• Smart Media Journal
• /
• v.9 no.3
• /
• pp.80-89
• /
• 2020

Out - of - distribution (OOD) samples are frequently encountered when deploying a classification model in plenty of real-world machine learning-based applications. Those samples are normally sampling far away from the training distribution, but many classifiers still assign them high reliability to belong to one of the training categories. In this study, we address the problem of removing OOD examples by estimating marginal density estimation using variational autoencoder (VAE). We also investigate other proper methods, such as temperature scaling, Gaussian discrimination analysis, and label smoothing. We use Chonnam National University (CNU) weeds dataset as the in - distribution dataset and CIFAR-10, CalTeach as the OOD datasets. Quantitative results show that the proposed framework can reject the OOD test samples with a suitable threshold.

• Food Science and Biotechnology
• /
• v.14 no.1
• /
• pp.168-170
• /
• 2005

The myeolchi-jeot samples were divided into different groups with or without the supplementation with biogenic amines. Subsequently, the samples were placed in an oven at $80^{\circ}C$ for 1 hr to allow the chemical reaction to proceed, and then were analyzed for N-nitrosamine contents using GC-TEA. N-nitrosamine was not detected in any of the myeolchi-jeot samples which had been treated with/without sodium nitrite. On the other hand, the yield of N-nitrosopyrrolidine from 1,000 mg/kg of putrescine and spermidine in the myeolchi-jeot samples (treated with 5 mg/kg of sodium nitrite) was 0.002 and 0.014%, respectively. N-nitrosamine was not produced from any other biogenic amines like, histamine, tyramine, cadaverine and spermine. In addition, curing and heating were the factors which influenced the formation of N-nitrosamine during the nitrosation of biogenic polyamines. For the formation of N-nitrosamine in the food systems, treatment with sodium nitrite and heating at appropriate temperature along with the satisfied supplementation of biogenic polyamines are required.

• Computers and Concrete
• /
• v.15 no.4
• /
• pp.503-514
• /
• 2015

Fibre-added concretes are frequently used in large site applications such as slab and airports as well as in bearing system elements or prefabricated elements. It is very difficult to determine the mechanical properties of the fibre-added concretes by experimental methods in situ. The purpose of this study is to develop an artificial neural network (ANN) model in order to predict the compressive and bending strengths of hybrid fibre-added and non-added concretes. The strengths have been predicted by means of the data that has been obtained from destructive (DT) and non-destructive tests (NDT) on the samples. NDTs are ultrasonic pulse velocity (UPV) and Rebound Hammer Tests (RH). 105 pieces of cylinder samples with a dimension of $150{\times}300mm$, 105 pieces of bending samples with a dimension of $100{\times}100{\times}400mm$ have been manufactured. The first set has been manufactured without fibre addition, the second set with the addition of %0.5 polypropylene and %0.5 steel fibre in terms of volume, and the third set with the addition of %0.5 polypropylene, %1 steel fibre. The water/cement (w/c) ratio of samples parametrically varies between 0.3-0.9. The experimentally measured compressive and bending strengths have been compared with predicted results by use of ANN method.

• Environmental Engineering Research
• /
• v.10 no.3
• /
• pp.112-121
• /
• 2005

The effectiveness of quicklime-based stabilization/solidification (S/S) in immobilizing lead (Pb) was assessed by performing semi-dynamic leaching tests (ANS16.1). In order to simulate landfill leaching conditions, the ANS 16.1 test was modified by using 0.014 N acetic acid (pH = 3.25) instead of distilled water. Artificial soil samples as well as field soil samples contaminated with Pb were tested. The effectiveness of quicklime treatment was evaluated by determining diffusion coefficients ($D_e$) and leachability indices (LX). A model developed by de Groot and van der Sloat was used to elucidate the controlling Pb leaching mechanisms. Overall, upon quicklime treatment Pb leachability was significantly reduced in a]l of the samples tested. The mean LX values were higher than 9 for an artificial soil sample containing 30% kaolinite treated with 10% quicklime and for a field soil sample treated with 10% quicklime, which suggests that S/S treated soils can be considered acceptable for "controlled utilization". Moreover, quicklime treatment was more effective in artificially contaminated soil with high kaolinite content (30%), indicating the amount of clay plays an important role in the success of the treatment. The controlling Pb leaching mechanism was found to be diffusion, in all quicklime treated samples.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.8
• /
• pp.3962-3980
• /
• 2019

To deal with single sample face recognition, this paper presents a patch based semi-supervised linear regression (PSLR) algorithm, which draws facial variation information from unlabeled samples. Each facial image is divided into overlapped patches, and a regression model with mapping matrix will be constructed on each patch. Then, we adjust these matrices by mapping unlabeled patches to $[1,1,{\cdots},1]^T$. The solutions of all the mapping matrices are integrated into an overall objective function, which uses ${\ell}_{2,1}$-norm minimization constraints to improve discrimination ability of mapping matrices and reduce the impact of noise. After mapping matrices are computed, we adopt majority-voting strategy to classify the probe samples. To further learn the discrimination information between probe samples and obtain more robust mapping matrices, we also propose a multistage PSLR (MPSLR) algorithm, which iteratively updates the training dataset by adding those reliably labeled probe samples into it. The effectiveness of our approaches is evaluated using three public facial databases. Experimental results prove that our approaches are robust to illumination, expression and occlusion.

• Journal of Microbiology and Biotechnology
• /
• v.28 no.12
• /
• pp.1971-1981
• /
• 2018

In this study, we investigated the altered enzymatic activities and metabolite profiles of koji fermented using varying permutations of Aspergillus oryzae and/or Bacillus amyloliquefaciens. Notably, the protease and ${\beta}$-glucosidase activities were manifold increased in co-inoculated (CO) koji samples (co-inoculation of A. oryzae and B. amyloliquefaciens). Furthermore, gas chromatography-mass spectrometry (GC-MS)-based metabolite profiling indicates that levels of amino acids, organic acids, sugars, sugar alcohols, fatty acids, nucleosides, and vitamins were distinctly higher in CO, SA (sequential inoculation of A. oryzae, followed by B. amyloliquefaciens), and SB (sequential inoculation of B. amyloliquefaciens, followed by A. oryzae). The multivariate principal component analysis (PCA) plot based on GC-MS datasets indicated a clustered pattern for MA and MB (koji samples inoculated either with A. oryzae or B. amyloliquefaciens) across PC2 (20.0%). In contrast, the CO, SA, and SB metabolite profiles displayed segregated patterns across PLS1 (22.2%) and PLS2 (21.1%) in the partial least-square discriminant analysis (PLS-DA) model. Intriguingly, the observed disparity in the levels of primary metabolites was engendered largely by higher relative levels of sugars and sugar alcohols in MA, SA, and CO koji samples, which was commensurate with the relative amylase activities in respective samples. Collectively, the present study emphasizes the utility of integrated biochemical and metabolomic approaches for achieving the optimal permutation of fermentative inocula for industrial koji preparation.

• Proceedings of the KSRS Conference
• /
• 2003.11a
• /
• pp.637-639
• /
• 2003

As the number of spectral bands of high spectral resolution data increases, the capability to detect more detailed classes should also increase, and the classification accuracy should increase as well. Often, it is impossible to access enough training pixels for supervise classification. For this reason, the performance of traditional classification methods isn't useful. In this paper, we propose a new model for classification that operates based on decision fusion. In this classifier, learning is performed at two steps. In first step, only training samples are used and in second step, this classifier utilizes semilabeled samples in addition to original training samples. At the beginning of this method, spectral bands are categorized in several small groups. Information of each group is used as a new source and classified. Each of this primary classifier has special characteristics and discriminates the spectral space particularly. With using of the benefits of all primary classifiers, it is made sure that the results of the fused local decisions are accurate enough. In decision fusion center, some rules are used to determine the final class of pixels. This method is applied to real remote sensing data. Results show classification performance is improved, and this method may solve the limitation of training samples in the high dimensional data and the Hughes phenomenon may be mitigated.

• The Korean Journal of Applied Statistics
• /
• v.13 no.2
• /
• pp.575-590
• /
• 2000

In this paper, we extended the Kim et al.'s two-stage unrelated question model(1992) and the Lee et al.'s improved unrelated question model(1998) to two sample unrelated question model of using two independent samples in the case of unknown $\pi_y$.

• Asian-Australasian Journal of Animal Sciences
• /
• v.19 no.10
• /
• pp.1490-1495
• /
• 2006

The purpose of this research was to develop an optimum composition model for a new synbiotic fermented dairy product with high probiotic cell counts, and to experimentally verify this model. The optimum composition model indicated the growth promoter ratio that could provide the highest growth rate for probiotics in this fermented product. Different levels of growth promoters were first blended with milk to improve the growth rates of probiotics, and the optimum composition model was determined. The probiotic viabilities and chemical properties were analyzed for the samples made using the optimal formula. The optimal combination of the growth promoters for the synbiotic fermented milk product was 1.12% peptides, 3% fructooligosaccharides (FOS), and 1.87% isomaltooligosaccharides (IMO). A product manufactured according to the formula of the optimum model was analyzed, showing that the model was effective in improving the viability of both Lactobacillus spp. and Bifidobacterium spp.

• Proceedings of the Korean Society of Near Infrared Spectroscopy Conference
• /
• 2001.06a
• /
• pp.1624-1624
• /
• 2001

Near-infrared spectroscopy (NIRS) was used to investigate the possibility for application in identification of apple cultivars. Three apple cultivars ‘Kamhong, Hwahong, and Fuji’ produced in Korea were scanned over the range of 1100-2500nm using NIRS (Infra Alzer 500). Two types of samples were used for scanning; one was apple with skin and the other was apple without skin. For cultivar identification, the NIR absorbance spectrums were analyzed by qualitative calibration in “Sesame” analysis program, and the various influence properties such as sugar contents, acidity, color, firmness, and micro-structure were compared in scanned samples. The ‘Kamhong’ cultivar could be identified from ‘Hwahong’ and ‘Fuji’ cultivars using the cluster model analysis. The test samples in calibration between ‘Kamhong’ and ‘Fuji’ cultivars could be completely identified. The test samples in calibration between ‘Kamhong’ and ‘Hwahong’ cultivars could be identified most of all. But, ‘Hwahong’ and ‘Fuji’ cultivars could not be quite classified each other. The apple skin influenced the identification process of apple cultivars. The samples without skin were more difficult to classify in calibration than the samples with skin. The physicochemical properties of apple cultivars showed like the result of identification in calibration using NIRS. Some physicochemical properties of ‘Kamhong’ cultivar were different from those of the other cultivars. Those of ‘Hwahong’ and ‘Fuji’ cultivars showed. similar to each other. The sucrose contents of ‘Kamhong’ cultivar were higher and the fructose contents and firmness of skin and flesh were lower than those of the others. The hypodermis layer of skin in ‘Kamhong’ cultivar was thinner than those of the others. In this studies, the identification of all apple cultivars by NIRS was not quite accurate because of the physicochemical properties which were different in the same cultivar, and inconsistent patterns by culivars in some properties. To solve these problems in NIRS application for apple cultivar identification, further study should be focused on the use of peculiar properties among the apple cultivars.