• Nuclear Engineering and Technology
• /
• v.52 no.2
• /
• pp.287-295
• /
• 2020

Group method of data handling (GMDH) is considered one of the earliest deep learning methods. Deep learning gained additional interest in today's applications due to its capability to handle complex and high dimensional problems. In this study, multi-layer GMDH networks are used to perform uncertainty quantification (UQ) and sensitivity analysis (SA) of nuclear reactor simulations. GMDH is utilized as a surrogate/metamodel to replace high fidelity computer models with cheap-to-evaluate surrogate models, which facilitate UQ and SA tasks (e.g. variance decomposition, uncertainty propagation, etc.). GMDH performance is validated through two UQ applications in reactor simulations: (1) low dimensional input space (two-phase flow in a reactor channel), and (2) high dimensional space (8-group homogenized cross-sections). In both applications, GMDH networks show very good performance with small mean absolute and squared errors as well as high accuracy in capturing the target variance. GMDH is utilized afterward to perform UQ tasks such as variance decomposition through Sobol indices, and GMDH-based uncertainty propagation with large number of samples. GMDH performance is also compared to other surrogates including Gaussian processes and polynomial chaos expansions. The comparison shows that GMDH has competitive performance with the other methods for the low dimensional problem, and reliable performance for the high dimensional problem.

• The Journal of the Korea Contents Association
• /
• v.19 no.11
• /
• pp.278-287
• /
• 2019

Spatial information extraction is to retrieve static and dynamic aspects in natural language text by explicitly marking spatial elements and their relational words. This paper proposes a deep learning approach for spatial information extraction for Korean language using a two-step bidirectional LSTM-CRF ensemble model. The integrated model of spatial element extraction and spatial relation attribute extraction is proposed too. An experiment with the Korean SpaceBank demonstrates the better efficiency of the proposed deep learning model than that of the previous CRF model, also showing that the proposed ensemble model performed better than the single model.

• International Journal of Aeronautical and Space Sciences
• /
• v.11 no.4
• /
• pp.326-337
• /
• 2010

This work addresses problems regarding trajectory planning for unmanned aerial vehicle sensors. Such sensors are used for taking measurements of large nonlinear systems. The sensor investigations presented here entails methods for improving estimations and predictions of large nonlinear systems. Thoroughly understanding the global system state typically requires probabilistic state estimation. Thus, in order to meet this requirement, the goal is to find trajectories such that the measurements along each trajectory minimize the expected error of the predicted state of the system. The considerable nonlinearity of the dynamics governing these systems necessitates the use of computationally costly Monte-Carlo estimation techniques, which are needed to update the state distribution over time. This computational burden renders planning to be infeasible since the search process must calculate the covariance of the posterior state estimate for each candidate path. To resolve this challenge, this work proposes to replace the computationally intensive numerical prediction process with an approximate covariance dynamics model learned using a nonlinear time-series regression. The use of autoregressive time-series featuring a regularized least squares algorithm facilitates the learning of accurate and efficient parametric models. The learned covariance dynamics are demonstrated to outperform other approximation strategies, such as linearization and partial ensemble propagation, when used for trajectory optimization, in terms of accuracy and speed, with examples of simplified weather forecasting.

• Journal of the Korean Society for Aeronautical & Space Sciences
• /
• v.35 no.3
• /
• pp.238-247
• /
• 2007

Genetic Algorithms(GA) which searches optimum solution using natural selection and the law of heredity has been applied to learning algorithms in order to estimate performance deterioration of the aircraft gas turbine engine. The compressor, gas generator turbine and power turbine are considered for engine performance deterioration and estimation for performance deterioration of a single component at design point was conducted. As a result of that, defect diagnostics has been conducted. The input criteria for the genetic algorithm to guarantee the high stability and reliability was discussed as increasing learning data sets. As a result, the accuracy of defect estimation and diagnostics were verified with its RMS error within 3%.

• Journal of IKEEE
• /
• v.24 no.4
• /
• pp.1148-1155
• /
• 2020

Fire must be extinguished as quickly as possible because they cause a lot of economic loss and take away precious human lives. Especially, the detection of smoke, which tends to be found first in fire, is of great importance. Smoke detection based on image has many difficulties in algorithm research due to the irregular shape of smoke. In this study, we introduce a new real-time smoke detection algorithm that reduces the detection of false positives generated by irregular smoke shape based on faster r-cnn of factory-installed surveillance cameras. First, we compute the global frame similarity and mean squared error (MSE) to detect the movement of smoke from the input surveillance camera. Second, we use deep learning algorithm (Faster r-cnn) to extract deferred candidate regions. Third, the extracted candidate areas for acting are finally determined using space and temporal features as smoke area. In this study, we proposed a new algorithm using the space and temporal features of global and local frames, which are well-proposed object information, to reduce false positives based on deep learning techniques. The experimental results confirmed that the proposed algorithm has excellent performance by reducing false positives of about 99.0% while maintaining smoke detection performance.

• Journal of the Korean Society for Precision Engineering
• /
• v.12 no.6
• /
• pp.120-127
• /
• 1995

A nonlinear empirical state-space model of the Artificial Neural Network(ANN) has been developed. The nonlinear model structure incorporates characteristic, so as to enable identification of the transient response, as well as the steady-state response of a dynamic system. A hybrid feedfoward/feedback neural network, namely a Local Time Delayed Recurrent Multi-layer Perception(RMLP), is the model structure developed in this paper. RMLP is used to identify nonlinear dynamic system in an input/output sense. The feedfoward protion of the network architecture provides with the well-known curve fitting factor, while local recurrent and cross-talk connections provides the dynamics of the system. A dynamic learning algorithm is used to train the proposed network in a supervised manner. The derived dynamic learning algorithm exhibit a computationally desirable characteristic; both network sweep involved in the algorithm are performed forward, enhancing its parallel implementation. RMLP state-space and its associate learning algorithm is demonstrated through a simple examples. The simulation results are very encouraging.

• Educational Technology International
• /
• v.9 no.1
• /
• pp.79-96
• /
• 2008

The issue of connection between entities has a lengthy history in educational research, especially since it provides the necessary bridge between base and target in analogical transfer. Recently, the connection has been viewed through the application of technology to bridge between sequences in order to be cognitively useful. This study reports the effect of sequence type (AT vs. TA) and connection type (fading vs. popping) on the achievement and analogical transfer in a multimedia application. In the current research, 10th -grade and 11th -grade biology students in Korea were randomly assigned to five groups to test the effects of presentation sequence and entity connection type on analogical transfer. Consistent with previous studies, sequence type has a significant effect: analogical transfer performance was better when base representations were presented first followed by target representations rather than the reverse order. This is probably because presenting a familiar base first helps in understanding a less familiar target. However, no fully significant differences were found with the entity connection types (fading vs. popping) in analogical transfer. According to the Markman and Gentner's (2005) spatial model, analogy in a space is influenced only by the differences between concepts, not by distance in space. Thus connection types fail on the basis of this spatial model in analogical transfer test. The findings and their implications for sequence and connection research and practice are discussed. Leveraging on the analogical learning process, specific implications for scaffolding learning processes and the development of adaptive expertise are drawn.

• Current Optics and Photonics
• /
• v.7 no.5
• /
• pp.545-556
• /
• 2023

This paper introduces volume-sharing multi-aperture imaging (VMAI), a potential approach proposed for volume reduction in space-borne imagers, with the aim of achieving high-resolution ground spatial imagery using deep learning methods, with reduced volume compared to conventional approaches. As an intermediate step in the VMAI payload development, we present a phase-1 design targeting a 1-meter ground sampling distance (GSD) at 500 km altitude. Although its optical imaging capability does not surpass conventional approaches, it remains attractive for specific applications on small satellite platforms, particularly surveillance missions. The design integrates one wide-field and three narrow-field cameras with volume sharing and no optical interference. Capturing independent images from the four cameras, the payload emulates a large circular aperture to address diffraction and synthesizes high-resolution images using deep learning. Computational simulations validated the VMAI approach, while addressing challenges like lower signal-to-noise (SNR) values resulting from aperture segmentation. Future work will focus on further reducing the volume and refining SNR management.

• Journal of the Korean Institute of Telematics and Electronics C
• /
• v.35C no.11
• /
• pp.76-88
• /
• 1998

While most of the research on reinforcement learning assumed a discrete control space, many of the real world control problems need to have continuous output. This can be achieved by using continuous mapping functions for the value and action functions of the reinforcement learning architecture. Two questions arise here however. One is what sort of function representation to use and the other is how to determine the amount of noise for search in action space. The ubiquitous neural network is used here to learn the value and policy functions. Next, the reinforcement predictor that is intended to predict the next reinforcement is introduced that also determines the amount of noise to add to the controller output. The proposed reinforcement learning architecture is found to have a sound on-line learning control performance especially at high-speed road following of high curvature road. Both computer simulation and actual experiments on a test vehicle have been performed and their efficiency and effectiveness has been verified.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.18 no.6
• /
• pp.1355-1360
• /
• 2014

In this paper, we apply the learning of the mobile robot insect augmented reality Digilog Book. In the era of electronic, book written in paper space just have moved to virtual reality space. The virtual reality, constraints spatial and physical, in the real world, it is a technique that enables to experience indirectly situation not experienced directly as user immersive experience type interface. Applied to the learning robot Digilog Book that allows the fusion of paper analog and digital content, using the augmented reality technology, to experience various interactions. Apply critical elements moving, three-dimensional images and animation to enrich the learning, for easier block assembly, designed to grasp more easily rank order between the blocks. Anywhere at any time, is capable of learning of the robot in Digilog Book to be executed by the mobile phone in particular.