• Journal of the Korea Convergence Society
• /
• v.12 no.5
• /
• pp.23-29
• /
• 2021

Recent deep learning-based natural language processing studies are conducting research to improve performance by training large amounts of data from various sources together. However, there is a possibility that the methodology of learning by combining data from various sources into one may prevent performance improvement. In the case of machine translation, data deviation occurs due to differences in translation(liberal, literal), style(colloquial, written, formal, etc.), domains, etc. Combining these corpora into one for learning can adversely affect performance. In this paper, we propose a new Corpus Weight Balance(CWB) method that considers the balance between parallel corpora in machine translation. As a result of the experiment, the model trained with balanced corpus showed better performance than the existing model. In addition, we propose an additional corpus construction process that enables coexistence with the human translation market, which can build high-quality parallel corpus even with a monolingual corpus.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.30B no.10
• /
• pp.77-85
• /
• 1993

In this paper, we present a method to solve a convexly combined fuzzy relational equation with generalized connectives. For this, we propose a neural network whose structure represents the fuzzy relational equation. Then we derive a learning algorithm by using the concept of back-propagation learning. Since the proposed method can be used for a general form of fuzzy relational equations, such fuzzy max-min or min-max relational equations can be treated as its special cases. Moreover, the relational structure adopted in the proposed neurocomputational approach can work in a highly parallel manner so that real-time applications of fuzzy sets are possibles as in fuzzy logic controllers, knowledge-based systems, and pattern recognition systems.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2003.09b
• /
• pp.122-125
• /
• 2003

By advent of NNC(Neural Network Chip), it is possible that process in parallel and discern the importance of signal with learning oneself by experience in external signal. So, the design of general purpose operation unit using VHDL(VHSIC Hardware Description Language) on the existing FPGA(Field Programmable Gate Array) can replaced EN(Expert Network) and learning algorithm. Also, neural network operation unit is possible various operation using learning of NN(Neural Network). This paper present general purpose operation unit using hierarchical structure of EN EN of presented structure learn from logical gate which constitute a operation unit, it relocated several layer The overall structure is hierarchical using a module, it has generality more than FPGA operation unit.

• 제어로봇시스템학회:학술대회논문집
• /
• 1992.10a
• /
• pp.632-637
• /
• 1992

Recently many dynamics control algorithms using robot dynamic equation have been proposed. One of them, Kawato's feedback error learning scheme requires neither an accurate model nor parameter estimation and makes the robot motion closer to the desired trajectory by repeating operation. In this paper, the feedback error learning algorithm is implemented to control a robot system, 5 DOF revolute type movemaster. For this purpose, an actuator dynamic model is constructed considering equivalent robot dynamics model with respect to actuator as well as friction model. The command input acquired from the actuator dynamic model is the sum of products of unknown parameters and known functions. To compute the control algorithm, a parallel processing computer, transputer, is used and real-time computing is achieved. The experiment is done for the three major link of movemaster and its result is presented.

• Proceedings of the KIEE Conference
• /
• 1998.07g
• /
• pp.2296-2298
• /
• 1998

In this paper, an on-chip learning pulse-mode digital neural network with a massively parallel yet compact and flexible network architecture is suggested. Algebraic neural operations are replaced by stochastic processes using pseudo-random sequences and simple logic gates are used as basic computing elements. Using Back-propagation algorithm both feed-forward and learning phases are efficiently implemented with simple logical gates. RNG architecture using LFSR and barrel shifter are adopted to avoid some correlation between pulse trains. Suggested network is designed in digital circuit and its performance is verified by computer simulation.

• Proceedings of the KIEE Conference
• /
• 1995.07b
• /
• pp.804-806
• /
• 1995

This paper discusses Self-organized Distributed Networks(SODN) as identifier of nonlinear dynamical systems. The structure of system identification employs series-parallel model. The identification procedure is based on a discrete-time formulation. The learning with the proposed SODN is fast and precise. Such properties arc caused from the local learning mechanism. Each local networks learns only data in a subregion. Large number of memory requirements and low generalization capability for the untrained region, which are drawbacks of conventional local network learning, are overcomed in the SODN. Through extensive simulation, SODN is shown to be effective for identification of nonlinear dynamical systems.

• The Transactions of the Korean Institute of Electrical Engineers D
• /
• v.53 no.8
• /
• pp.578-584
• /
• 2004

The CSVM(Current Support Vector Machine) that is a digital architecture performing all phases of recognition process including kernel computing, learning, and recall of SVM(Support Vector Machine) on a chip is proposed. Concurrent operation by parallel architecture of elements generates high speed and throughput. The classification problems of bio data having high dimension are solved fast and easily using the CSVM. Quadratic programming in original SVM learning algorithm is not suitable for hardware implementation, due to its complexity and large memory consumption. Hardware-friendly SVM learning algorithms, kernel adatron and kernel perceptron, are embedded on a chip. Experiments on fixed-point algorithm having quantization error are performed and their results are compared with floating-point algorithm. CSVM implemented on FPGA chip generates fast and accurate results on high dimensional cancer data.

• Journal of the Korean Society for Precision Engineering
• /
• v.16 no.11
• /
• pp.9-15
• /
• 1999

A piezoelectric actuator is widely used in precision positioning applications due to its excellent positioning resolution. However, the piezoelectric actuator lacks in repeatability because of its inherently high hysteresis characteristic between voltage and displacement. In this paper, a controller is proposed to compensate the hysteresis nonlinearity. The controller is composed of a PID and a neural network part in parallel manner. The output of the PID controller is used to teach the neural network controller by the unsupervised learning method. In addition, the PID controller stabilizes the piezoelectric actuator in the beginning of the learning process, when the neural network controller is not learned. However, after the learning process the piezoelectric actuator is mainly controlled by the neural netwok controller. In this paper, the excellent tracking performance of the proposed controller was verified by experiments and was compared with the classical PID controller.

• Journal of Platform Technology
• /
• v.11 no.2
• /
• pp.3-14
• /
• 2023

This paper proposes a big data sentiment analysis method and deep learning implementation method to provide a webtoon comment analysis web page for convenient comment confirmation and feedback of webtoon writers for the development of the cartoon industry in the video animation field. In order to solve the difficulty of automatic analysis due to the nature of Internet comments and provide various sentiment analysis information, LSTM(Long Short-Term Memory) algorithm, ranking algorithm, and word2vec algorithm are applied in parallel, and actual popular works are used to verify the validity. If the analysis method of this paper is used, it is easy to expand to other domestic and overseas platforms, and it is expected that it can be used in various video animation content fields, not limited to the webtoon field

• ETRI Journal
• /
• v.45 no.1
• /
• pp.93-104
• /
• 2023

Coreference resolution is a task in discourse analysis that links several headwords used in any document object. We suggest pointer networks-based coreference resolution for Korean using multi-task learning (MTL) with an attention mechanism for a hierarchical structure. As Korean is a head-final language, the head can easily be found. Our model learns the distribution by referring to the same entity position and utilizes a pointer network to conduct coreference resolution depending on the input headword. As the input is a document, the input sequence is very long. Thus, the core idea is to learn the word- and sentence-level distributions in parallel with MTL, while using a shared representation to address the long sequence problem. The suggested technique is used to generate word representations for Korean based on contextual information using pre-trained language models for Korean. In the same experimental conditions, our model performed roughly 1.8% better on CoNLL F1 than previous research without hierarchical structure.