• Journal of Information Technology Applications and Management
• /
• v.25 no.4
• /
• pp.145-156
• /
• 2018

This study proposes a content-based recommendation algorithm (NRA) for pet food. The proposed algorithm tries to recommend appropriate or inappropriate feed by using collective intelligence based on user experience and prior knowledge of experts. Based on the physical and health status of the dogs, this study suggests what kind of nutrients are necessary for the dogs and the most recommended pet food containing these nutrients. Performance evaluation was performed in terms of recall, precision, F1 and AUC. As a result of the performance evaluation, the AUC and F1 value of the proposed NRA was 15% and 42% higher than that of the baseline model, respectively. In addition, the performance of NRA is shown higher for recommendation of normal dogs than disease dogs.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.14 no.4
• /
• pp.314-322
• /
• 2021

Skeleton-based action recognition has attracted considerable attention in human action recognition. Recent methods for skeleton-based action recognition employ spatiotemporal graph convolutional networks (GCNs) and have remarkable performance. However, most of them have heavy computational complexity for robust action recognition. To solve this problem, we propose a shuffle graph convolutional network (SGCN) which is a lightweight graph convolutional network using pointwise group convolution rather than pointwise convolution to reduce computational cost. Our SGCN is composed of spatial and temporal GCN. The spatial shuffle GCN contains pointwise group convolution and part shuffle module which enhances local and global information between correlated joints. In addition, the temporal shuffle GCN contains depthwise convolution to maintain a large receptive field. Our model achieves comparable performance with lowest computational cost and exceeds the performance of baseline at 0.3% and 1.2% on NTU RGB+D and NTU RGB+D 120 datasets, respectively.

• Journal of Broadcast Engineering
• /
• v.27 no.7
• /
• pp.1011-1020
• /
• 2022

Recent high-performance panoptic segmentation models are based on transformer architectures. However, transformer-based panoptic segmentation methods are basically slower than convolution-based methods, since the attention mechanism in the transformer requires quadratic complexity w.r.t. image resolution. Also, sine and cosine computation for positional embedding in the transformer also yields a bottleneck for computation time. To address these problems, we adopt three modules to speed up the inference runtime of the transformer-based panoptic segmentation. First, we perform channel-level reduction using depth-wise separable convolution for inputs of the transformer decoder. Second, we replace sine and cosine-based positional encoding with convolution operations, called conv-embedding. We also apply a separable self-attention to the transformer encoder to lower quadratic complexity to linear one for numbers of image pixels. As result, the proposed model achieves 44% faster frame per second than baseline on ADE20K panoptic validation dataset, when we use all three modules.

• The Journal of Korea Robotics Society
• /
• v.18 no.1
• /
• pp.122-126
• /
• 2023

An unexpected occlusion in a real life, not in a laboratory, can be more fatal to neural networks than expected. In addition, it is virtually impossible to create a network that learns all the environmental changes as well as occlusions. Therefore, we propose an alternative approach in which the architecture and number of parameters remain unchanged while adapting to occlusion circumstances. Learning method with the term Conversion Learning classifies them more robustly by converting the weights from various occlusion situations. The experiments on MNIST dataset showed a 3.07 [%p] performance improvement over the baseline CNN model in a situation where most objects are occluded and unknowing what occlusion will appear in advance. The experimental results suggest that Conversion Learning is an efficient method to respond to environmental changes such as occluded images.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.11a
• /
• pp.136-139
• /
• 2020

본 논문에서는 최근 대두되고 있는 심층신경망 압축 연구에서 가중치 공유와 관련하여 심층신경망 모델 압축방법 Inter-Layer Kernel Prediction을 제안한다. 제안 방법은 영상 압축에서 사용되는 프레임 간 prediction 방법을 응용한 컨볼루션 신경망 가중치 공유 및 모델 압축 방법이다. 본 논문은 레이어 간 유사한 kernel들이 존재한다는 것을 발견하고 이를 기반으로 Inter-Layer Kernel Prediction을 사용하여 기존 모델 가중치를 보다 더 적은 비트로 표현하여 저장하는 방법을 제안한다. 제안 방법은 CIFAR10/100으로 학습된 ResNet에서 약 4.1 배의 압축률을 달성했으며 CIFAR10으로 학습된 ResNet110에서는 오히려 기존 Baseline 모델에 비해 0.04%의 성능 향상을 기록했다.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2023.01a
• /
• pp.311-313
• /
• 2023

병렬 기계 작업장에서 자원을 효율적으로 활용하기 위해서는 처리할 작업을 적절한 기계에 할당해야 한다. 특정 작업을 처리할 기계를 선택할 때 휴리스틱을 사용할 수도 있으나, 특정 작업장에 맞춤화된 휴리스틱을 개발하는 것은 쉽지 않다. 반면, 본 논문에서는 이종 병렬 기계 작업장을 위한 작업 할당 모형을 개발하는데 강화학습을 응용하고자 한다. 작업 할당 모형을 학습하는데 필요한 에피소드들은 상용 시뮬레이션 소프트웨어인 FlexSim을 이용하여 생성하였다. 아울러, stable-baseline3 라이브러리를 이용하여 강화학습 알고리즘을 생성된 에피소드들에 적용하였다. 실험 결과를 통해 시뮬레이션과 강화학습이 작업장 운영관리에 유용함을 알 수 있었다.

• Annual Conference of KIPS
• /
• 2024.10a
• /
• pp.565-568
• /
• 2024

Parkinson's disease (PD) is a progressive disorder. In this study, we proposed a deep learning model that utilized participants' baseline clinical features and deformation-based morphometry (DBM) to predict long-term cognitive trajectory over four years. A total of 216 participants from the PPMI (Parkinson's Progression Markers Initiative) dataset were included, with 157 being PD patients and 59 healthy controls. We identified brain connectivity patterns associated with long-term cognitive decline using DBM and independent component analysis (ICA) techniques. Results of the cognitive prediction indicated that using only clinical features, DBM features, and multimodal features yielded average accuracies of 76 ± 4%, 70 ± 6%, and 78 ± 2%, and average AUC (Area Under the Curve) of 0.71 ± 0.06, 0.62 ± 0.04, and 0.76 ± 0.06, respectively. Our study demonstrated that the potential of using DBM features to better predict disease progression.

• Maxillofacial Plastic and Reconstructive Surgery
• /
• v.29 no.6
• /
• pp.520-526
• /
• 2007

The accuracy of model surgery is one of important factors which can influence the outcome of orthognathic surgery. To evaluate the accuracy of digitalized model surgery, we tried the model surgery on a software after transferring the mounted model block into a digital model, and compared the results with that of classical manual model surgery. We could get the following results, which can be used as good baseline analysis for the clinical application. 1. We made the 3D scanning of dental model blocks, and mounted on a software. And we performed the model surgery according to the previously arranged surgical plans, and let the rapid prototyping machine produce the surgical wafer. All through these process, we could confirm that the digital model surgery is feasible without difficulties. 2. The digital model surgery group (Group 2) showed a mean error of $0.0{\sim}0.1mm$ for moving the maxillary model block to the target position. And Group 1, which was done by manual model surgery, presented a mean error of $0.1{\sim}1.2mm$, which is definitely greater than those of Group 2. 3. Remounted maxillary model block with the wafers produced by digital model surgery from Group 2 showed the less mean error (0.2 to 0.4 mm) than that produced by manual model surgery in Group 1 (0.3 to 1.4 mm). From these results, we could confirm that the digital model surgery in Group 2 presented less error than manual model surgery of Group 1. And the model surgery by digital manipulation is expected to have less influence from the individual variation or degree of expertness. So the increased accuracy and enhanced manipulability will serve the digital model surgery as the good candidate for the improvement and replacement of the classical model surgery, if careful preparation works for the clinical adjustment is accompanied.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.27 no.1
• /
• pp.693-700
• /
• 2009

For real-time precise GPS data processing such as a long baseline network RTK (Real-Time Kinematic) survey, PPP (Precise Point Positioning) and monitoring of ionospheric/tropospheric delays, it is necessary to guarantee accuracy comparable to IGS (International GNSS Service) precise orbit with no latency. As a preliminary study for determining near real-time satellite orbits, the general procedures of satellite orbit determination, especially the dynamic approach, were studied. In addition, the transformation between terrestrial and inertial reference frames was tested to integrate acceleration. The IAU 1976/1980 precession/nutation model showed a consistency of 0.05 mas with IAU 2000A model. Since the IAU 2000A model has a large number of nutation components, it took more time to compute the transformation matrix. The classical method with IAU 2000A model was two times faster than the NRO (non-rotating origin) approach, while there is no practical difference between two transformation matrices.

• The Korean Journal of Pain
• /
• v.23 no.3
• /
• pp.166-171
• /
• 2010

Background: Neuropathic pain resulting from diverse causes is a chronic condition for which effective treatment is lacking. The goal of this study was to test whether dexamethasone exerts a preemptive analgesic effect with bupivacaine when injected perineurally in the spared nerve injury model. Methods: Fifty rats were randomly divided into five groups. Group 1 (control) was ligated but received no drugs. Group 2 was perineurally infiltrated (tibial and common peroneal nerves) with 0.4% bupivacaine (0.2 ml) and dexamethasone (0.8 mg) 10 minutes before surgery. Group 3 was infiltrated with 0.4% bupivacaine (0.2 ml) and dexamethasone (0.8 mg) after surgery. Group 4 was infiltrated with normal saline (0.2 ml) and dexamethasone (0.8 mg) 10 minutes before surgery. Group 5 was infiltrated with only 0.4% bupivacaine (0.2 ml) before surgery. Rat paw withdrawal thresholds were measured using the von Frey hair test before surgery as a baseline measurement and on postoperative days 3, 6, 9, 12, 15, 18 and 21. Results: In the group injected preoperatively with dexamethasone and bupivacaine, mechanical allodynia did not develop and mechanical threshold forces were significantly different compared with other groups, especially between postoperative days 3 and 9 (P < 0.05). Conclusions: In conclusion, preoperative infiltration of both dexamethasone and bupivacaine showed a significantly better analgesic effect than did infiltration of bupivacaine or dexamethasone alone in the spared nerve injury model, especially early on after surgery.