A5063999144- Natural Language Processing Techniques
- Biomedical Text Mining and Ontologies
- Topic Modeling
- Handwritten Text Recognition Techniques
- Domain Adaptation and Few-Shot Learning
- Multimodal Machine Learning Applications
- Advanced Neural Network Applications
- Image Retrieval and Classification Techniques
- Image Processing and 3D Reconstruction
- Bone Tissue Engineering Materials
- Semantic Web and Ontologies
- Dental materials and restorations
- Endodontics and Root Canal Treatments
- Advanced Image and Video Retrieval Techniques
- Visual Attention and Saliency Detection
- Dental Implant Techniques and Outcomes
- Bioinformatics and Genomic Networks
- Additive Manufacturing Materials and Processes
- Robotics and Sensor-Based Localization
- Dental Radiography and Imaging
- Metal and Thin Film Mechanics
- COVID-19 diagnosis using AI
- Sparse and Compressive Sensing Techniques
- Speech and dialogue systems
- Titanium Alloys Microstructure and Properties
Naver (South Korea)
2021
Showa University
2012-2014
The University of Tokyo
2009-2013
Tokyo University of Science
2012
Korea University
2008-2009
We report the Gene Normalization (GN) challenge in BioCreative III where participating teams were asked to return a ranked list of identifiers genes detected full-text articles. For training, 32 fully and 500 partially annotated articles prepared. A total 507 selected as test set. Due high annotation cost, it was not feasible obtain gold-standard human annotations for all Instead, we developed an Expectation Maximization (EM) algorithm approach choosing small number manual that most capable...
Abstract Motivation: Event extraction using expressive structured representations has been a significant focus of recent efforts in biomedical information extraction. However, event resources and methods have so far focused almost exclusively on molecular-level entities processes, limiting their applicability. Results: We extend the approach to encompass all levels biological organization from molecular whole organism. present ontological foundations, target types guidelines for entity...
Domain generalization (DG) methods aim to achieve generalizability an unseen target domain by using only training data from the source domains. Although a variety of DG have been proposed, recent study shows that under fair evaluation protocol, called DomainBed, simple empirical risk minimization (ERM) approach works comparable or even outperforms previous methods. Unfortunately, simply solving ERM on complex, non-convex loss function can easily lead sub-optimal seeking sharp minima. In this...
Mineral trioxide aggregate (MTA) cement is an attractive material in endodontic dentistry. The purpose of this study was to produce calcium silicate, which a major component MTA, from waste materials. A dental alginate impression gel and used chalks were selected mixed suitable ratio (Code: EXP). As control, CaCO3 commercial diatomite CON). Each powder heated 850。C 1000。C, then kneaded with water. TG-DTA, compressive tests, SEM observations, elemental mapping analyses, XRD analyses...
Recent open-vocabulary detection methods aim to detect novel objects by distilling knowledge from vision-language models (VLMs) trained on a vast amount of image-text pairs. To improve the effectiveness these methods, researchers have utilized datasets with large vocabulary that contains number object classes, under assumption such data will enable extract comprehensive relationships between various and better generalize unseen classes. In this study, we argue more fine-grained labels are...
Recent end-to-end scene text spotters have achieved great improvement in recognizing arbitrary-shaped instances. Common approaches for spotting use region of interest pooling or segmentation masks to restrict features single However, this makes it hard the recognizer decode correct sequences when detection is not accurate i.e. one more characters are cropped out. Considering that accurately decide word boundaries with only detector, we propose a novel Detection-agnostic End-to-End...
In this paper, we propose a system for biomedical event extraction using multi-phase approach. It consists of trigger detector, type classifier, and relation recognizer compositor. The firstly identifies triggers in given sentence. Then, it classifies the into one nine predefined classes. Lastly, examines each whether has with participant candidates, composites events extracted relations. official score proposed recorded 61.65 precision, 9.40 recall 16.31 f-score approximate span matching....
Calcium phosphate is known as a major component of biological hard tissues. This study aimed to produce calcium by recycling kneaded surplus gypsum. β-dihydrate gypsum was derived from commercial dental β-hemihydrate gypsum, which mechanically powdered and mixed with the liquid zinc cement. mixture fired at 1,200°C evaluated XRD analysis, thermal analysis scanning electron microscopy (SEM). An acceptable ratio mixing 4 g powder 1.5 mL phosphoric acid liquid. peaks were monotonic below 800°C,...
This study aimed to develop a dental investment for titanium casting. ZrO2 and Al2O3 were selected as refractory materials prepare three investments (Codes: A-C) according the quantity of Zr. cement was used binder at ratio 15%, they mixed with special mixing liquid. B1 control water. Fundamental examinations statistically evaluated. A casting test performed B. Fluidities, setting times, green strengths showed no remarkable differences; however, significantly different from those B1....
In this paper, we describe a syntactic and semantic dependency parsing system submitted to the shared task of CoNLL 2008. The proposed consists five modules: parser, predicate identifier, local role labeler, global sequence candidate generator, selector. parser is based on Malt Parser generator CKY style algorithm. remaining three modules are implemented by using maximum entropy classifiers. achieves 76.90 labeled F1 for overall task, 84.82 attachment, 68.71 WSJ+Brown test set.
Recent advances in deep learning have enabled complex real-world use cases comprised of multiple vision tasks and detection are being shifted to the edge side as a pre-processing step entire workload. Since running model on resource-constraint devices is challenging, techniques for efficient inference methods demanded. In this paper, we present an objectness-aware object method reduce computational cost by sparsifying activation values background regions where target objects don't exist....
Most NLP applications work under the assumption that a user input is error-free; thus, word segmentation (WS) for written languages use boundary markers (WBMs), such as spaces, has been regarded trivial issue. However, noisy real-world texts, blogs, e-mails, and SMS, may contain spacing errors require correction before further processing take place. For Korean language, many researchers have adopted traditional WS approach, which eliminates all spaces in re-inserts proper boundaries....
Recently, self-supervised methods show remarkable achievements in image-level representation learning. Nevertheless, their self-supervisions lead the learned to sub-optimal for dense prediction tasks, such as object detection, instance segmentation, etc. To tackle this issue, several recent learning have extended single embedding pixel-level embeddings. Unlike learning, due spatial deformation of augmentation, it is difficult sample positive pairs. Previous studies sampled pairs using...
For successful scene text recognition (STR) models, synthetic image generators have alleviated the lack of annotated images from real world. Specifically, they generate multiple with diverse backgrounds, font styles, and shapes enable STR models to learn visual patterns that might not be accessible manually data. In this paper, we introduce a new generator, SynthTIGER, by analyzing techniques used for synthesis integrating effective ones under single algorithm. Moreover, propose two...