• Journal of Information Processing Systems
• /
• v.18 no.2
• /
• pp.245-257
• /
• 2022

In the Internet of Things (IoT) era, the types of devices used by one user are becoming more diverse and the number of devices is also increasing. However, a forensic investigator is restricted to exploit or collect all the user's devices; there are legal issues (e.g., privacy, jurisdiction) and technical issues (e.g., computing resources, the increase in storage capacity). Therefore, in the digital forensics field, it has been a challenge to acquire information that remains on the devices that could not be collected, by analyzing the seized devices. In this study, we focus on the fact that multiple devices share data through account synchronization of the online platform. We propose a novel way of identifying the user's interest through analyzing the remnants of targeted advertising which is provided based on the visited websites or search terms of logged-in users. We introduce a detailed methodology to pick out the targeted advertising from cache data and infer the user's interest using deep learning. In this process, an improved learning model considering the unique characteristics of advertisement is implemented. The experimental result demonstrates that the proposed method can effectively identify the user interest even though only one device is examined.

• Nuclear Engineering and Technology
• /
• v.53 no.2
• /
• pp.522-531
• /
• 2021

We describe a new deep learning model - Convolutional eXtreme Gradient Boosting (ConvXGB) for classification problems based on convolutional neural nets and Chen et al.'s XGBoost. As well as image data, ConvXGB also supports the general classification problems, with a data preprocessing module. ConvXGB consists of several stacked convolutional layers to learn the features of the input and is able to learn features automatically, followed by XGBoost in the last layer for predicting the class labels. The ConvXGB model is simplified by reducing the number of parameters under appropriate conditions, since it is not necessary re-adjust the weight values in a back propagation cycle. Experiments on several data sets from UCL Repository, including images and general data sets, showed that our model handled the classification problems, for all the tested data sets, slightly better than CNN and XGBoost alone and was sometimes significantly better.

• International Journal of Computer Science & Network Security
• /
• v.21 no.12spc
• /
• pp.526-538
• /
• 2021

Machine and deep learning-based models are emerging techniques that are being used to address prediction problems in biomedical data analysis. DNA sequence prediction is a critical problem that has attracted a great deal of attention in the biomedical domain. Machine and deep learning-based models have been shown to provide more accurate results when compared to conventional regression-based models. The prediction of the gene sequence that leads to cancerous diseases, such as prostate cancer, is crucial. Identifying the most important features in a gene sequence is a challenging task. Extracting the components of the gene sequence that can provide an insight into the types of mutation in the gene is of great importance as it will lead to effective drug design and the promotion of the new concept of personalised medicine. In this work, we extracted the exons in the prostate gene sequences that were used in the experiment. We built a Deep Neural Network (DNN) and Bi-directional Long-Short Term Memory (Bi-LSTM) model using a k-mer encoding for the DNA sequence and one-hot encoding for the class label. The models were evaluated using different classification metrics. Our experimental results show that DNN model prediction offers a training accuracy of 99 percent and validation accuracy of 96 percent. The bi-LSTM model also has a training accuracy of 95 percent and validation accuracy of 91 percent.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.17 no.6
• /
• pp.1075-1080
• /
• 2022

Recently, the effort is continuously applied in machine learning and deep learning algorithm which is represented as artificial intelligence algorithm in the varies field such as prediction, classification and clustering. In this paper, we propose detection algorithm for un-peeling status of PCB protection film by using Dectron2. We use 42 images of data as training and 19 images of data as testing based on 61 images which was taken under the condition of a critical reflection angel of 42.8°. As a result, we get 16 images that was detected and 3 images that was not detected among 19 images of testing data.

• Journal of the Korea Computer Graphics Society
• /
• v.27 no.5
• /
• pp.73-79
• /
• 2021

Recently, in the field of computer animation, a method for generating motion using deep learning has been studied away from conventional finite-state machines or graph-based methods. The expressiveness of the network required for learning motions is more influenced by the diversity of motion contained in it than by the simple length of motion to be learned. This study aims to find an efficient network structure when the types of motions to be learned are diverse. In this paper, we train and compare three types of networks: basic fully-connected structure, mixture of experts structure that uses multiple fully-connected layers in parallel, recurrent neural network which is widely used to deal with seq2seq, and transformer structure used for sequence-type data processing in the natural language processing field.

• Journal of Biomedical Engineering Research
• /
• v.43 no.2
• /
• pp.102-108
• /
• 2022

Purpose: To propose a deep learning-based clinical decision support technique for laryngeal disease on epiglottis, tongue and vocal cords. Materials and Methods: A total of 873 laryngeal endoscopic images were acquired from the PACS database of Pusan N ational University Yangsan Hospital. and VGG16 model was applied with transfer learning and fine-tuning. Results: The values of precision, recall, accuracy and F1-score for test dataset were 0.94, 0.97, 0.95 and 0.95 for epiglottis images, 0.91, 1.00, 0.95 and 0.95 for tongue images, and 0.90, 0.64, 0.73 and 0.75 for vocal cord images, respectively. Conclusion: Experimental results demonstrated that the proposed model have a potential as a tool for decision-supporting of otolaryngologist during manual inspection of laryngeal endoscopic images.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.11
• /
• pp.3780-3797
• /
• 2022

The estimation of radio map (RM) is a fundamental and critical task for the network planning and optimization performance of mobile communication. In this paper, a RM estimation method is proposed based on a deep dual learning structure. This method can simultaneously and accurately reconstruct the urban building map (UBM) and estimate the RM of the whole cell by only part of the measured reference signal receiving power (RSRP). Our proposed method implements UBM reconstruction task and RM estimation task by constructing a dual U-Net-based structure, which is named RadioCycle. RadioCycle jointly trains two symmetric generators of the dual structure. Further, to solve the problem of interference negative transfer in generators trained jointly for two different tasks, RadioCycle introduces a dynamic weighted averaging method to dynamically balance the learning rate of these two generators in the joint training. Eventually, the experiments demonstrate that on the UBM reconstruction task, RadioCycle achieves an F1 score of 0.950, and on the RM estimation task, RadioCycle achieves a root mean square error of 0.069. Therefore, RadioCycle can estimate both the RM and the UBM in a cell with measured RSRP for only 20% of the whole cell.

• International Journal of Internet, Broadcasting and Communication
• /
• v.14 no.1
• /
• pp.136-141
• /
• 2022

A very complex task in deep learning such as image classification must be solved with the help of neural networks and activation functions. The backpropagation algorithm advances backward from the output layer towards the input layer, the gradients often get smaller and smaller and approach zero which eventually leaves the weights of the initial or lower layers nearly unchanged, as a result, the gradient descent never converges to the optimum. We propose a two-factor non-saturating activation functions known as Bea-Mish for machine learning applications in deep neural networks. Our method uses two factors, beta (𝛽) and alpha (𝛼), to normalize the area below the boundary in the Mish activation function and we regard these elements as Bea. Bea-Mish provide a clear understanding of the behaviors and conditions governing this regularization term can lead to a more principled approach for constructing better performing activation functions. We evaluate Bea-Mish results against Mish and Swish activation functions in various models and data sets. Empirical results show that our approach (Bea-Mish) outperforms native Mish using SqueezeNet backbone with an average precision (AP50val) of 2.51% in CIFAR-10 and top-1accuracy in ResNet-50 on ImageNet-1k. shows an improvement of 1.20%.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.32 no.3
• /
• pp.487-499
• /
• 2022

Machine learning such as deep learning have been widely used in recent years. Recently deep learning is performed in a trusted execution environment such as ARM TrustZone to improve security in edge devices and embedded devices with low computing resource. To mitigate this problem, we propose TPMP that efficiently uses the limited memory of TEE through DNN model partitioning. TPMP achieves high confidentiality of DNN by performing DNN models that could not be run with existing memory scheduling methods in TEE through optimized memory scheduling. TPMP required a similar amount of computational resources to previous methodologies.

• International Journal of Computer Science & Network Security
• /
• v.22 no.12
• /
• pp.98-106
• /
• 2022

A condition of the retina known as hypertensive retinopathy (HR) is connected to high blood pressure. The severity and persistence of hypertension are directly correlated with the incidence of HR. To avoid blindness, it is essential to recognize and assess HR as soon as possible. Few computer-aided systems are currently available that can diagnose HR issues. On the other hand, those systems focused on gathering characteristics from a variety of retinopathy-related HR lesions and categorizing them using conventional machine-learning algorithms. Consequently, for limited applications, significant and complicated image processing methods are necessary. As seen in recent similar systems, the preciseness of classification is likewise lacking. To address these issues, a new CAD HR-diagnosis system employing the advanced Deep Dense CNN Learning (DD-CNN) technology is being developed to early identify HR. The HR-diagnosis system utilized a convolutional neural network that was previously trained as a feature extractor. The statistical investigation of more than 1400 retinography images is undertaken to assess the accuracy of the implemented system using several performance metrics such as specificity (SP), sensitivity (SE), area under the receiver operating curve (AUC), and accuracy (ACC). On average, we achieved a SE of 97%, ACC of 98%, SP of 99%, and AUC of 0.98. These results indicate that the proposed DD-CNN classifier is used to diagnose hypertensive retinopathy.