung and Blood Institute
2009
Inha University
2007-2009
Zero to Three
2005
National Cancer Center
2005
Genome Research Foundation
2000
SIRT1, an NAD+ (nicotinamide adenine dinucleotide)-dependent deacetylase, protects cells from stress-induced apoptosis, and its orthologues delay aging in lower eukaryotes. SIRT1 increases survival response to stress such as DNA damage by deacetylating a number of substrates including pro-apoptotic protein p53. The molecular mechanism which DNA-damage activates is not known. By screening kinase inhibitor library, we identified CK2 kinase. pleiotropic with more than 300 well-known...
In training of modern large natural language processing (NLP) models, it has become a common practice to split models using 3D parallelism multiple GPUs. Such technique, however, suffers from high overhead inter-node communication. Compressing the communication is one way mitigate by reducing traffic volume; existing compression techniques have critical limitations be applied for NLP with in that 1) only data targeted, and 2) schemes already harm model quality too much.
The growing memory and computational demands of large language models (LLMs) for generative inference present significant challenges practical deployment. One promising solution to address these is offloading-based batched inference, which leverages host disk as an extended hierarchy GPUs. While the approach cost-effectively enables LLM its performance limited by substantial I/O overhead, primarily due key-value (KV) cache sizes, increase with batch size context window length. In this paper,...
Rho-like GTPases, including Cdc42, Rac1 and RhoA, regulate distinct actin cytoskeleton changes required for cell adhesion, migration invasion. In the present study, we examined role of Rac signaling in inherent migration, as well radiation-induced rat glioma cells. Stable overexpression dominant-negative Rac1N17 a C6 line (C6-RacN17) promoted ionizing radiation further increased this migration. Migration was accompanied by decreased expression focal adhesion molecules FAK paxillin. Focal...
The molecular mechanism underlying the initiation of somatic cell reprogramming into induced pluripotent stem cells (iPSCs) has not been well described. Thus, we generated single-cell-derived clones by using a combination drug-inducible vectors encoding transcription factors (Oct4, Sox2, Klf4 and Myc) single-cell expansion strategy. This system achieved high efficiency after metabolic epigenetic remodeling. Functional analyses cloned revealed that extracellular signal-regulated kinase (ERK)...
The circadian clock organizes the physiology and behavior of organisms to their daily environmental rhythms. central timekeeping mechanism in eukaryotic cells is transcriptional-translational feedback loop (TTFL). In <i>Drosophila</i> TTFL, transcription factors CLOCK (CLK) CYCLE (CYC) play crucial roles activating expression core genes clock-controlled genes. Many signaling pathways converge on CLK/CYC complex regulate its activity fine-tune cellular oscillator time cues. We aimed identify...
We examined genomic DNA from each of three human-derived gastric cancer cell lines, using the technique restriction landmark scanning (RLGS) which allows monitoring approximately 2, 000 NotI landmarks. The resulting spots were compared with those in normal mucosa or primary tumor. In all, 9 intense detected two lines. Two highly intensified common lines and proven to be originated region containing human c-myc proto-oncogene on chromosome 8. degree amplification was similar other estimated...
Training large language models (LLMs) is known to be challenging because of the huge computational and memory capacity requirements. To address these issues, it common use a cluster GPUs with 3D parallelism, which splits model along data batch, pipeline stage, intra-layer tensor dimensions. However, parallelism produces additional challenge finding optimal number ways on each dimension mapping split onto GPUs. Several previous studies have attempted automatically find configuration, but many...
Despite the impressive capabilities of Large Language Models (LLMs) in various tasks, their vulnerability to unsafe prompts remains a critical issue. These can lead LLMs generate responses on illegal or sensitive topics, posing significant threat safe and ethical use. Existing approaches attempt address this issue using classification models, but they have several drawbacks. With increasing complexity prompts, similarity search-based techniques that identify specific features provide more...
The recent huge advance of Large Language Models (LLMs) is mainly driven by the increase in number parameters. This has led to substantial memory capacity requirements, necessitating use dozens GPUs just meet capacity. One popular solution this storage-offloaded training, which uses host and storage as an extended hierarchy. However, obviously comes at cost bandwidth bottleneck because devices have orders magnitude lower compared that GPU device memories. Our work, Smart-Infinity, addresses...
Adversarial robustness of the neural network is a significant concern when it applied to security-critical domains. In this situation, adversarial distillation promising option which aims distill teacher improve small student network. Previous works pretrain make robust examples aimed at itself. However, are dependent on parameters target The fixed inevitably degrades its against unseen transferred targets in process. We propose PeerAiD peer learn instead an that trains and simultaneously...
Training large deep neural network models is highly challenging due to their tremendous computational and mem-ory requirements. Blockwise distillation provides one promising method towards faster convergence by splitting a model into multiple smaller models. In state-of-the-art blockwise methods, training performed block-by-block in data-parallel manner using GPUs. To produce inputs for the student blocks, teacher executed from beginning until current block under training. However, this...
The nitrogen cycle is a biogeochemical primarily associated with the microbial activity that occurs in various environments, including soil. Various genes related to have been studied for different purposes by many researchers; therefore, polymerase chain reaction (PCR) conditions and gene compositions differ among reports, making comparisons difficult. In this study, we compare PCR methods amplify 13 cycle-related (amo (amoA amoB), norB (cnorB qnorB), hzs, napA, narG, nifH, nirK, nirS,...