• Phonetics and Speech Sciences
• /
• v.2 no.4
• /
• pp.145-151
• /
• 2010

In this paper, a robust feature compensation method is proposed for improving the performance of speech recognition. The proposed method is incorporated into the auto-regressive moving average (ARMA) based feature compensation. We employ variable weights for the ARMA filter according to the degree of speech activity, and pass the normalized cepstral sequence through the weighted ARMA filter. Additionally when normalizing the cepstral sequences in training, the cepstral means and variances are estimated from total training utterances. Experimental results show the proposed method significantly improves the speech recognition performance in the noisy and reverberant environments.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2002.12a
• /
• pp.179-182
• /
• 2002

In this paper, this system makes use of the analog sensor and converts the feature of fish analog signal when sensor is operating with CPU(80C196KC). Then, After signal processing, this feature Is classified a special feature and a outline of fish by using the neural network, one of the artificial intelligence scheme. This neural network classifies fish pattern of very simple and short calculation. This has linear activation function and the error backpropagation is used as a learning algorithm. And the neural network is learned in off-line process. Because an adaptation period of neural network is too long time when random initial weights are used, off-line learning Is induced to decrease the Progress time We confirmed this method has better performance than somewhat outdated machines.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.11 no.10
• /
• pp.5023-5038
• /
• 2017

Big data is an emerging technology which deals with wide range of data sets with sizes beyond the ability to work with software tools which is commonly used for processing of data. When we consider a huge network, we have to process a large amount of network information generated, which consists of both normal and abnormal activity logs in large volume of multi-dimensional data. Intrusion Detection System (IDS) is required to monitor the network and to detect the malicious nodes and activities in the network. Massive amount of data makes it difficult to detect threats and attacks. Sequential Pattern mining may be used to identify the patterns of malicious activities which have been an emerging popular trend due to the consideration of quantities, profits and time orders of item. Here we propose a sequential pattern mining algorithm with fuzzy logic feature selection and fuzzy weighted support for huge volumes of network logs to be implemented in Apache Hadoop YARN, which solves the problem of speed and time constraints. Fuzzy logic feature selection selects important features from the feature set. Fuzzy weighted supports provide weights to the inputs and avoid multiple scans. In our simulation we use the attack log from NS-2 MANET environment and compare the proposed algorithm with the state-of-the-art sequential Pattern Mining algorithm, SPADE and Support Vector Machine with Hadoop environment.

• The Journal of the Acoustical Society of Korea
• /
• v.13 no.6
• /
• pp.83-90
• /
• 1994

In this paper, we propose a new learning algorithm, ASOFM(Adaptive Self Organizing Feature Map), to solve the defects of Kohonen's Self Organiaing Feature Map. Kohonen's algorithm is sometimes stranded on local minima for the initial weights. The proposed algorithm uses an object function which can evaluate the state of network in learning and adjusts the learning rate adaptively according to the evaluation of the object function. As a result, it is always guaranteed that the state of network is converged to the global minimum value and it has a capacity of generalized learning by adaptively. It is reduce that the learning time of our algorithm is about $30\%$ of Kohonen's.

• Journal of the Korean Society for information Management
• /
• v.18 no.2
• /
• pp.203-230
• /
• 2001

The purpose of this study is to develop a document clustering model for automatic classification of knowledge. Two test collections of newspaper article texts and journal article abstracts are built for the clustering experiment. Various feature reduction criteria as well as term weighting methods are applied to the term sets of the test collections, and cosine and Jaccard coefficients are used as similarity measures. The performances of complete linkage and K-means clustering algorithms are compared using different feature selection methods and various term weights. It was found that complete linkage clustering outperforms K-means algorithm and feature reduction up to almost 10% of the total feature sets does not lower the performance of document clustering to any significant extent.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.42 no.4 s.304
• /
• pp.43-50
• /
• 2005

This paper proposes a new pattern recognition system combining the new adaptive feature weighting based on the genetic algorithm and the modified KNN(K Nearest-Neighbor) rules. The new feature weighting proposed herein avoids the overfitting and finds the Proper feature weighting value by determining the middle value of weights using GA. New GA operators are introduced to obtain the high performance of the system. Moreover, a class dependent feature weighting strategy is employed. Whilst the classical methods use the same feature space for all classes, the Proposed method uses a different feature space for each class. The KNN rule is modified to estimate the class of test pattern using adaptive feature space. Experiments were performed with the unconstrained handwritten numeral database of Concordia University in Canada to show the performance of the proposed method.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.46 no.1
• /
• pp.121-127
• /
• 2009

The Multistage Feature-based Classification Model(MFCM) is proposed in this paper. MFCM does not use whole feature vectors extracted from the original data at once to classify each data, but use only groups related to each feature vector to classify separately. In the training stage, the contribution rate calculated from each feature vector group is drew throughout the accuracy of each feature vector group and then, in the testing stage, the final classification result is obtained by applying weights corresponding to the contribution rate of each feature vector group. In this paper, the proposed MFCM algorithm is applied to the problem of music genre classification. The results demonstrate that the proposed MFCM outperforms conventional algorithms by 7% - 13% on average in terms of classification accuracy.

• KIPS Transactions on Software and Data Engineering
• /
• v.6 no.3
• /
• pp.149-154
• /
• 2017

The MCT-Adaboost algorithm chooses an optimal set of features in each rounds. On each round, it chooses the best feature by calculate minimizing error rate using feature index and MCT kernel distribution. The involved process of weak classification executed by a hard decision. This decision occurs some problems when it chooses ambiguous kernel feature. In this paper, we propose the modified MCT-Adaboost classification using soft decision. The typical MCT-Adaboost assigns a same initial weights to each datum. This is because, they assume that all information of database is blind. We assign different initial weights with our propose new algorithm using some statistical properties of involved features. In experimental results, we confirm that our method shows better performance than the traditional one.

• Journal of the Korea Society of Computer and Information
• /
• v.14 no.11
• /
• pp.73-79
• /
• 2009

In this paper, a method is proposed that enhances the performance of the facial expression recognition using template matching of Local Binary Pattern(LBP) histogram. In this method, the face image is segmented into blocks, and the LBP histogram is constructed to be used as the feature of the block. Block dissimilarity is calculated between a block of input image and the corresponding block of the model image. Image dissimilarity is defined as the weighted sum of the block dissimilarities. In conventional methods, the block weights are assigned by intuition. In this paper a new method is proposed that optimizes the weights from training samples. An experiment shows the recognition rate is enhanced by the proposed method.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.18 no.5
• /
• pp.217-223
• /
• 2023

In this paper, we propose a new model for the conditional generation of music, considering key and rhythm, fundamental elements of music. MIDI sheet music is converted into a WAV format, which is then transformed into a Mel Spectrogram using the Short-Time Fourier Transform (STFT). Using this information, key and rhythm details are classified by passing through two Convolutional Neural Networks (CNNs), and this information is again fed into the Music Transformer. The key and rhythm details are combined by differentially multiplying the weights and the embedding vectors of the MIDI events. Several experiments are conducted, including a process for determining the optimal weights. This research represents a new effort to integrate essential elements into music generation and explains the detailed structure and operating principles of the model, verifying its effects and potentials through experiments. In this study, the accuracy for rhythm classification reached 94.7%, the accuracy for key classification reached 92.1%, and the Negative Likelihood based on the weights of the embedding vector resulted in 3.01.