• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제13권8호
• /
• pp.4285-4299
• /
• 2019

Adversarial attack is a technique that causes a malfunction of classification models by adding noise that cannot be distinguished by humans, which poses a threat to a deep learning model. In this paper, we propose an efficient method to detect adversarial images using Gaussian process regression. Existing deep learning-based adversarial detection methods require numerous adversarial images for their training. The proposed method overcomes this problem by performing classification based on the statistical features of adversarial images and clean images that are extracted by Gaussian process regression with a small number of images. This technique can determine whether the input image is an adversarial image by applying Gaussian process regression based on the intermediate output value of the classification model. Experimental results show that the proposed method achieves higher detection performance than the other deep learning-based adversarial detection methods for powerful attacks. In particular, the Gaussian process regression-based detector shows better detection performance than the baseline models for most attacks in the case with fewer adversarial examples.

• Structural Engineering and Mechanics
• /
• 제75권6호
• /
• pp.771-784
• /
• 2020

Reliability analysis techniques combining with various surrogate models have attracted increasing attention because of their accuracy and great efficiency. However, they primarily focus on the structures with continuous response, while very rare researches on the reliability analysis for structures with discontinuous response are carried out. Furthermore, existing adaptive reliability analysis methods based on importance sampling (IS) still have some intractable defects when dealing with small failure probability, and there is no related research on reliability analysis for structures involving discontinuous response and small failure probability. Therefore, this paper proposes a novel reliability analysis method called AGPC-IS for such structures, which combines adaptive Gaussian process classification (GPC) and adaptive-kernel-density-estimation-based IS. In AGPC-IS, an efficient adaptive strategy for design of experiments (DoE), taking into consideration the classification uncertainty, the sampling uniformity and the regional classification accuracy improvement, is developed with the purpose of improving the accuracy of Gaussian process classifier. The adaptive kernel density estimation is introduced for constructing the quasi-optimal density function of IS. In addition, a novel and more precise stopping criterion is also developed from the perspective of the stability of failure probability estimation. The efficiency, superiority and practicability of AGPC-IS are verified by three examples.

• 한국정보과학회:학술대회논문집
• /
• 한국정보과학회 2007년도 가을 학술발표논문집 Vol.34 No.2 (C)
• /
• pp.485-489
• /
• 2007

Gaussian process classifiers (GPCs) are fully statistical kernel classification models which have a latent function with Gaussian process prior Recently, EP approximation method has been proposed to infer the posterior over the latent function. It can have a special hyperparameter which can treat outliers potentially. In this paper, we propose the outlier robust algorithm which alternates EP and the hyperparameter updating until convergence. We also show its usefulness with the simulation results.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 2006년 학술대회 논문집 정보 및 제어부문
• /
• pp.470-472
• /
• 2006

Sensor networks are the results of convergence of very important technologies such as wireless communication and micro electromechanical systems. In recent years, sensor networks found a wide applicability in various fields such as environment and health, industry scene system monitoring, etc. A very important step for these many applications is pattern classification and recognition of data collected by sensors installed or deployed in different ways. But, pattern classification and recognition are sometimes difficult to perform. Systematic approach to pattern classification based on modem learning techniques like Multivariate Gaussian mixture models, can greatly simplify the process of developing and implementing real-time classification models. This paper proposes a new recognition system which is hierarchically composed of many sensor nodes having the capability of simple processing and wireless communication. The proposed system is able to perform context classification of sensed data using the Multivariate Gaussian function. In order to verify the usefulness of the proposed system, it was applied to intelligent dust collecting system.

• IEIE Transactions on Smart Processing and Computing
• /
• 제4권4호
• /
• pp.202-208
• /
• 2015

In this study, we propose a new inference algorithm for a multiclass Gaussian process classification model using a variational EM framework and the Laplace approximation (LA) technique. This is performed in two steps, called expectation and maximization. First, in the expectation step (E-step), using Bayes' theorem and the LA technique, we derive the approximate posterior distribution of the latent function, indicating the possibility that each observation belongs to a certain class in the Gaussian process classification model. In the maximization step, we compute the maximum likelihood estimators for hyper-parameters of a covariance matrix necessary to define the prior distribution of the latent function by using the posterior distribution derived in the E-step. These steps iteratively repeat until a convergence condition is satisfied. Moreover, we conducted the experiments by using synthetic data and Iris data in order to verify the performance of the proposed algorithm. Experimental results reveal that the proposed algorithm shows good performance on these datasets.

• 제어로봇시스템학회논문지
• /
• 제11권12호
• /
• pp.991-995
• /
• 2005

Sensor networks are the results of convergence of very important technologies such as wireless communication and micro electromechanical systems. In recent years, sensor networks found a wide applicability in various fields such as health, environment and habitat monitoring, military, etc. A very important step for these many applications is pattern classification and recognition of data collected by sensors installed or deployed in different ways. But, pattern classification and recognition are sometimes difficult to perform. Systematic approach to pattern classification based on modern teaming techniques like Multivariate Gaussian mixture models, can greatly simplify the process of developing and implementing real-time classification models. This paper proposes a new recognition system which is hierarchically composed of many sensor nodes haying the capability of simple processing and wireless communication. The proposed system is able to perform classification of sensed data using the Multivariate Gaussian function. In order to verify the usefulness of the proposed system, it was applied to hand gesture recognition system.

• 반도체디스플레이기술학회지
• /
• 제10권2호
• /
• pp.1-5
• /
• 2011

In recent years, worldwide production of solar wafers increased rapidly. Therefore, the solar wafer technology in the developed countries already has become an industry, and related industries such as solar wafer manufacturing equipment have developed rapidly. In this paper we propose the color classification method of the polycrystalline solar wafer that needed in manufacturing equipment. The solar wafer produced in the manufacturing process does not have a uniform color. Therefore, the solar wafer panels made with insensitive color uniformity will fall off the aesthetics. Gaussian mixture models (GMM) are among the most statistically mature methods for clustering and we use the Gaussian mixture models for the classification of the polycrystalline solar wafers. In addition, we compare the performance of the color feature vector from various color space for color classification. Experimental results show that the feature vector from YCbCr color space has the most efficient performance and the correct classification rate is 97.4%.

• 한국정보과학회논문지:소프트웨어및응용
• /
• 제36권8호
• /
• pp.633-641
• /
• 2009

본 논문에서는 아날로그 데이터 처리가 가능하고, 온라인 학습, 학습 중 새로운 클래스 추가등의 특징을 가진 패턴 인식기를 제안하였다. 제안한 패턴 인식기는 계층적 구조를 가지고 있으며, 각 레벨별로 서로 다른 메트릭을 적용하여 분류 성능을 향상 시켰다. 제안한 패턴 인식기는 신경망 기반의 패턴 인식 알고리즘인 Gaussian ARTMAP 모델을 기반으로 하고 있다. Gaussian ARTMAP 모델을 계층적으로 구성하고, 계층마다 서로 다른 특징을 학습하도록 하기 위하여 Principal Component Emphasis (P.C.E) 방법을 제안하였으며, 이를 이용하여 새로운 메트릭을 생성하는 방법을 제안하였다. P.C.E는 학습된 입력 데이터들의 분산을 이용하여 클래스 내의 공통 속성을 나타내는 분산이 작은 차원을 제거하고 패턴 간의 서로 다른 속성을 나타내는 분산이 큰 차원만 유지하는 방법이다. 제안한 알고리즘의 학습 과정에서 교사 신호와 다르게 분류된 패턴이 발생하면 잘못 분류 된 클래스와 입력된 패턴을 분리하기 위하여 P.C.E를 수행하고 하위 노드에서 학습하게 된다. 실험 결과 제안한 모델은 기존에 제안된 패턴 인식 모델들 보다 높은 분류 성능을 가지고 있음을 확인하였다.

• 대한전자공학회논문지SP
• /
• 제46권3호
• /
• pp.138-142
• /
• 2009

본 논문에서는 보이스 피싱 (Voice Pishing) 예방을 위한 알고리즘을 최소 분류 오차 기법 (Minimum Classification Error)을 기반으로 제한하다. 휴대폰으로 전송되어진 신호를 기반으로 3GPP2 Selectable Mode Vocoder (SMV)의 복호화 과정에서 자동적으로 추출되는 중요 특징벡터를 사용하여 Gaussian Mixture Model (GMM)을 구성하고 이를 기반으로 구해지는 로그(Log) 기반의 우도 (Likelihood)를 사용한 변별적 가중치 학습을 사용하여 보이스 피싱 예방을 위한 검출 알고리즘을 제안하다. 실험 결과 제안된 보이스 피싱 알고리즘이 기존의 방법에 비해 우수한 성능을 보인 것을 알 수 있었다.

• 제어로봇시스템학회논문지
• /
• 제4권3호
• /
• pp.372-377
• /
• 1998

This paper deals with the detection, feature extraction and classification of surface defects in cold rolled strips. Inspection systems are one of the most important fields in factory automation. Defects such as slipmark and dullmark can be effectively detected with a Gaussian matched filter because their shapes are similar to Gaussian. It is justified that the proposed WF(Wavelet Frame) method could be regarded as multiscale Gaussian matched filter which can be applied to the inspection of cold rolled strip. After a wavelet frame transform, the entropies and moments are computed for each subband which pass through both local low pass filter and nonlinear operator. With these features as input, a MLP(Multi Layer Perceptron) is used as a classifier. The proposed inspection method was applied to the real images with defects, and hence showed good performance. The role of each extracted feature is analyzed by KLT(Karhunen-Loeve Transform).