• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.9
• /
• pp.3583-3597
• /
• 2020

In this paper, a parallel deep learning model using a convolutional neural network and a dilated convolutional neural network is proposed to classify Alzheimer's disease with high accuracy in PET/CT images. The developed model consists of two pipelines, a conventional CNN pipeline, and a dilated convolution pipeline. An input image is sent through both pipelines, and at the end of both pipelines, extracted features are concatenated and used for classifying Alzheimer's disease. Complimentary abilities of both networks provide better overall accuracy than single conventional CNNs in the dataset. Moreover, instead of performing binary classification, the proposed model performs three-class classification being Alzheimer's disease, mild cognitive impairment, and normal control. Using the data received from Dong-a University, the model performs classification detecting Alzheimer's disease with an accuracy of up to 95.51%.

• The Journal of the Convergence on Culture Technology
• /
• v.9 no.6
• /
• pp.301-307
• /
• 2023

Deep learning requires building a deep neural network, collecting a large amount of training data, and then training the built neural network for a long time. If training does not proceed properly or overfitting occurs, training will fail. When using deep learning tools that have been developed so far, it takes a lot of time to collect training data and learn. However, due to the rapid advent of the mobile environment and the increase in sensor data, the demand for real-time deep learning technology that can dramatically reduce the time required for neural network learning is rapidly increasing. In this study, a real-time deep learning system was implemented using an Arduino system equipped with a fine dust sensor. In the implemented system, fine dust data is measured every 30 seconds, and when up to 120 are accumulated, learning is performed using the previously accumulated data and the newly accumulated data as a dataset. The neural network for learning was composed of one input layer, one hidden layer, and one output. To evaluate the performance of the implemented system, learning time and root mean square error (RMSE) were measured. As a result of the experiment, the average learning error was 0.04053796, and the average learning time of one epoch was about 3,447 seconds.

• 한국정보교육학회:학술대회논문집
• /
• 2021.08a
• /
• pp.385-391
• /
• 2021

In this paper, a case of deep learning artificial neural network education was studied for artificial intelligence literacy education for preservice teachers and incumbent teachers. In addition, through the proposed educational case, we tried to explore the contents of artificial neural network principle education that elementary, middle and high school students can experience. To this end, first, an example of training on the principle of operation of an artificial neural network that recognizes two types of images is presented. And as an artificial neural network extension application education case, an artificial neural network education case for recognizing three types of images was presented. The number of output layers was changed according to the number of images to be recognized by the artificial neural network, and the cases implemented in a spreadsheet were divided and explained. In addition, in order to experience the operation results of the artificial neural network, we presented the educational contents to directly write the learning data necessary for the artificial neural network of the supervised learning method. In this paper, the implementation of the artificial neural network and the recognition test results are visually presented using a spreadsheet.

• The Journal of the Convergence on Culture Technology
• /
• v.9 no.4
• /
• pp.487-494
• /
• 2023

In this study, performance comparison was performed on R and TensorFlow, which are free deep learning tools. In the experiment, six types of deep neural networks were built using each tool, and the neural networks were trained using the 10-year Korean temperature dataset. The number of nodes in the input layer of the constructed neural network was set to 10, the number of output layers was set to 5, and the hidden layer was set to 5, 10, and 20 to conduct experiments. The dataset includes 3600 temperature data collected from Gangnam-gu, Seoul from March 1, 2013 to March 29, 2023. For performance comparison, the future temperature was predicted for 5 days using the trained neural network, and the root mean square error (RMSE) value was measured using the predicted value and the actual value. Experiment results shows that when there was one hidden layer, the learning error of R was 0.04731176, and TensorFlow was measured at 0.06677193, and when there were two hidden layers, R was measured at 0.04782134 and TensorFlow was measured at 0.05799060. Overall, R was measured to have better performance. We tried to solve the difficulties in tool selection by providing quantitative performance information on the two tools to users who are new to machine learning.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.21 no.2
• /
• pp.189-194
• /
• 2020

This paper analyzes the gradient descent method, which is the one most used for learning neural networks. Learning means updating a parameter so the loss function is at its minimum. The loss function quantifies the difference between actual and predicted values. The gradient descent method uses the slope of the loss function to update the parameter to minimize error, and is currently used in libraries that provide the best deep learning algorithms. However, these algorithms are provided in the form of a black box, making it difficult to identify the advantages and disadvantages of various gradient descent methods. This paper analyzes the characteristics of the stochastic gradient descent method, the momentum method, the AdaGrad method, and the Adadelta method, which are currently used gradient descent methods. The experimental data used a modified National Institute of Standards and Technology (MNIST) data set that is widely used to verify neural networks. The hidden layer consists of two layers: the first with 500 neurons, and the second with 300. The activation function of the output layer is the softmax function, and the rectified linear unit function is used for the remaining input and hidden layers. The loss function uses cross-entropy error.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.23 no.8
• /
• pp.975-980
• /
• 2019

In this paper, we propose a deep learning-based scheduling scheme for user selection in multi-user multi-antenna networks which is considered one of key technologies for the next generation mobile communication systems. We obtained 90,000 data samples from the conventional optimal scheme to train the proposed neural network and verified the trained neural network to check if the trained neural network is over-fitted. Although the proposed neural network-based scheduling algorithm requires considerable complexity and time for training in the initial stage, it does not cause any extra complexity once it has been trained successfully. On the other hand, the conventional optimal scheme continuously requires the same complexity of computations for every scheduling. According to extensive computer-simulations, the proposed deep learning-based scheduling algorithm yields about 88~96% average sum-rates of the conventional scheme for SNRs lower than 10dB, while it can achieve optimal average sum-rates for SNRs higher than 10dB.

• International Journal of Computer Science & Network Security
• /
• v.22 no.3
• /
• pp.61-66
• /
• 2022

Object classification is one of the main fields in neural networks and has attracted the interest of many researchers. Although there have been vast advancements in this area, still there are many challenges that are faced even in the current era due to its inefficiency in handling large data, linguistic and dimensional complexities. Powerful hardware and software approaches in Neural Networks such as Deep Neural Networks present efficient mechanisms and contribute a lot to the field of object recognition as well as to handle time series classification. Due to the high rate of accuracy in terms of prediction rate, a neural network is often preferred in applications that require identification, segmentation, and detection based on features. Neural networks self-learning ability has revolutionized computing power and has its application in numerous fields such as powering unmanned self-driving vehicles, speech recognition, etc. In this paper, the experiment is conducted to implement a neural approach to identify numbers in different formats without human intervention. Measures are taken to improve the efficiency of the machines to classify and identify numbers. Experimental results show the importance of having training sets to achieve better recognition accuracy.

• Journal of Information Processing Systems
• /
• v.13 no.2
• /
• pp.204-214
• /
• 2017

Artificial intelligence, especially deep learning technology, is penetrating the majority of research areas, including the field of bioinformatics. However, deep learning has some limitations, such as the complexity of parameter tuning, architecture design, and so forth. In this study, we analyze these issues and challenges in regards to its applications in bioinformatics, particularly genomic analysis and medical image analytics, and give the corresponding approaches and solutions. Although these solutions are mostly rule of thumb, they can effectively handle the issues connected to training learning machines. As such, we explore the tendency of deep learning technology by examining several directions, such as automation, scalability, individuality, mobility, integration, and intelligence warehousing.

• Genomics & Informatics
• /
• v.17 no.2
• /
• pp.15.1-15.7
• /
• 2019

Automatically detecting mentions of pharmaceutical drugs and chemical substances is key for the subsequent extraction of relations of chemicals with other biomedical entities such as genes, proteins, diseases, adverse reactions or symptoms. The identification of drug mentions is also a prior step for complex event types such as drug dosage recognition, duration of medical treatments or drug repurposing. Formally, this task is known as named entity recognition (NER), meaning automatically identifying mentions of predefined entities of interest in running text. In the domain of medical texts, for chemical entity recognition (CER), techniques based on hand-crafted rules and graph-based models can provide adequate performance. In the recent years, the field of natural language processing has mainly pivoted to deep learning and state-of-the-art results for most tasks involving natural language are usually obtained with artificial neural networks. Competitive resources for drug name recognition in English medical texts are already available and heavily used, while for other languages such as Spanish these tools, although clearly needed were missing. In this work, we adapt an existing neural NER system, NeuroNER, to the particular domain of Spanish clinical case texts, and extend the neural network to be able to take into account additional features apart from the plain text. NeuroNER can be considered a competitive baseline system for Spanish drug and CER promoted by the Spanish national plan for the advancement of language technologies (Plan TL).

• Journal of Semiconductor Engineering
• /
• v.2 no.2
• /
• pp.130-135
• /
• 2021

Recently, deep neural networks (DNNs) are actively used for action control so that an autonomous system, such as the robot, can perform human-like behaviors and operations. Unlike recognition tasks, the real-time operation is essential in action control, and it is too slow to use remote learning on a server communicating through a network. New learning techniques, such as reinforcement learning (RL), are needed to determine and select the correct robot behavior locally. In this paper, we propose an energy-efficient DNN processor with a LUT-based processing engine and near-zero skipper. A CNN-based facial emotion recognition and an RNN-based emotional dialogue generation model is integrated for natural HRI system and tested with the proposed processor. It supports 1b to 16b variable weight bit precision with and 57.6% and 28.5% lower energy consumption than conventional MAC arithmetic units for 1b and 16b weight precision. Also, the near-zero skipper reduces 36% of MAC operation and consumes 28% lower energy consumption for facial emotion recognition tasks. Implemented in 65nm CMOS process, the proposed processor occupies 1784×1784 um2 areas and dissipates 0.28 mW and 34.4 mW at 1fps and 30fps facial emotion recognition tasks.