• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.32 no.5
• /
• pp.817-826
• /
• 2022

AI technology is being successfully introduced in many fields, and models deployed as a service are deployed with black box environment that does not expose the model's information to protect intellectual property rights and data. In a black box environment, attackers try to steal data or parameters used during training by using model output. This paper proposes a method of inferring the type of model to directly find out the composition of layer of the target model, based on the fact that there is no attack to infer the information about the type of model from the deep learning model. With ResNet, VGGNet, AlexNet, and simple convolutional neural network models trained with MNIST datasets, we show that the types of models can be inferred using the output values in the gray box and black box environments of the each model. In addition, we inferred the type of model with approximately 83% accuracy in the black box environment if we train the big and small relationship feature that proposed in this paper together, the results show that the model type can be infrerred even in situations where only partial information is given to attackers, not raw probability vectors.

• Communications for Statistical Applications and Methods
• /
• v.19 no.3
• /
• pp.405-414
• /
• 2012

In a stratified sample, the sampling frame is divided into non-overlapping groups or strata (e.g. geographical areas, age-groups, and genders). A sample is taken from each stratum, if this sample is a simple random sample it is referred to as stratified random sampling. In this paper, we study the bootstrap inference (including confidence interval) and test for a stratified population mean. We also introduce the bootstrap consistency based on limiting distribution related to the plug-in estimator of the population mean. We suggest three bootstrap confidence intervals such as standard bootstrap method, percentile bootstrap method and studentized bootstrap method. We also suggest a bootstrap test method computing the $ASL_{boot}$(Achieved Significance Level). The results of estimation are verified using simulation.

• Journal of KIISE:Computing Practices and Letters
• /
• v.15 no.1
• /
• pp.72-76
• /
• 2009

Recently, as the interest of ubiquitous computing has been increased there has been lots of research about recognizing human activities to provide services in this environment. Especially, in mobile environment, contrary to the conventional vision based recognition researches, lots of researches are sensor based recognition. In this paper we propose to recognize the user's activity with multi-modal sensors using hierarchical dynamic Bayesian networks. Dynamic Bayesian networks are trained by the OVR(One-Versus-Rest) strategy. The inferring part of this network uses less calculation cost by selecting the activity with the higher percentage of the result of a simpler Bayesian network. For the experiment, we used an accelerometer and a physiological sensor recognizing eight kinds of activities, and as a result of the experiment we gain 97.4% of accuracy recognizing the user's activity.

• Journal of the Institute of Electronics Engineers of Korea SC
• /
• v.47 no.5
• /
• pp.18-23
• /
• 2010

In this paper, we propose a computationally efficient method detecting the P300 wave for brain-machine interface. Electrophysiological researches have shown that the P300 wave's potential is decreased when human intention matches visual stimulation. Motivated by this fact, we can infer human intention for brain-machine interface by detecting the P300 wave's potential decrease. The P300 wave is recorded from EEG(electroencephalogram) electrodes attached on human brain skull after giving alphabetical stimulation. To detect the potential decrease in P300, firstly we statistically model the P300 wave's negative potential. Then we infer human intention based on maximum likelihood estimation. The proposed method was evaluated on the data recorded from three healthy human subjects. The method achieved an averaging accuracy of 98% from subject k, 90% from subject j and 79.8% from subject h.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.16 no.4
• /
• pp.2401-2406
• /
• 2015

Pipelines are used by large heavy industries to deliver various types of fluids. Since this is important to maintain the performance of large systems, it is necessary to accurately predict remaining life of the corroded pipeline. However, predicting the remaining life is difficult due to uncertainties in the associated variables, such as geometries, material properties, corrosion rate, etc. In this paper, a statistical method for predicting corrosion remaining life is proposed using Bayesian inference. To accomplish this, pipeline failure probability was calculated using prior information about pipeline failure pressure according to elapsed time, and the given experimental data based on Bayes' rule. The corrosion remaining life was calculated as the elapsed time with 10 % failure probability. Using 10 and 50 samples generated from random variables affecting the corrosion of the pipe, the pipeline failure probability was estimated, after which the estimated remaining useful life was compared with the assumed true remaining useful life.

• Journal of Korean Society for Geospatial Information Science
• /
• v.18 no.4
• /
• pp.61-69
• /
• 2010

The collapse of a slope or cut embankment brings much damage to life and property. Accordingly, it is very important to analyze the spatial distribution by calculating the landslide susceptibility in the estimation of the risk of landslide occurrence. The heuristic, statistic, deterministic, and probabilistic methods have been introduced to make landslide susceptibility maps. In many cases, however, the reliability is low due to insufficient field data, and the qualitative experience and knowledge of experts could not be combined with the quantitative mechanical?analysis model in the existing methods. In this paper, new modeling method for a probabilistic landslide susceptibility analysis combined Bayesian Network with ontology model about experts' knowledge and spatial data was proposed. The ontology model, which was made using the reasoning engine, was automatically converted into the Bayesian Network structure. Through conditional probabilistic reasoning using the created Bayesian Network, landslide susceptibility with uncertainty was analyzed, and the results were described in maps, using GIS. The developed Bayesian Network was then applied to the test-site to verify its effect, and the result corresponded to the landslide traces boundary at 86.5% accuracy. We expect that general users will be able to make a landslide susceptibility analysis over a wide area without experts' help.

• Korean Journal of Cognitive Science
• /
• v.17 no.4
• /
• pp.303-321
• /
• 2006

According to Harris'(1966) distributional hypothesis, understanding the meaning of a word is thought to be dependent on its context. Under this hypothesis about human language ability, this paper proposes a computational model for native speaker's language processing mechanism concerning word sense disambiguation, based on two sets of experiments. Among the three computational models discussed in this paper, namely, the logic model, the probabilistic model, and the probabilistic inference model, the experiment shows that the logic model is first applied fer semantic disambiguation of the key word. Nexr, if the logic model fails to apply, then the probabilistic model becomes most relevant. The three models were also compared with the test results in terms of Pearson correlation coefficient value. It turns out that the logic model best explains the human decision behaviour on the ambiguous words, and the probabilistic inference model tomes next. The experiment consists of two pans; one involves 30 sentences extracted from 1 million graphic-word corpus, and the result shows the agreement rate anong native speakers is at 98% in terms of word sense disambiguation. The other pm of the experiment, which was designed to exclude the logic model effect, is composed of 50 cleft sentences.

• School Mathematics
• /
• v.13 no.3
• /
• pp.363-384
• /
• 2011

This study is based on the recognition that the school mathematics education should reinforce the heuristic and constructional aspects related with discoveries of mathematical rules and understanding of mathematical concepts from real world situations as well as the deductive and formal aspects emphasizing on mathematical contents precisely. The 11th grade students of one class from a city high school with average were chosen. They were given time to learn various functions of Excel in regular classes of "Information Society and Computer" subject. They don't have difficulty using cells, mathematical functions and statistical functions in spreadsheet. Experiment was performed for six weeks and there were two hours of classes in a week. Considering the results of this research, teaching materials using spreadsheets play an important role in helping students to experience probabilistic and statistical reasoning and construct mathematical thinking. This implies that teaching materials using spreadsheet provide students with an opportunity to interact with probabilistic and statistical situations by adopting engineering which can encourage students to observe and experience various aspects of real world in authentic situations.

• Journal of the Korean Society of Hazard Mitigation
• /
• v.9 no.6
• /
• pp.143-151
• /
• 2009

Extreme hydrologic events cause serious disaster, such as flood and drought. Many researchers have an effort to estimate design rainfalls or discharges. This study evaluated parameter estimation methods to estimate probability rainfalls with low uncertainty which will be used in design rainfalls. This study collected rainfall data from Incheon, Gangnueng, Gwangju, Busan, and Chupungryong gage station, and generated synthetic rainfall data using ARMA model. This study employed the maximum likelihood method and the Bayesian inference method for estimating parameters of the Gumbel and GEV distribution. Using a bootstrap resampling method, this study estimated the confidence intervals of estimated probability rainfalls. Based on the comparison of the confidence intervals, this study recommended a proper parameter estimation method for estimating probability rainfalls which have a low uncertainty.

• Journal of the Korean Data and Information Science Society
• /
• v.11 no.2
• /
• pp.189-199
• /
• 2000

A major deficiency of laboratory testing of rock structure is that the structures are limited in size and therefore present a very small and highly selective sample of the rock mass from which were removed. In a typical engineering project, the samples tested in the laboratory represent only a very small fraction of one percent of the volume of the rock mass. In this paper, we calculate the representative orientation of the resultant vector, the measure of the degree of clustering, the volume of rock mass, the trace length of discontinuity spacing under underlying distributions. And we generate the random fracture networks using real data. We propose the calculating the trace length.