• The Journal of the Acoustical Society of Korea
• /
• v.18 no.8
• /
• pp.94-99
• /
• 1999

It is the hybrid structure of HMM and neural network(NN) that shows high recognition rate in speech recognition algorithms. And it is a method which has majorities of statistical model and neural network model respectively. In this study, we propose a new style of the hybrid structure of semi-continuous HMM(SCHMM) and radial basis function(RBF), which re-estimates weighting coefficients probability affecting observation probability after Baum-Welch estimation. The proposed method takes account of the similarity of basis Auction of RBF's hidden layer and SCHMM's probability density functions so as to discriminate speech signals sensibly through the learned and estimated weighting coefficients of RBF. As simulation results show that the recognition rates of the hybrid structure SCHMM/RBF are higher than those of SCHMM in unlearned speakers' recognition experiment, the proposed method has been proved to be one which has more sensible property in recognition than SCHMM.

• Journal of Digital Contents Society
• /
• v.18 no.1
• /
• pp.183-190
• /
• 2017

This paper proposes an audio event detection method using Deep Neural Networks (DNN). The proposed method applies Feed Forward Neural Network (FFNN) to generate output probabilities of twenty audio events for each frame. Mel scale filter bank (FBANK) features are extracted from each frame, and its five consecutive frames are combined as one vector which is the input feature of the FFNN. The output layer of FFNN produces audio event probabilities for each input feature vector. More than five consecutive frames of which event probability exceeds threshold are detected as an audio event. An audio event continues until the event is detected within one second. The proposed method achieves as 71.8% accuracy for 20 classes of the UrbanSound8K and the BBC Sound FX dataset.

• Journal of Internet Computing and Services
• /
• v.19 no.6
• /
• pp.101-112
• /
• 2018

A lot of manpower and budgets are being used to prevent fires, and only a small portion of the data generated during this process is used for disaster prevention activities. This study develops a prediction model of fire occurrence probability based on data mining in order to more actively use these data for disaster prevention activities. For this purpose, variables for predicting fire occurrence probability of various buildings were selected and data of construction administrative system, national fire information system, and Korea Fire Insurance Association were collected and integrated data set was constructed. After appropriate data cleansing and preprocessing, various data mining methodologies such as artificial neural network, decision trees, SVM, and Naive Bayesian were used to develop a prediction model of the fire occurrence probability of buildings. The most accurate model among the derived models is Linear SVM model which shows 68.42% as experimental data and 63.54% as verification data and it is the best model to predict fire occurrence probability of buildings. As this study develops the prediction model which uses only the set values of the specific ranges, future studies may explore more opportunites to use various setting values not shown in this study.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.24 no.12B
• /
• pp.2447-2454
• /
• 1999

This paper proposes a N-APPEM(Nonlinear A Posteriori Probability Estimation Method) with a frame normalization method to conventional classification network to increase speech recognition ability. It also tests the recognition ability of the classification and prediction neural networks for the Korean isolated digits. From the experimental results, the prediction network with MLP(Multi-Layer Perceptron) achieves the highest recognition ability of 98.0%. The prediction requires very complicated networks increased linearly with the number of incoming speech categories. However, the classification network with the N-APPEM and the normalization improves the recognition ability up to 85.5% with a sin81e network, which is almost 12.0% improvement.

• International Journal of Computer Science & Network Security
• /
• v.21 no.11
• /
• pp.89-96
• /
• 2021

This work is an extension of my work presented a robust and economically efficient method for the Discrimination of four Mung-Beans [1] varieties based on quantitative parameters. Due to the advancement of technology, users try to find the solutions to their daily life problems using smartphones but still for computing power and memory. Hence, there is a need to find the best classifier to classify the Mung-Beans using already suggested features in previous work with minimum memory requirements and computational power. To achieve this study's goal, we take the experiments on various supervised classifiers with simple architecture and calculations and give the robust performance on the most relevant 10 suggested features selected by Fisher Co-efficient, Probability of Error, Mutual Information, and wavelet features. After the analysis, we replace the Artificial Neural Network and Deep learning with a classifier that gives approximately the same classification results as the above classifier but is efficient in terms of resources and time complexity. This classifier is easily implemented in the smartphone environment.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.33 no.2
• /
• pp.175-182
• /
• 2023

Differential cryptanalysis is one of the analysis techniques for block ciphers, and uses the property that the output difference with respect to the input difference exists with a high probability. If random data and differential data can be distinguished, data complexity for differential cryptanalysis can be reduced. For this, many studies on deep learning-based neural distinguisher have been conducted. In this paper, a deep learning-based neural distinguisher for PIPO 64/128 is proposed. As a result of experiments with various input differences, the 3-round neural distinguisher for the differential characteristics for 0, 1, 3, and 5-rounds achieved accuracies of 0.71, 0.64, 0.62, and 0.64, respectively. This work allows distinguishing attacks for up to 8 rounds when used with the classical distinguisher. Therefore, scalability was achieved by finding a distinguisher that could handle the differential of each round. To improve performance, we plan to apply various neural network structures to construct an optimal neural network, and implement a neural distinguisher that can use related key differential or process multiple input differences simultaneously.

• The Journal of Korea Robotics Society
• /
• v.12 no.4
• /
• pp.425-431
• /
• 2017

As drones gain more popularity these days, drone detection becomes more important part of the drone systems for safety, privacy, crime prevention and etc. However, existing drone detection systems are expensive and heavy so that they are only suitable for industrial or military purpose. This paper proposes a novel approach for training Convolutional Neural Networks to detect drones from images that can be used in embedded systems. Unlike previous works that consider the class probability of the image areas where the class object exists, the proposed approach takes account of all areas in the image for robust classification and object detection. Moreover, a novel loss function is proposed for the CNN to learn more effectively from limited amount of training data. The experimental results with various drone images show that the proposed approach performs efficiently in real drone detection scenarios.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 1997.11a
• /
• pp.259-262
• /
• 1997

This paper presents a visual servoing of RV-M2 robot manipulators to track and grasp moving object, using pruned dynamic recurrent neural networks(DRNN). The object is stationary in the robot work space and the robot is tracking and grasping the object by using CCD camera mounted on the end-effector. In order to optimize the structure of DRNN, we decide the node whether delete or add, by mutation probability, first in case of delete node, the node which have minimum sum of input weight is actually deleted, and then in case of add node, the weight is connected according to the number of case which added node can reach the other nodes. Using evolutionary programming(EP) that search the struture and weight of the DRNN, and evolution strategies(ES) which train the weight of neuron, we pruned the net structure of DRNN. We applied the DRNN to the Visual Servoing of a robot manipulators to control position and orientation of end-effector, and the validity and effectiveness of the pro osed control scheme will be verified by computer simulations.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.10 no.3
• /
• pp.270-279
• /
• 2000

This paper describes the on-chip learning algorithm of neural networks using the stochastic pulse arithmetic. Stochastic pulse arithmetic is the computation using the numbers represented by the probability of 1' and 0's occurrences in a random pulse stream. This stochastic arithmetic has the merits when applied to neural network ; reduction of the area of the implemented hardware and getting a global solution escaping from local minima by virtue of the stochastic characteristics. And in this study, the on-chip learning algorithm is derived from the backpropagation algorithm for effective hardware implementation. We simulate the nonlinear separation problem of the some character patterns to verify the proposed learning algorithm. We also had good results after applying this algorithm to recognize printed and handwritten numbers.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.3
• /
• pp.1450-1463
• /
• 2019

The problem towards crowd behavior recognition in a serious clustered scene is extremely challenged on account of variable scales with non-uniformity. This paper aims to propose a crowed behavior classification framework based on a transferring hybrid network blending 3D res-net with inception-v3. First, the 3D res-inception network is presented so as to learn the augmented visual feature of UCF 101. Then the target dataset is applied to fine-tune the network parameters in an attempt to classify the behavior of densely crowded scenes. Finally, a transferred entropy function is used to calculate the probability of multiple labels in accordance with these features. Experimental results show that the proposed method could greatly improve the accuracy of crowd behavior recognition and enhance the accuracy of multiple label classification.