• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.25 no.4
• /
• pp.501-507
• /
• 2021

Neural networks with differentiable activation functions are differentiable with respect to input variables. We improve the approximation capability of neural networks by using the gradient and Hessian of neural networks to satisfy the differential equations of the problems of interest. We apply differential neural networks to the pricing of financial options, where stochastic differential equations and the Black-Scholes partial differential equation represent the differential relation of price of option and underlying assets, and the first and second derivatives of option price play an important role in financial engineering. The proposed neural network learns - (a) the sample paths of option prices generated by stochastic differential equations and (b) the Black-Scholes equation at each time and asset price. Experimental results show that the proposed method gives accurate option values and the first and second derivatives.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2002.12a
• /
• pp.502-505
• /
• 2002

본 논문은 다치(MVL:Multiple Valued Logic) 신경망의 BP(Backpropagation) 학습 알고리즘을 이용하여 패턴 인식에 이용하는 방법을 제안한다. MVL 신경망을 이용하여 패턴 인식에 이용함으로서, 네트워크에 필요한 시간 및 기억 공간을 최소화할 수 있고 환경 변화에 적응할 수 있는 가능성을 제시하였다. MVL 신경망은 다치 논리 함수를 기반으로 신경망을 구성하였으며, 입력은 리터럴 함수로 변환시키고, 출력은 MIN과 MAX 연산을 사용하여 구하였고, 학습을 하기 위해 다치 논리식의 편 미분을 사용하였다.

• Journal of Digital Convergence
• /
• v.16 no.12
• /
• pp.217-229
• /
• 2018

This paper aims at applying the data mining such as decision tree, neural network, and logistic regression to an unsold apartment complex in Wirye new town and developing the model forecasting the result of initial sale contract by house unit. Raw data are divided into training data and test data. The order of predictability in training data is neural network, decision tree, and logistic regression. On the contrary, the results of test data show that logistic regression is the best model. This means that logistic regression has more data adaptability than neural network which is developed as the model optimized for training data. Determinants of initial sale are the location of floor, direction, the location of unit, the proximity of electricity and generator room, subscriber's residential region and the type of subscription. This suggests that using two models together is more effective in exploring determinants of initial sales. This paper contributes to the development of convergence field by expanding the scope of data mining.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.24 no.7
• /
• pp.842-848
• /
• 2020

Music markets have entered the era of streaming. In order to select and propose music that suits the taste of music consumers, there is an active demand and research on an automatic music genre classification system. We propose a method to improve the accuracy of genre unclassified songs, which was a lack of the previous system, by using a generative adversarial network (GAN) to further develop the automatic voting system for deep learning music genre using Softmax proposed in the previous paper. In the previous study, if the spectrogram of the song was ambiguous to grasp the genre of the song, it was forced to leave it as an unclassified song. In this paper, we proposed a system that increases the accuracy of genre classification of unclassified songs by converting the spectrogram of unclassified songs into an easy-to-read spectrogram using GAN. And the result of the experiment was able to derive an excellent result compared to the existing method.

• Journal of Digital Contents Society
• /
• v.18 no.5
• /
• pp.969-976
• /
• 2017

This paper proposes the learning method of an artificial neural network and a convolutional neural network using the WFSO algorithm developed as an optimization algorithm. Since the optimization algorithm searches based on a number of candidate solutions, it has a drawback in that it is generally slow, but it rarely falls into the local optimal solution and it is easy to parallelize. In addition, the artificial neural networks with non-differentiable activation functions can be trained and the structure and weights can be optimized at the same time. In this paper, we describe how to apply WFSO algorithm to artificial neural network learning and compare its performances with error back-propagation algorithm in multilayer artificial neural networks and convolutional neural networks.

• Proceedings of the Korea Multimedia Society Conference
• /
• 2002.05c
• /
• pp.325-329
• /
• 2002

본 연구에서는 할선법과 모멘트를 조합한 학습알고리즘의 신경망 기반 독립성분분석 기법을 제안하였다. 제안된 알고리즘은 할선법과 모멘트에 기초를 둔 고정점 알고리즘의 독립성분분석 기법이다. 여기서 할선법은 독립성분 상호간의 정보를 최소화하기 위해 negentropy를 최대화는 과정에서 요구되는 1차 미분에 따른 계산량을 줄이기 위함이고, 모멘트는 최대화 과정에서 발생하는 발진을 억제하여 보다 빠른 학습을 위함이다. 제안된 기법을 256×256 픽셀의 8개 지문영상에서 임의 혼합행렬에 따라 발생되는 혼합지문들을 각각 대상으로 시뮬레이션한 결과, 할선법만에 기초한 기법보다 우수한 분리성능과 빠른 학습속도가 있음을 확인하였다.

• Geophysics and Geophysical Exploration
• /
• v.26 no.3
• /
• pp.114-125
• /
• 2023

The physics-informed neural network (PINN) has been proposed to overcome the limitations of various numerical methods used to solve partial differential equations (PDEs) and the drawbacks of purely data-driven machine learning. The PINN directly applies PDEs to the construction of the loss function, introducing physical constraints to machine learning training. This technique can also be applied to wave equation modeling. However, to solve the wave equation using the PINN, second-order differentiations with respect to input data must be performed during neural network training, and the resulting wavefields contain complex dynamical phenomena, requiring careful strategies. This tutorial elucidates the fundamental concepts of the PINN and discusses considerations for wave equation modeling using the PINN approach. These considerations include spatial coordinate normalization, the selection of activation functions, and strategies for incorporating physics loss. Our experimental results demonstrated that normalizing the spatial coordinates of the training data leads to a more accurate reflection of initial conditions in neural network training for wave equation modeling. Furthermore, the characteristics of various functions were compared to select an appropriate activation function for wavefield prediction using neural networks. These comparisons focused on their differentiation with respect to input data and their convergence properties. Finally, the results of two scenarios for incorporating physics loss into the loss function during neural network training were compared. Through numerical experiments, a curriculum-based learning strategy, applying physics loss after the initial training steps, was more effective than utilizing physics loss from the early training steps. In addition, the effectiveness of the PINN technique was confirmed by comparing these results with those of training without any use of physics loss.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.14 no.3
• /
• pp.375-382
• /
• 2004

This paper presents control designs using neural networks (NN) for a XY positioning table. The proposed neuro-controller is composed of an outer PD tracking loop for stabilization of the fast flexible-mode dynamics and an NN inner loop used to compensate for the system nonlinearities. A tuning algorithm is given for the NN weights, so that the NN compensation scheme becomes adaptive, guaranteeing small tracking errors and bounded weight estimates. Formal nonlinear stability proofs are given to show that the tracking error is small. The proposed neuro-controller is implemented and tested on an IBM PC-based XY positioning table, and is applicable to many precision XY tables. The algorithm, simulation, and experimental results are described. The experimental results are shown to be superior to those of conventional control.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2002.04b
• /
• pp.1037-1040
• /
• 2002

본 연구에서는 효율적인 학습알고리즘을 가지는 신경망 기반 독립성분분석 기법을 제안하였다. 제안된 기법은 할선법에 기초를 둔 fixed point 알고리즘의 신경망 기반 독립성분분석 기법이다. 여기서 할선법은 독립성분 상호간의 정보를 최소화하기 위해 negentropy를 최대화는 과정에서 요구되는 1차 미분에 따른 계산량을 줄이기 위함이다. 제안된 기법을 500개의 데이터를 가지는 4개 신호들로부터 임의의 혼합 행렬에 따라 발생되는 혼합신호들을 각각 대상으로 시뮬레이션 한 결과, 우수한 분리성능과 빠른 학습 속도가 있음을 확인하였다.

• Proceedings of the Korean Society for Emotion and Sensibility Conference
• /
• 1997.11a
• /
• pp.168-173
• /
• 1997

시선의 추출을 통해 사용자의 관심 방향을 알고자하는 연구는 여러 분야에 응용될 수 있는데, 대표적인 것이 장애인의 컴퓨터 이용이나, 다중 윈도우에서 마우스의 기능 대용 및, VR에서의 위치 추적 장비의 대용 그리고 원격 회의 시스템에서의 view controlling등이다. 기존의 대부분의 연구들에서는 얼굴의 입력된 동영상으로부터 얼굴의 3차원 움직임량(rotation, translation)을 구하는데 중점을 두고 있으나 [1][2], 모니터, 카메라, 얼굴 좌표계간의 복잡한 변환 과정때문에 이를 바탕으로 사용자의 응시 위치를 파악하고자하는 연구는 거으 이루어지지 않고 있다. 본 논문에서는 일반 사무실 환경에서 입력된 얼굴 동영상으로부터 얼굴 영역 및 얼굴내의 눈, 코, 입 영역 등을 추출함으로써 모니터의 일정 영역을 응시하는 순간 변화된 특징점들의 위치 및 특징점들이 형성하는 기하학적 모양의 변화를 바탕으로 응시 위치를 계산하였다. 이때 앞의 세 좌표계간의 복잡한 변환 관계를 해결하기 위하여, 신경망 구조(다층 퍼셉트론)을 이용하였다. 신경망의 학습 과정을 위해서는 모니터 화면을 15영역(가로 5등분, 세로 3등분)으로 분할하여 각 영역의 중심점을 응시할 때 추출된 특징점들을 사용하였다. 이때 학습된 15개의 응시 위치이외에 또 다른 응시 영역에 대한 출력값을 얻기 위해, 출력 함수로 연속적이고 미분가능한 함수(linear output function)를 사용하였다. 실험 결과 신경망을 이용한 응시위치 파악 결과가 선형 보간법[3]을 사용한 결과보다 정확한 성능을 나타냈다.