• ETRI Journal
• /
• v.39 no.3
• /
• pp.398-405
• /
• 2017

In this paper, an artificial stereo extension method that creates stereophonic sound from a mono sound source is proposed. The proposed method first trains deep neural networks (DNNs) that model the nonlinear relationship between the dominant and residual signals of the stereo channel. In the training stage, the band-wise log spectral magnitude and unwrapped phase of both the dominant and residual signals are utilized to model the nonlinearities of each sub-band through deep architecture. From that point, stereo extension is conducted by estimating the residual signal that corresponds to the input mono channel signal with the trained DNN model in a sub-band domain. The performance of the proposed method was evaluated using a log spectral distortion (LSD) measure and multiple stimuli with a hidden reference and anchor (MUSHRA) test. The results showed that the proposed method provided a lower LSD and higher MUSHRA score than conventional methods that use hidden Markov models and DNN with full-band processing.

• Journal of Korea Multimedia Society
• /
• v.20 no.8
• /
• pp.1406-1420
• /
• 2017

Artificial neural networks (ANNs) play an important role in the fields of function approximation, prediction, and classification. ANN performance is critically dependent on the input parameters, including the number of neurons in each layer, and the optimal values of weights and biases assigned to each neuron. In this study, we apply the particle swarm optimization method, a popular optimization algorithm for determining the optimal values of weights and biases for every neuron in different layers of the ANN. Several regression models, including general linear regression, Fourier regression, smoothing spline, and polynomial regression, are conducted to evaluate the proposed method's prediction power compared to multiple linear regression (MLR) methods. In addition, residual analysis is conducted to evaluate the optimized ANN accuracy for both training and test datasets. The experimental results demonstrate that the proposed method can effectively determine optimal values for neuron weights and biases, and high accuracy results are obtained for prediction applications. Evaluations of the proposed method reveal that it can be used for prediction and estimation purposes, with a high accuracy ratio, and the designed model provides a reliable technique for optimization. The simulation results show that the optimized ANN exhibits superior performance to MLR for prediction purposes.

• International Journal of Computer Science & Network Security
• /
• v.22 no.7
• /
• pp.301-307
• /
• 2022

Energy consumption has grown alongside dramatic population increases. Statistics show that buildings in particular utilize a significant amount of energy, worldwide. Because of this, building energy prediction is crucial to best optimize utilities' energy plans and also create a predictive model for consumers. To improve energy prediction performance, this paper proposes a ResNet-LSTM model that combines residual networks (ResNets) and long short-term memory (LSTM) for energy consumption prediction. ResNets are utilized to extract complex and rich features, while LSTM has the ability to learn temporal correlation; the dense layer is used as a regression to forecast energy consumption. To make our model more robust, we employed Huber loss during the optimization process. Huber loss obtains high efficiency by handling minor errors quadratically. It also takes the absolute error for large errors to increase robustness. This makes our model less sensitive to outlier data. Our proposed system was trained on historical data to forecast energy consumption for different time series. To evaluate our proposed model, we compared our model's performance with several popular machine learning and deep learning methods such as linear regression, neural networks, decision tree, and convolutional neural networks, etc. The results show that our proposed model predicted energy consumption most accurately.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.19 no.4
• /
• pp.574-579
• /
• 2009

There have been much difficulties to construct an optimized neural network in complex nonlinear regression problems such as selecting the networks structure and avoiding overtraining problem generated by noise. In this paper, we propose a stepwise constructive method for neural networks using a flexible incremental algorithm. When the hidden nodes are added, the flexible incremental algorithm adaptively controls the number of hidden nodes by a validation dataset for minimizing the prediction residual error. Here, the ELM (Extreme Learning Machine) was used for fast training. The proposed neural network can be an universal approximator without user intervene in the training process, but also it has faster training and smaller number of hidden nodes. From the experimental results with various benchmark datasets, the proposed method shows better performance for real-world regression problems than previous methods.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.54 no.5
• /
• pp.35-41
• /
• 2017

In this paper, we analyze the performance of the recently introduced Hint-knowledge distillation (KD) training approach based on the teacher-student framework for knowledge distillation and knowledge transfer. As a deep neural network (DNN) considered in this paper, the deep residual network (ResNet), which is currently regarded as the latest DNN, is used for the teacher-student framework. Therefore, when implementing the Hint-KD training, we investigate the impact on the weight of KD information based on the soften factor in terms of classification accuracy using the widely used open deep learning frameworks, Caffe. As a results, it can be seen that the recognition accuracy of the student model is improved when the fixed value of the KD information is maintained rather than the gradual decrease of the KD information during training.

• Smart Media Journal
• /
• v.9 no.1
• /
• pp.23-29
• /
• 2020

This paper presents an approach for dynamic hand gesture recognition by using algorithm based on 3D Convolutional Neural Network (3D_CNN), which is later extended to 3D Residual Networks (3D_ResNet), and the neural network based key frame selection. Typically, 3D deep neural network is used to classify gestures from the input of image frames, randomly sampled from a video data. In this work, to improve the classification performance, we employ key frames which represent the overall video, as the input of the classification network. The key frames are extracted by SegNet instead of conventional clustering algorithms for video summarization (VSUMM) which require heavy computation. By using a deep neural network, key frame selection can be performed in a real-time system. Experiments are conducted using 3D convolutional kernels such as 3D_CNN, Inflated 3D_CNN (I3D) and 3D_ResNet for gesture classification. Our algorithm achieved up to 97.8% of classification accuracy on the Cambridge gesture dataset. The experimental results show that the proposed approach is efficient and outperforms existing methods.

• The Journal of Korea Robotics Society
• /
• v.13 no.1
• /
• pp.1-7
• /
• 2018

The image-to-image translation is one of the deep learning applications using image data. In this paper, we aim at improving the performance of object transfiguration which transforms a specific object in an image into another specific object. For object transfiguration, it is required to transform only the target object and maintain background images. In the existing results, however, it is observed that other parts in the image are also transformed. In this paper, we have focused on the structure of artificial neural networks that are frequently used in the existing methods and have improved the performance by adding constraints to the exiting structure. We also propose the advanced structure that combines the existing structures to maintain their advantages and complement their drawbacks. The effectiveness of the proposed methods are shown in experimental results.

• Smart Structures and Systems
• /
• v.22 no.2
• /
• pp.175-183
• /
• 2018

Time-series data often contain one of the most valuable pieces of information in many fields including manufacturing. Because time-series data are relatively cheap to acquire, they (e.g., vibration signals) have become a crucial part of big data even in manufacturing shop floors. Recently, deep-learning models have shown state-of-art performance for analyzing big data because of their sophisticated structures and considerable computational power. Traditional models for a machinery-monitoring system have highly relied on features selected by human experts. In addition, the representational power of such models fails as the data distribution becomes complicated. On the other hand, deep-learning models automatically select highly abstracted features during the optimization process, and their representational power is better than that of traditional neural network models. However, the applicability of deep-learning models to the field of prognostics and health management (PHM) has not been well investigated yet. This study integrates the "residual fitting" mechanism inherently embedded in the wavelet transform into the convolutional neural network deep-learning structure. As a result, the architecture combines a signal smoother and classification procedures into a single model. Validation results from rotor vibration data demonstrate that our model outperforms all other off-the-shelf feature-based models.

• Journal of Information Processing Systems
• /
• v.15 no.2
• /
• pp.305-319
• /
• 2019

The prediction of the sum of container is very important in the field of container transport. Many influencing factors can affect the prediction results. These factors are usually composed of many variables, whose composition is often very complex. In this paper, we use gray relational analysis to set up a proper forecast index system for the prediction of the sum of containers in foreign trade. To address the issue of the low accuracy of the traditional prediction models and the problem of the difficulty of fully considering all the factors and other issues, this paper puts forward a prediction model which is combined with a back-propagation (BP) neural networks and the support vector machine (SVM). First, it gives the prediction with the data normalized by the BP neural network and generates a preliminary forecast data. Second, it employs SVM for the residual correction calculation for the results based on the preliminary data. The results of practical examples show that the overall relative error of the combined prediction model is no more than 1.5%, which is less than the relative error of the single prediction models. It is hoped that the research can provide a useful reference for the prediction of the sum of container and related studies.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.1
• /
• pp.1-15
• /
• 2022

The MobileNetV3 is specially designed for mobile devices with limited memory and computing power. To reduce the network parameters and improve the network inference speed, a new lightweight network is proposed based on MobileNetV3. Firstly, to reduce the computation of residual blocks, a partial residual structure is designed by dividing the input feature maps into two parts. The designed partial residual structure is used to replace the residual block in MobileNetV3. Secondly, a dual-path feature extraction structure is designed to further reduce the computation of MobileNetV3. Different convolution kernel sizes are used in the two paths to extract feature maps with different sizes. Besides, a transition layer is also designed for fusing features to reduce the influence of the new structure on accuracy. The CIFAR-100 dataset and Image Net dataset are used to test the performance of the proposed partial residual structure. The ResNet based on the proposed partial residual structure has smaller parameters and FLOPs than the original ResNet. The performance of improved MobileNetV3 is tested on CIFAR-10, CIFAR-100 and ImageNet image classification task dataset. Comparing MobileNetV3, GhostNet and MobileNetV2, the improved MobileNetV3 has smaller parameters and FLOPs. Besides, the improved MobileNetV3 is also tested on CPU and Raspberry Pi. It is faster than other networks