• IEIE Transactions on Smart Processing and Computing
• /
• 제2권5호
• /
• pp.277-281
• /
• 2013

Modern speaker verification systems based on support vector machines (SVMs) use Gaussian mixture model (GMM) supervectors as their input feature vectors, and the maximum a posteriori (MAP) adaptation is a conventional method for generating speaker-dependent GMMs by adapting a universal background model (UBM). MAP adaptation requires the appropriate amount of input utterance due to the number of model parameters to be estimated. On the other hand, with limited utterances, unreliable MAP adaptation can be performed, which causes adaptation noise even though the Bayesian priors used in the MAP adaptation smooth the movements between the UBM and speaker dependent GMMs. This paper proposes a sparse MAP adaptation method, which is known to perform well in the automatic speech recognition area. By introducing sparse MAP adaptation to the GMM-SVM-based speaker verification system, the adaptation noise can be mitigated effectively. The proposed method utilizes the L0 norm as a regularizer to induce sparsity. The experimental results on the TIMIT database showed that the sparse MAP-based GMM-SVM speaker verification system yields a 42.6% relative reduction in the equal error rate with few additional computations.

• 방송공학회논문지
• /
• 제17권4호
• /
• pp.634-639
• /
• 2012

본 논문에서는 다수의 영상에서 발견된 특징점의 정확한 정합을 위한 비선형 최적화 기법을 제안한다. 영상에서 발견된 특징점은 선형 해법에 의해 다수의 영상간의 변환을 구할 수 있지만 큰 오차를 수반하게 된다. 이는 영상이 생성되는 모델이 비선형이며, 다수시점간의 운동역시 비선형의 형태를 띄기 때문이다. 하지만 다수의 영상의 비선형 최적화는 일반적인 비선형 해법을 도입하였을 때에는 복잡도가 지수적으로 증가하는 단점이 있다. 본 논문에서는 Levenberg-Marquardt 비선형 최적화 방법의 희박해법(Sparse solution)을 이용하여 다수의 특징점간의 변환을 구하는 방법을 보인다.

• Journal of Computing Science and Engineering
• /
• 제10권2호
• /
• pp.58-67
• /
• 2016

With the large amount of complex network data that is increasingly available on the Web, link prediction has become a popular data-mining research field. The focus of this paper is on a link-prediction task that can be formulated as a binary classification problem in complex networks. To solve this link-prediction problem, a sparse-classification algorithm called "Truncated Kernel Projection Machine" that is based on empirical-feature selection is proposed. The proposed algorithm is a novel way to achieve a realization of sparse empirical-feature-based learning that is different from those of the regularized kernel-projection machines. The algorithm is more appealing than those of the previous outstanding learning machines since it can be computed efficiently, and it is also implemented easily and stably during the link-prediction task. The algorithm is applied here for link-prediction tasks in different complex networks, and an investigation of several classification algorithms was performed for comparison. The experimental results show that the proposed algorithm outperformed the compared algorithms in several key indices with a smaller number of test errors and greater stability.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제11권5호
• /
• pp.2590-2606
• /
• 2017

Generally, research of classical image classification algorithms assume that training data and testing data are derived from the same domain with the same distribution. Unfortunately, in practical applications, this assumption is rarely met. Aiming at the problem, a domain adaption image classification approach based on multi-sparse representation is proposed in this paper. The existences of intermediate domains are hypothesized between the source and target domains. And each intermediate subspace is modeled through online dictionary learning with target data updating. On the one hand, the reconstruction error of the target data is guaranteed, on the other, the transition from the source domain to the target domain is as smooth as possible. An augmented feature representation produced by invariant sparse codes across the source, intermediate and target domain dictionaries is employed for across domain recognition. Experimental results verify the effectiveness of the proposed algorithm.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제11권3호
• /
• pp.1700-1721
• /
• 2017

The problem of blocking artifacts is very common in block-based image and video compression, especially at very low bit rates. In this paper, we propose a post-processing method for JPEG-coded image deblocking via sparse representation and adaptive residual threshold. This method includes three steps. First, we obtain the dictionary by online dictionary learning and the compressed images. The dictionary is then modified by the histogram of oriented gradient (HOG) feature descriptor and K-means cluster. Second, an adaptive residual threshold for orthogonal matching pursuit (OMP) is proposed and used for sparse coding by combining blind image blocking assessment. At last, to take advantage of human visual system (HVS), the edge regions of the obtained deblocked image can be further modified by the edge regions of the compressed image. The experimental results show that our proposed method can keep the image more texture and edge information while reducing the image blocking artifacts.

• 디지털콘텐츠학회 논문지
• /
• 제16권4호
• /
• pp.605-613
• /
• 2015

본 논문은 패턴인식에서 자주 사용되는 투영행렬을 희소화하는 문제를 다룬다. 최근 임베디드 시스템이 널리 사용됨에 따라 탑재되는 프로그램의 용량이 제한받는 경우가 빈번히 발생한다. 개발된 프로그램은 상수 데이터를 포함하는 경우가 많다. 예를 들어, 얼굴인식과 같은 패턴인식 프로그램의 경우 고차원 벡터를 저차원 벡터로 차원을 축소하는 투영행렬을 사용하는 경우가 많다. 인식성능 향상을 위해 영상으로부터 매우 높은 차원의 고차원 특징벡터를 추출하는 경우 투영행렬의 사이즈는 매우 크다. 최근 라소 회귀분석 방법을 이용한 RSR(rotated sparse regression) 방법론[1]이 제안되었다. 이 방법론은 여러 실험을 통해 희소행렬을 구하는 가장 우수한 알고리즘 중 하나로 평가받고 있다. 우리는 본 논문에서 RSR을 개선할 수 있는 세 가지 방법론을 제안한다. 즉, 학습데이터에서 이상치를 제거하여 일반화 성능을 높이는 방법, 학습데이터를 랜덤 샘플링하여 희소율을 높이는 방법, RSR의 목적함수에 엘라스틱 넷 회귀분석의 패널티 항을 사용한 E-RSR(elastic net-RSR) 방법을 제안한다. 우리는 실험을 통해 제안한 방법론이 인식률을 희생하지 않으며 희소율을 크게 증가시킴으로써 기존 RSR 방법론을 개선할 수 있음을 보였다.

• 한국정보과학회:학술대회논문집
• /
• 한국정보과학회 2003년도 가을 학술발표논문집 Vol.30 No.2 (1)
• /
• pp.4-6
• /
• 2003

A holistic representation, such as sparse ceding or independent component analysis (ICA), was successfully applied to explain early auditory processing and sound classification. In contrast, Part-based representation is an alternative way of understanding object recognition in brain. In this paper. we employ the non-negative matrix factorization (NMF)［1］which learns parts-based representation for sound classification. Feature extraction methods from spectrogram using NMF are explained. Experimental results show that NMF-based features improve the performance of sound classification over ICA-based features.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제12권7호
• /
• pp.3194-3216
• /
• 2018

Slow Feature Discriminant Analysis (SFDA) is a supervised feature extraction method inspired by biological mechanism. In this paper, a novel method called Two Dimensional Slow Feature Discriminant Analysis via $L_{2,1}$ norm minimization ($2DSFDA-L_{2,1}$) is proposed. $2DSFDA-L_{2,1}$ integrates $L_{2,1}$ norm regularization and 2D statically uncorrelated constraint to extract discriminant feature. First, $L_{2,1}$ norm regularization can promote the projection matrix row-sparsity, which makes the feature selection and subspace learning simultaneously. Second, uncorrelated features of minimum redundancy are effective for classification. We define 2D statistically uncorrelated model that each row (or column) are independent. Third, we provide a feasible solution by transforming the proposed $L_{2,1}$ nonlinear model into a linear regression type. Additionally, $2DSFDA-L_{2,1}$ is extended to a bilateral projection version called $BSFDA-L_{2,1}$. The advantage of $BSFDA-L_{2,1}$ is that an image can be represented with much less coefficients. Experimental results on three face databases demonstrate that the proposed $2DSFDA-L_{2,1}/BSFDA-L_{2,1}$ can obtain competitive performance.

• KNOM Review
• /
• 제24권2호
• /
• pp.24-34
• /
• 2021

엣지 컴퓨팅을 사용하는 서비스 공급업체는 높은 수준의 서비스를 제공한다. 이에 따라 다양하고 중요한 정보들이 단말 장치에 저장되면서 탐지하기 더욱 어려운 최신 사이버 공격의 핵심 목표가 됐다. 보안을 위해 침입 탐지시스템과 같은 보안 시스템이 자주 활용되지만, 기존의 침입 탐지 시스템은 탐지 정확도가 낮은 문제점이 존재한다. 따라서 본 논문에서는 엣지 컴퓨팅에서 단말 장치의 더욱 정확한 침입 탐지를 위한 기계 학습 모델을 제안한다. 제안하는 모델은 희소성 제약을 사용하여 입력 데이터의 중요한 특징 벡터들을 추출하는 stacked sparse autoencoder (SSAE)와 convolutional neural network (CNN)를 결합한 하이브리드 모델이다. 최적의 모델을 찾기 위해 SSAE의 희소성 계수를 조절하면서 모델의 성능을 비교 및 분석했다. 그 결과 희소성 계수가 일 때 96.9%로 가장 높은 정확도를 보여주었다. 따라서 모델이 중요한 특징들만 학습할 경우 더 높은 성능을 얻을 수 있었다.

• Journal of Computing Science and Engineering
• /
• 제10권4호
• /
• pp.103-110
• /
• 2016

Emotion recognition is an important component of affective computing, and is significant in the implementation of natural and friendly human-computer interaction. An effective approach to recognizing emotion from text is based on a machine learning technique, which deals with emotion recognition as a classification problem. However, in emotion recognition, the texts involved are usually very short, leaving a very large, sparse feature space, which decreases the performance of emotion classification. This paper proposes to resolve the problem of feature sparseness, and largely improve the emotion recognition performance from short texts by doing the following: representing short texts with word cluster features, offering a novel word clustering algorithm, and using a new feature weighting scheme. Emotion classification experiments were performed with different features and weighting schemes on a publicly available dataset. The experimental results suggest that the word cluster features and the proposed weighting scheme can partly resolve problems with feature sparseness and emotion recognition performance.