• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.58 no.7
• /
• pp.1443-1449
• /
• 2009

This paper presents a profiling model of a wide-window superscalar microprocessor using multiple branch prediction. The key idea is to apply statistical profiling technique to the superscalar microprocessor with a wide instruction window and a multiple branch predictor. The statistical profiling data are used to obtain a synthetical instruction trace, and the consecutive multiple branch prediction rates are utilized for running trace-driven simulation on the synthesized instruction trace. We describe our design and evaluate it with the SPEC 2000 integer benchmarks. Our performance model can achieve accuracy of 8.5 % on the average.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.30 no.6
• /
• pp.1291-1300
• /
• 2020

Recently, a deep learning-based non-profiling side-channel analysis was proposed. The deep learning-based non-profiling analysis is a technique that trains a neural network model for all guessed keys and then finds the correct secret key through the difference in the training metrics. As the performance of non-profiling analysis varies greatly depending on the neural network training model design, a correct model design criterion is required. This paper describes the two types of loss functions and eight labeling methods used in the training model design. It predicts the analysis performance of each labeling method in terms of non-profiling analysis and power consumption model. Considering the characteristics of non-profiling analysis and the HW (Hamming Weight) power consumption model is assumed, we predict that the learning model applying the HW label without One-hot encoding and the Correlation Optimization (CO) loss will have the best analysis performance. And we performed actual analysis on three data sets that are Subbytes operation part of AES-128 1 round. We verified our prediction by non-profiling analyzing two data sets with a total 16 of MLP-based model, which we describe.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.18 no.3
• /
• pp.89-100
• /
• 2023

Recently, on-device artificial intelligence (AI) solutions using mobile devices and embedded edge devices have emerged in various fields, such as computer vision, to address network traffic burdens, low-energy operations, and security problems. Although vision transformer deep learning models have outperformed conventional convolutional neural network (CNN) models in computer vision, they require more computations and parameters than CNN models. Thus, they are not directly applicable to embedded edge devices with limited hardware resources. Many researchers have proposed various model compression methods or lightweight architectures for vision transformers; however, there are only a few studies evaluating the effects of model compression techniques of vision transformers on performance. Regarding this problem, this paper presents a performance evaluation of vision transformers on embedded platforms. We investigated the behaviors of three vision transformers: DeiT, LeViT, and MobileViT. Each model performance was evaluated by accuracy and inference time on edge devices using the ImageNet dataset. We assessed the effects of the quantization method applied to the models on latency enhancement and accuracy degradation by profiling the proportion of response time occupied by major operations. In addition, we evaluated the performance of each model on GPU and EdgeTPU-based edge devices. In our experimental results, LeViT showed the best performance in CPU-based edge devices, and DeiT-small showed the highest performance improvement in GPU-based edge devices. In addition, only MobileViT models showed performance improvement on EdgeTPU. Summarizing the analysis results through profiling, the degree of performance improvement of each vision transformer model was highly dependent on the proportion of parts that could be optimized in the target edge device. In summary, to apply vision transformers to on-device AI solutions, either proper operation composition and optimizations specific to target edge devices must be considered.

• 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.

• Nuclear Engineering and Technology
• /
• v.54 no.8
• /
• pp.3027-3033
• /
• 2022

Background: In this study, various types of deep-learning models for predicting in vitro radiosensitivity from gene-expression profiling were compared. Methods: The clonogenic surviving fractions at 2 Gy from previous publications and microarray gene-expression data from the National Cancer Institute-60 cell lines were used to measure the radiosensitivity. Seven different prediction models including three distinct multi-layered perceptrons (MLP), four different convolutional neural networks (CNN) were compared. Folded cross-validation was applied to train and evaluate model performance. The criteria for correct prediction were absolute error < 0.02 or relative error < 10%. The models were compared in terms of prediction accuracy, training time per epoch, training fluctuations, and required calculation resources. Results: The strength of MLP-based models was their fast initial convergence and short training time per epoch. They represented significantly different prediction accuracy depending on the model configuration. The CNN-based models showed relatively high prediction accuracy, low training fluctuations, and a relatively small increase in the memory requirement as the model deepens. Conclusion: Our findings suggest that a CNN-based model with moderate depth would be appropriate when the prediction accuracy is important, and a shallow MLP-based model can be recommended when either the training resources or time are limited.

• Journal of KIISE
• /
• v.41 no.12
• /
• pp.1117-1125
• /
• 2014

Folksonomy data, which is derived from social tagging systems, is a useful source for understanding a user's intention and interest. Using the folksonomy data, it is possible to create an accurate user profile which can be utilized to build a personalized search system. However there are limitations in some of the traditional methods such as Vector Space Model(VSM) for user profiling and similarity computation. This paper suggests a novel method with graph-based user and document profile which uses the proximity information of query terms to improve personalized search. We demonstrate the performance of the suggested method by comparing its performance with several state-of-the-art VSM based personalization models in two different folksonomy datasets. The results show that the proposed model constantly outperforms the other state-of-the-art personalization models. Furthermore, the parameter sensitivity results show that the proposed model is parameter-free in that it is not affected by the idiosyncratic nature of datasets.

• Industrial Engineering and Management Systems
• /
• v.12 no.2
• /
• pp.118-129
• /
• 2013

In this study, we intend to identify key financial variables that can accurately classify Green IT leaders against Green IT followers. In particular, we build and compare single and meta-classifiers to identify the relationship between environmental performance and financial performance, while focusing on selecting and interpreting a final prediction model with a smaller set of financial performance indicators. Our experimental results demonstrate that several key variables representing the size, financial resources, operational efficiency, and risk-taking tendency of an organization can successfully identify Green IT leaders with approximately 90% of accuracy. In addition, we find that Green IT leaders show a higher utilization rate of Web pages as a green marketing channel than Green IT followers while they share common layouts of Web publication to build green IT brands with some differences.

• Journal of Korean Society of Industrial and Systems Engineering
• /
• v.37 no.3
• /
• pp.51-63
• /
• 2014

Because of the changing educational environments, Korean universities have to stand up to the challenge of enhancing their performances in educational activities. Therefore, Korean universities should seek ways to improve its competitiveness. The evaluation of Korean universities has been conducted by the Government and other media agencies like JoongAng's Daily to promote the development of universities. The purpose of this study is to evaluate and compare the relative efficiency of Korean universities using DEA (Data Envelopment Analysis) and AHP (Analytic Hierarchy Process) models. To produce reasonable results, major input and output variables have been determined in a step using a Profiling method, Spearman correlation analysis and AHP model. Also, this study shows that the relative efficiency for Korean universities is obtained not only based on quantity of output, but also on the basis of the quality of the output. Drawing upon the results of the analysis, this study provides a comprehensive and practical method for evaluating the performance of universities in terms of efficiency.

• Journal of Distribution Science
• /
• v.20 no.6
• /
• pp.1-10
• /
• 2022

Purpose: This research proposes a new modified Recency-Frequency-Monetary (RFM) model by extending the model with spatial analysis for supporting decision-makers in discovering the promotional target market. Research design, data and methodology: This quantitative research utilizes data-mining techniques and the RFM model to cluster a university's provider schools. The RFM model was modified by adapting its variables to the university's marketing context and adding a district's potential (D) variable based on heatmap analysis using Geographic Information System (GIS) and K-means clustering. The K-prototype algorithm and the Elbow method were applied to find provider school clusters using the proposed RFM-D model. After profiling the clusters, the target segment was assigned. The model was validated using empirical data from an Indonesian university, and its performance was compared to the Customer Lifetime Value (CLV)-based RFM utilizing accuracy, precision, recall, and F1-score metrics. Results: This research identified five clusters. The target segment was chosen from the highest-value and high-value clusters that comprised 17.80% of provider schools but can contribute 75.77% of students. Conclusions: The proposed model recommended more targeted schools in higher-potential districts and predicted the target segment with 0.99 accuracies, outperforming the CLV-based model. The empirical findings help university management determine the promotion location and allocate resources for promotional information distribution and student recruitment.

• Health Policy and Management
• /
• v.27 no.2
• /
• pp.149-156
• /
• 2017

Background: This study was conducted to evaluate the performance of the Hierarchical Condition Category (HCC) model, identify potentially high-cost patients, and examine the effects of adding prior utilization to the risk model using Korean claims data. Methods: We incorporated 2 years of data from the National Health Insurance Services-National Sample Cohort. Five risk models were used to predict health expenditures: model 1 (age/sex groups), model 2 (the Center for Medicare and Medicaid Services-HCC with age/sex groups), model 3 (selected 54 HCCs with age/sex groups), model 4 (bed-days of care plus model 3), and model 5 (medication-days plus model 3). We evaluated model performance using $R^2$ at individual level, predictive positive value (PPV) of the top 5% of high-cost patients, and predictive ratio (PR) within subgroups. Results: The suitability of the model, including prior use, bed-days, and medication-days, was better than other models. $R^2$ values were 8%, 39%, 37%, 43%, and 57% with model 1, 2, 3, 4, and 5, respectively. After being removed the extreme values, the corresponding $R^2$ values were slightly improved in all models. PPVs were 16.4%, 25.2%, 25.1%, 33.8%, and 53.8%. Total expenditure was underpredicted for the highest expenditure group and overpredicted for the four other groups. PR had a tendency to decrease from younger group to older group in both female and male. Conclusion: The risk adjustment models are important in plan payment, reimbursement, profiling, and research. Combined prior use and diagnostic data are more powerful to predict health costs and to identify high-cost patients.