• The KIPS Transactions:PartB
• /
• v.16B no.4
• /
• pp.333-340
• /
• 2009

Most of the previous automatic word spacing systems are not suitable to use for mobile devices with relatively low computing powers because they require many system resources. We propose an automatic word spacing system that requires reasonable memory usage and simple numerical computations for mobile devices with low computing powers. The proposed system is a two step model that consists of a statistical system and a rule-based system. To reduce the memory usage, the statistical system first corrects word spacing errors by using a modified hidden Markov model based on character unigrams. Then, to increase the accuracy, the rule-based system re-corrects miscorrected word spaces by using lexical rules based on character bigrams or more. In the experiments, the proposed system showed relatively high accuracy of 94.14% in spite of small memory usage of about 1MB.

• Journal of KIISE
• /
• v.44 no.1
• /
• pp.51-56
• /
• 2017

The research area for solving question answering (QA) problems using artificial intelligence models is in a methodological transition period, and one such architecture, the dynamic memory network (DMN), is drawing attention for two key attributes: its attention mechanism defined by neural network operations and its modular architecture imitating cognition processes during QA of human. In this paper, we increased accuracy of the inferred answers, by adapting an automatic data augmentation method for lacking amount of training data, and by improving the ability of time perception. The experimental results showed that in the 1K-bAbI tasks, the modified DMN achieves 89.21% accuracy and passes twelve tasks which is 13.58% higher with passing four more tasks, as compared with one implementation of DMN. Additionally, DMN's word embedding vectors form strong clusters after training. Moreover, the number of episodic passes and that of supporting facts shows direct correlation, which affects the performance significantly.

• Journal of Institute of Control, Robotics and Systems
• /
• v.13 no.4
• /
• pp.279-285
• /
• 2007

The objective of the research is two fold. The first is to design and propose a stable and robust learning control algorithm. The controller is CMAC Learning Controller which consists of a model-based controller, such as LQR or PID, as a reference control and a CMAC. The second objective is to implement a reference control and CMAC at two different sampling rates. Generally, a conventional controller is designed based on a mathematical plant model. However, increasing complexity of the plant and accuracy requirement on mathematical models nearly prohibits the application of the conventional controller design approach. To avoid inherent complexity and unavoidable uncertainty in modeling, biology mimetic methods have been developed. One of such attempts is Cerebellar Model Articulation Computer(CMAC) developed by Albus. CMAC has two main disadvantages. The first disadvantage of CMAC is increasing memory requirement with increasing number of input variables and with increasing accuracy demand. The memory needs can be solved with cheap memories due to recent development of new memory technology. The second disadvantage is a demand for processing powers which could be an obstacle especially when CMAC should be implemented in real-time. To overcome the disadvantages of CMAC, we propose CMAC learning controller with multiple sampling rates. With this approach a conventional controller which is a reference to CMAC at high enough sampling rate but CMAC runs at the processor's unoccupied time. To show efficiency of the proposed method, an inverted pendulum controller is designed and implemented. We also demonstrate it's possibility as an industrial control solution and robustness against a modeling uncertainty.

• Journal of the Korean Society for Aeronautical & Space Sciences
• /
• v.48 no.4
• /
• pp.277-284
• /
• 2020

The remaining range between missile and target can be used to calculate the guidance command as well as to determine the explosion time of the warhead. Since the range, however, is not directly measured by on-board sensors of the missile, it is usually estimated by filter-based techniques using angle-only measurements. Conventional filter-based techniques are complex and require huge computation. In this paper, we propose a range estimation algorithm based on the geometrical triangulation principle for two points of missiles and a target. The proposed algorithm has a simple structure but the accuracy is largely dependent on the measurement errors. To improve the accuracy of estimation, Digital Fading Memory Filter (DFMF) is applied. The performance of the proposed algorithm is analyzed through numerical simulations.

• Korean Journal of Remote Sensing
• /
• v.36 no.4
• /
• pp.515-525
• /
• 2020

This study investigates the potential of bidirectional long short-term memory (Bi-LSTM) for efficient modeling of temporal information in crop classification using multitemporal remote sensing images. Unlike unidirectional LSTM models that consider only either forward or backward states, Bi-LSTM could account for temporal dependency of time-series images in both forward and backward directions. This property of Bi-LSTM can be effectively applied to crop classification when it is difficult to obtain full time-series images covering the entire growth cycle of crops. The classification performance of the Bi-LSTM is compared with that of two unidirectional LSTM architectures (forward and backward) with respect to different input image combinations via a case study of crop classification in Anbadegi, Korea. When full time-series images were used as inputs for classification, the Bi-LSTM outperformed the other unidirectional LSTM architectures; however, the difference in classification accuracy from unidirectional LSTM was not substantial. On the contrary, when using multitemporal images that did not include useful information for the discrimination of crops, the Bi-LSTM could compensate for the information deficiency by including temporal information from both forward and backward states, thereby achieving the best classification accuracy, compared with the unidirectional LSTM. These case study results indicate the efficiency of the Bi-LSTM for crop classification, particularly when limited input images are available.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.41 no.6
• /
• pp.663-671
• /
• 2016

According to rapidly increasing of network traffics, necessity of high-speed router also increased. For various purposes, like traffic statistic and security, traffic measurement function should performed by router. However, because of the nature of high-speed router, memory resource of router was limited. RCC proposed a way to measure traffics with high speed and accuracy. Additional quadratic probing hashing table used for accumulating elephant flows in RCC. However, in our experiment, quadratic probing performed many overheads when allocated small memory space or load factor was high. Especially, quadratic requested many calculations in update and lookup. To face this kind of problem, we use a cuckoo hashing which performed a good performance in update and loop for enhancing the RCC. As results, RCC with cuckoo hashing performed high accuracy and speed even when load factor of memory was high.

• 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.

• The Journal of Korean Institute of Next Generation Computing
• /
• v.15 no.5
• /
• pp.7-19
• /
• 2019

Ransomware is a malicious program that requires the cost of decryption after encrypting files on the user's desktop. Since the frequency and the financial damage of ransomware attacks are increasing each year, ransomware prevention, detection and recovery system are needed. Baek et al. proposed SSD-Insider, an algorithm for detecting ransomware within SSD. In this paper, we propose an AdvanSSD-Insider algorithm that substitutes a hash table used for the overwriting check with a bloom filter in the SSD-Insider. Experimental results show that the AdvanSSD-Insider algorithm reduces memory usage by up to 90% and execution time by up to 77% compared to the SSD-Insider algorithm and achieves the same detection accuracy. In addition, the AdvanSSD-Insider algorithm can monitor 10 times longer than the SSD-Insider algorithm in same memory condition. As a result, detection accuracy is increased for some ransomware which was difficult to detect using previous algorithm.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.14 no.5
• /
• pp.277-286
• /
• 2019

In this paper, we propose an efficient hyperplane generation technique to classify human activity from combination of events and sequence information obtained from multiple-event sensors. By generating hyperplane efficiently, our machine learning algorithm classify with less memory and run time than the LSVM (Linear Support Vector Machine) for embedded system. Because the fact that light weight and high speed algorithm is one of the most critical issue in the IoT, the study can be applied to smart home to predict human activity and provide related services. Our approach is based on reducing numbers of hyperplanes and utilizing robust string comparing algorithm. The proposed method results in reduction of memory consumption compared to the conventional ML (Machine Learning) algorithms; 252 times to LSVM and 34,033 times to LSTM (Long Short-Term Memory), although accuracy is decreased slightly. Thus our method showed outstanding performance on accuracy per hyperplane; 240 times to LSVM and 30,520 times to LSTM. The binarized image is then divided into groups, where each groups are converted to binary number, in order to reduce the number of comparison done in runtime process. The binary numbers are then converted to string. The test data is evaluated by converting to string and measuring similarity between hyperplanes using Levenshtein algorithm, which is a robust dynamic string comparing algorithm. This technique reduces runtime and enables the proposed algorithm to become 27% faster than LSVM, and 90% faster than LSTM.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.25 no.11
• /
• pp.1649-1654
• /
• 2021

In this paper, we introduce a regularization of long short-term memory (LSTM) based fall detection system using TensorFlow that can detect falls that can occur in the elderly. Fall detection uses data from a 3-axis acceleration sensor attached to the body of an elderly person and learns about a total of 7 behavior patterns, each of which is a pattern that occurs in daily life, and the remaining 3 are patterns for falls. During training, a normalization process is performed to effectively reduce the loss function, and the normalization performs a maximum-minimum normalization for data and a L2 regularization for the loss function. The optimal regularization conditions of LSTM using several falling parameters obtained from the 3-axis accelerometer is explained. When normalization and regularization rate λ for sum vector magnitude (SVM) are 127 and 0.00015, respectively, the best sensitivity, specificity, and accuracy are 98.4, 94.8, and 96.9%, respectively.