• International Journal of Contents
• /
• v.15 no.4
• /
• pp.8-15
• /
• 2019

This paper presents a solution of the 'Quick, Draw! Doodle Recognition Challenge' hosted by Google. Doodles are drawings comprised of concrete representational meaning or abstract lines creatively expressed by individuals. In this challenge, a doodle is presented as a sequence of sketches. From the view of at the sketch level, to learn the pattern of strokes representing a doodle, we propose a sequential model stacked with multiple convolution layers and Long Short-Term Memory (LSTM) cells following the attention mechanism [15]. From the view at the image level, we use multiple models pre-trained on ImageNet to recognize the doodle. Finally, an ensemble and a post-processing method using the minimum cost flow algorithm are introduced to combine multiple models in achieving better results. In this challenge, our solutions garnered 11th place among 1,316 teams. Our performance was 0.95037 MAP@3, only 0.4% lower than the winner. It demonstrates that our method is very competitive. The source code for this competition is published at: https://github.com/ngxbac/Kaggle-QuickDraw.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.43 no.4 s.310
• /
• pp.37-45
• /
• 2006

In this paper, we propose the development of MHLA(Modulatory Hippocampus Learning Algorithm) which remodel a principle of brain of hippocampus. Hippocampus takes charge auto-associative memory and controlling functions of long-term or short-term memory strengthening. We organize auto-associative memory based 3 steps system(DG, CA3, CAl) and improve speed of learning by addition of modulator to long-term memory learning. In hippocampal system, according to the 3 steps order, information applies statistical deviation on Dentate Gyrus region and is labelled to responsive pattern by adjustment of a good impression. In CA3 region, pattern is reorganized by auto-associative memory. In CAI region, convergence of connection weight which is used long-term memory is learned fast by neural networks which is applied modulator. To measure performance of MHLA, PCA(Principal Component Analysis) is applied to face images which are classified by pose, expression and picture quality. Next, we calculate feature vectors and learn by MHLA. Finally, we confirm cognitive rate. The results of experiments, we can compare a proposed method of other methods, and we can confirm that the proposed method is superior to the existing method.

• Journal of Digital Contents Society
• /
• v.19 no.4
• /
• pp.789-799
• /
• 2018

This article presents machine learning based approach on Big data to analyzing time series data for anomaly detection in such industrial complex system. Long Short-Term Memory (LSTM) network have been demonstrated to be improved version of RNN and have become a useful aid for many tasks. This LSTM based model learn the higher level temporal features as well as temporal pattern, then such predictor is used to prediction stage to estimate future data. The prediction error is the difference between predicted output made by predictor and actual in-coming values. An error-distribution estimation model is built using a Gaussian distribution to calculate the anomaly in the score of the observation. In this manner, we move from the concept of a single anomaly to the idea of the collective anomaly. This work can assist the monitoring and management of Smart Factory in minimizing failure and improving manufacturing quality.

• Journal of KIISE
• /
• v.44 no.6
• /
• pp.607-612
• /
• 2017

This paper presents an attempt to apply Deep Learning technology to solve the problem of forecasting floods in urban areas. We employ Recurrent Neural Networks (RNNs), which are suitable for analyzing time series data, to learn observed data of river water and to predict the water level. To test the model, we use water observation data of a station in the Trinity river, Texas, the U.S., with data from 2013 to 2015 for training and data in 2016 for testing. Input of the neural networks is a 16-record-length sequence of 15-minute-interval time-series data, and output is the predicted value of the water level at the next 30 minutes and 60 minutes. In the experiment, we compare three Deep Learning models including standard RNN, RNN trained with Back Propagation Through Time (RNN-BPTT), and Long Short-Term Memory (LSTM). The prediction quality of LSTM can obtain Nash Efficiency exceeding 0.98, while the standard RNN and RNN-BPTT also provide very high accuracy.

• Journal of Korean Society for Quality Management
• /
• v.49 no.4
• /
• pp.581-594
• /
• 2021

Purpose: The purpose of this study is to compare machine learning models for anomaly detection of mechanical facility equipment and suggest an anomaly detection system for mechanical facility equipment in subway stations. It helps to predict failures and plan the maintenance of facility. Ultimately it aims to improve the quality of facility equipment. Methods: The data collected from Daejeon Metropolitan Rapid Transit Corporation was used in this experiment. The experiment was performed using Python, Scikit-learn, tensorflow 2.0 for preprocessing and machine learning. Also it was conducted in two failure states of the equipment. We compared and analyzed five unsupervised machine learning models focused on model Long Short-Term Memory Variational Autoencoder(LSTM-VAE). Results: In both experiments, change in vibration and current data was observed when there is a defect. When the rotating body failure was happened, the magnitude of vibration has increased but current has decreased. In situation of axis alignment failure, both of vibration and current have increased. In addition, model LSTM-VAE showed superior accuracy than the other four base-line models. Conclusion: According to the results, model LSTM-VAE showed outstanding performance with more than 97% of accuracy in the experiments. Thus, the quality of mechanical facility equipment will be improved if the proposed anomaly detection system is established with this model used.

• Korean Journal of Cognitive Science
• /
• v.20 no.4
• /
• pp.535-554
• /
• 2009

Information collected on mobile devices might be utilized to support user's memory, but it is difficult to effectively retrieve them because of the enormous amount of information. In order to organize information as an episodic approach that mimics human memory for the effective search, it is required to detect important event like landmarks. For providing new services with users, in this paper, we propose the prediction model to find landmarks automatically from various context log information based on attributed Bayesian networks. The data are divided into daily and weekly ones, and are categorized into attributes according to the source, to learn the Bayesian networks for the improvement of landmark prediction. The experiments on the Nokia log data showed that the Bayesian method outperforms SVMs, and the proposed attributed Bayesian networks are superior to the Bayesian networks modelled daily and weekly.

• Journal of KIISE:Software and Applications
• /
• v.37 no.7
• /
• pp.558-567
• /
• 2010

There is a large body of ongoing research to develop efficient prediction methods for data streams. These methods provide single prediction with a fixed time interval. It is necessary to develop a method for multi-interval prediction (MIP) because different prediction results may be obtained based on different intervals in many cases. In this paper, we propose a solution for MIP based on the Hierarchical Temporal Memory (HTM) model. In order to solve the problem of MIP with HTM, we present an Integrated Hierarchical Temporal Memory (IHTM) network by introducing a new node type Zeta1LastNode to the original HTM network. Using the hierarchical characteristic of the IHTM network, different levels in the network learn and model the features of a data stream with different intervals and generate prediction results for different intervals. Performance evaluation shows that the IHTM is efficient in the memory and time consumption compared with the original HTM network in MIP.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.6
• /
• pp.2444-2463
• /
• 2020

Significant detection of video can more rationally allocate computing resources and reduce the amount of computation to improve accuracy. Deep learning can extract the edge features of the image, providing technical support for video saliency. This paper proposes a new detection method. We combine the Convolutional Neural Network (CNN) and the Deep Bidirectional LSTM Network (DB-LSTM) to learn the spatio-temporal features by exploring the object motion information and object motion information to generate video. A continuous frame of significant images. We also analyzed the sample database and found that human attention and significant conversion are time-dependent, so we also considered the significance detection of video cross-frame. Finally, experiments show that our method is superior to other advanced methods.

• English Language & Literature Teaching
• /
• no.6
• /
• pp.1-24
• /
• 2000

Listening comprehension is an integrative and creative process of interaction through which listeners receive speakers' production of linguistic or non-linguistic knowledge. Compared with reading comprehension, it may arouse difficulties and thus impose more burdens on foreign learners. The Audio-Lingual Method focused primarily on speaking. Mimicry, repetition, rote memory, and transformation drills actually interfered with listening comprehension. So learners lost interest and were not highly motivated. Improving listening comprehension requires continual attentiveness and interest. Listening skill can be extended systematically only when students are frequently exposed to a wide range of listening materials with an affective, cultural, social, and psycholinguistic approach. Therefore, teachers should help students learn how to comprehend intactly the overall meaning of intended messages. The literature on teaching listening skill suggests various useful activities: TPR, dictation, role playing, singing, picture recognition, completion, prediction, seeking specific information, summarizing, labeling, humor, jokes, cartoons, media, and so on. Practical classroom teaching necessitates a systematic procedure in which students should take part in meaningful tasks/activities. In addition to this, learners must practice listening comprehension trough a self-study process.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2003.05a
• /
• pp.325-329
• /
• 2003

This paper discuss the method of the system's efficient control using a Intelligent hybrid algorithm in nonlinear dynamics systems. Existing neural network and genetic algorithm for the control of non-linear systems work well in static states. but it be not particularly good in changeable states and must re-learn for the control of the system in the changed state. This time spend a lot of time. For the solution of this problem we suggest the intelligent hybrid self-tuning controller. it includes neural network, genetic algorithm and immune system. it is based on neural network, and immune system and genetic algorithm are added against a changed factor. We will call a change factor an antigen. When an antigen broke out, immune system come into action and genetic algorithm search an antibody. So the system is controled more stably and rapidly. Moreover, The Genetic algorithm use the memory address of the immune bank as a genetic factor. So it brings an advantage which the realization of a hardware easy.