• International journal of advanced smart convergence
• /
• v.9 no.4
• /
• pp.115-119
• /
• 2020

Recently, CCTV installations are rapidly increasing in the public and private sectors to prevent various crimes. In accordance with the increasing number of CCTVs, video-based abnormal behavior detection in control systems is one of the key technologies for safety. This is because it is difficult for the surveillance personnel who control multiple CCTVs to manually monitor all abnormal behaviors in the video. In order to solve this problem, research to recognize abnormal behavior using deep learning is being actively conducted. In this paper, we propose a model for detecting abnormal behavior based on the deep learning model that is currently widely used. Based on the abnormal behavior video data provided by AI Hub, we performed a comparative experiment to detect anomalous behavior through violence learning and fainting in videos using 2D CNN-LSTM, 3D CNN, and I3D models. We hope that the experimental results of this abnormal behavior learning model will be helpful in developing intelligent CCTV.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2022.05a
• /
• pp.370-373
• /
• 2022

Recently, various deep learning framework models such as Tensorflow, Pytorch, Keras, etc. have appeared. In addition, CNN (Convolutional Neural Network) is applied to image recognition using frameworks such as Tensorflow, Pytorch, and Keras, and the optimization model in image classification is mainly used. In this paper, based on the results of training the CNN model with the Paitotchi and tensor flow frameworks most often used in the field of deep learning image recognition, the two frameworks are compared and analyzed for image analysis. Derived an optimized framework.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2021.05a
• /
• pp.72-74
• /
• 2021

The deep learning process currently utilized in various fields consists of data preparation, data preprocessing, model generation, model learning, and model evaluation. In the process of model learning, the loss function compares the value of the model with the actual value and outputs the difference. In this paper, we analyze various loss functions used in the deep learning model for biomarker extraction, which measure the degree of loss of neural network output values, and try to find the best loss function through experiments.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.30 no.3
• /
• pp.337-347
• /
• 2020

In CRYPTO 2019, Gohr presents that Deep-learning can be used for cryptanalysis. In this paper, we verify whether Deep-learning can identify the structures of S-box. To this end, we conducted two experiments. First, we use DDT and LAT of S-boxes as the learning data, whose structure is one of mainly used S-box structures including Feistel, MISTY, SPN, and multiplicative inverse. Surprisingly, our Deep-learning algorithms can identify not only the structures but also the number of used rounds. The second application verifies the pseudo-randomness of and structures by increasing the nuber of rounds in each structure. Our Deep-learning algorithms outperform the theoretical distinguisher in terms of the number of rounds. In general, the design rationale of ciphers used for high level of confidentiality, such as for military purposes, tends to be concealed in order to interfere cryptanalysis. The methods presented in this paper show that Deep-learning can be utilized as a tool for analyzing such undisclosed design rationale.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.33 no.6
• /
• pp.1055-1065
• /
• 2023

Recently, various studies using deep reinforcement learning (deep RL) technology have been conducted to solve complex problems using big data collected at industrial internet of things. Deep RL uses reinforcement learning"s trial-and-error algorithms and cumulative compensation functions to generate and learn its own data and quickly explore neural network structures and parameter decisions. However, studies so far have shown that the larger the size of the learning data is, the higher are the memory usage and search time, and the lower is the accuracy. In this study, model-agnostic learning for efficient federated deep RL was utilized to solve privacy invasion by increasing robustness as 55.9% and achieve 97.8% accuracy, an improvement of 5.5% compared with the comparative optimization-based meta learning models, and to reduce the delay time by 28.9% on average.

• Journal of Advanced Navigation Technology
• /
• v.25 no.6
• /
• pp.569-575
• /
• 2021

Deep learning shows differences in prediction performance depending on data quality and model. This study uses various input data and multiple deep learning models to build an optimal deep learning model for predicting solar radiation, which has the most influence on power generation prediction. did. As the input data, the weather data of the Korea Meteorological Administration and the clairvoyant meteorological image were used by segmenting the image of the Korea Meteorological Agency. , comparative evaluation, and predicting solar radiation by constructing multiple deep learning models connecting the models with the best error rate in each model. As an experimental result, the RMSE of model A, which is a multiple deep learning model, was 0.0637, the RMSE of model B was 0.07062, and the RMSE of model C was 0.06052, so the error rate of model A and model C was better than that of a single model. In this study, the model that connected two or more models through experiments showed improved prediction rates and stable learning results.

• Electronics and Telecommunications Trends
• /
• v.33 no.4
• /
• pp.54-60
• /
• 2018

In this paper, we analyze the trends of deep-learning based plant data processing technologies. In recent years, the deep-learning technology has been widely applied to various AI tasks, such as vision (image classification, image segmentation, and so on) and natural language processing because it shows a higher performance on such tasks. The deep-leaning method is also applied to plant data processing tasks and shows a significant performance. We analyze and show how the deep-learning method is applied to plant data processing tasks and related industries.

• Electronics and Telecommunications Trends
• /
• v.33 no.5
• /
• pp.103-110
• /
• 2018

Studies on artificial intelligence have been developed for the past couple of decades. After a few periods of prosperity and recession, a new machine learning method, so-called Deep Learning, has been introduced. This is the result of high-quality big- data, an increase in computing power, and the development of new algorithms. The main targets for deep learning are 1D audio and 2D images. The application domain is being extended from a discriminative model, such as classification/segmentation, to a generative model. Currently, deep learning is used for processing 3D data. However, unlike 2D, it is not easy to acquire 3D learning data. Although low-cost 3D data acquisition sensors have become more popular owing to advances in 3D vision technology, the generation/acquisition of 3D data remains a very difficult problem. Moreover, it is not easy to directly apply an existing network model, such as a convolution network, owing to the variety of 3D data representations. In this paper, we summarize the 3D deep learning technology that have started to be developed within the last 2 years.

• 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.

• Journal of Internet Computing and Services
• /
• v.21 no.6
• /
• pp.33-39
• /
• 2020

The paper is to build recommendation systems leveraging Deep Learning and Big Data platform, Spark to predict item ratings of the Amazon e-commerce site. Recommendation system in e-commerce has become extremely popular in recent years and it is very important for both customers and sellers in daily life. It means providing the users with products and services they are interested in. Therecommendation systems need users' previous shopping activities and digital footprints to make best recommendation purpose for next item shopping. We developed the recommendation models in Amazon AWS Cloud services to predict the users' ratings for the items with the massive data set of Amazon customer reviews. We also present Big Data architecture to afford the large scale data set for storing and computation. And, we adopted deep learning for machine learning community as it is known that it has higher accuracy for the massive data set. In the end, a comparative conclusion in terms of the accuracy as well as the performance is illustrated with the Deep Learning architecture with Spark ML and the traditional Big Data architecture, Spark ML alone.