• Journal of Intelligence and Information Systems
• /
• v.29 no.1
• /
• pp.249-263
• /
• 2023

Collaborative Filtering, a representative recommendation system methodology, consists of two approaches: neighbor methods and latent factor models. Among these, the latent factor model using matrix factorization decomposes the user-item interaction matrix into two lower-dimensional rectangular matrices, predicting the item's rating through the product of these matrices. Due to the factor vectors inferred from rating patterns capturing user and item characteristics, this method is superior in scalability, accuracy, and flexibility compared to neighbor-based methods. However, it has a fundamental drawback: the need to reflect the diversity of preferences of different individuals for items with no ratings. This limitation leads to repetitive and inaccurate recommendations. The Adaptive Deep Latent Factor Model (ADLFM) was developed to address this issue. This model adaptively learns the preferences for each item by using the item description, which provides a detailed summary and explanation of the item. ADLFM takes in item description as input, calculates latent vectors of the user and item, and presents a method that can reflect personal diversity using an attention score. However, due to the requirement of a dataset that includes item descriptions, the domain that can apply ADLFM is limited, resulting in generalization limitations. This study proposes a Generalized Adaptive Deep Latent Factor Recommendation Model, G-ADLFRM, to improve the limitations of ADLFM. Firstly, we use item ID, commonly used in recommendation systems, as input instead of the item description. Additionally, we apply improved deep learning model structures such as Self-Attention, Multi-head Attention, and Multi-Conv1D. We conducted experiments on various datasets with input and model structure changes. The results showed that when only the input was changed, MAE increased slightly compared to ADLFM due to accompanying information loss, resulting in decreased recommendation performance. However, the average learning speed per epoch significantly improved as the amount of information to be processed decreased. When both the input and the model structure were changed, the best-performing Multi-Conv1d structure showed similar performance to ADLFM, sufficiently counteracting the information loss caused by the input change. We conclude that G-ADLFRM is a new, lightweight, and generalizable model that maintains the performance of the existing ADLFM while enabling fast learning and inference.

• Journal of IKEEE
• /
• v.23 no.4
• /
• pp.1302-1308
• /
• 2019

This paper presents an approach for classifying image made by acquired position data from a 2D Lidar sensor with a convolutional neural network (CNN). Lidar sensor has been widely used for unmanned devices owing to advantages in term of data accuracy, robustness against geometry distortion and light variations. A CNN algorithm consists of one or more convolutional and pooling layers and has shown a satisfactory performance for image classification. In this paper, different types of CNN architectures based on training methods, Gradient Descent(GD) and Levenberg-arquardt(LM), are implemented. The LM method has two types based on the frequency of approximating Hessian matrix, one of the factors to update training parameters. Simulation results of the LM algorithms show better classification performance of the image data than that of the GD algorithm. In addition, the LM algorithm with more frequent Hessian matrix approximation shows a smaller error than the other type of LM algorithm.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.23 no.10
• /
• pp.1290-1297
• /
• 2019

The paper proposes a distance estimation technique for ultra-wideband (UWB) systems using convolutional neural network (CNN). To estimate the distance from the transmitter and the receiver in the proposed method, 1 dimensional vector consisted of the magnitudes of the received samples is reshaped into a 2 dimensional matrix, and by using this matrix, the distance is estimated through the CNN regressor. The received signal for CNN training is generated by the UWB channel model in the IEEE 802.15.4a, and the CNN model is trained. Next, the received signal for CNN test is generated by filed experiments in indoor environments, and the distance estimation performance is verified. The proposed technique is also compared with the existing threshold based method. According to the results, the proposed CNN based technique is superior to the conventional method and specifically, the proposed method shows 0.6 m root mean square error (RMSE) at distance 10 m while the conventional technique shows much worse 1.6 m RMSE.

• Proceedings of the Korean Society for Cognitive Science Conference
• /
• 2005.05a
• /
• pp.177-181
• /
• 2005

기계학습에서 커널을 이용한 방법은 그 응용범위가 기계학습의 전반에 걸쳐 다양하게 이용되고 있으며, 그 성능 또한 기존의 방법들을 앞지르고 있다. 이는 기존의 비선형적 접근을 커널을 이용한 고차원 공간에서의 선형적 접근법으로 바꿈으로써 가능하게 되는 것이다. 다양한 분야에 적용되는 많은 커널들이 존재하며 각 커널들은 특별한 분야에 적용되기 쉽도록 다른 형태를 띠고 있기도 하지만, 커널로서 작용하기 위해 양한정 조건(positive definiteness)을 만족해야 한다. 본 연구에서는 DNA 문제에 직접 적용시킬 수 있는 방법으로서의 새로운 커널을 제시한다. 또한 매트로폴리스(Metropolis) 알고리즘을 이용하여 DNA의 hybridization과정을 모사함으로써 새로운 종류의 커널이 양한정(positive definite) 조건을 만족시킬 수 있는 방법을 제시한다. 새로 만들어진 커널이 행렬값을 형성해 나가는 과정을 살펴보면 인간이 예(instance)로부터 개념을 형성해 나가는 과정과 흡사한 양상을 보이는 것을 알 수 있다. 개념을 나타내는 좋은 예로서의 표본(prototype)으로부터 개념이 형성되어 가는 과정은 표본(prototype)이 아닌 예로부터 개념이 형성되는 과정과 다른 양상을 띠는 것과 같은 모양을 보인다.

• Proceedings of the Korean Information Science Society Conference
• /
• 2006.10b
• /
• pp.324-329
• /
• 2006

Non-negative matrix factorization(NMF) 기법은 음이 아닌 값으로 구성된 데이터를 두 종류의 양의 행렬의 곱의 형식으로 분할하는 데이터 분석기법으로서, 텍스트마이닝, 바이오인포매틱스, 멀티미디어 데이터 분석 등에 활용되었다. 본 연구에서는 기본 NMF 기법에 기반하여 텍스트 문서로부터 토픽을 추출하고 동시에 이를 가시적으로 도시하기 위한 Topographic NMF (TNMF) 기법을 제안한다. TNMF에 의한 토픽 가시화는 데이터를 전체적인 관점에서 보다 직관적으로 파악하는데 도움이 될 수 있다. TNMF는 생성모델 관점에서 볼 때, 2개의 은닉층을 갖는 계층적 모델로 표현할 수 있으며, 상위 은닉층에서 하위 은닉층으로의 연결은 토픽공간상에서 토픽간의 전이확률 또는 이웃함수를 정의한다. TNMF에서의 학습은 전이확률값의 연속적 스케줄링 과정 속에서 반복적 파리미터 갱신 과정을 통해 학습이 이루어지는데, 파라미터 갱신은 기본 NMF 기반 학습 과정으로부터 유사한 형태로 유도될 수 있음을 보인다. 추가적으로 Probabilistic LSA에 기초한 토픽 가시화 기법 및 희소(sparse)한 해(解) 도출을 목적으로 한 non-smooth NMF 기법과의 연관성을 분석, 제시한다. NIPS 학회 논문 데이터에 대한 실험을 통해 제안된 방법론이 문서 내에 내재된 토픽들을 효과적으로 가시화 할 수 있음을 제시한다.

• Communications of Mathematical Education
• /
• v.21 no.4
• /
• pp.621-646
• /
• 2007

In our new learning environment, we were asked to change our teaching method in our Linear Algebra class. In mathematics class, we could use several math-softwares such as MATHEMATICA, MATLAB, MAPLE, Drive etc.. MATLAB was quite well fit with our Linear Algebra class. In this paper we introduce an efficient way of delivery on important concepts in linear algebra by using well-known MATLAB/ATLAST M-files which we downloded from http://www.umassd.edu/specialprograms/atlast/.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.07a
• /
• pp.667-670
• /
• 2020

최근에는 증강현실, 로봇공학 등의 분야에서 객체의 위치 검출 이외에도, 객체의 자세에 대한 추정이 요구되고 있다. 객체의 자세 정보가 포함된 데이터셋은 위치 정보만 포함된 데이터셋에 비하여 상대적으로 매우 적기 때문에 인공 신경망 구조를 활용하기 어려운 측면이 있으나, 최근에 들어서는 기계학습 기반의 자세 추정 알고리즘들이 여럿 등장하고 있다. 본 논문에서는 이 가운데 Dense 6d Pose Object detector (DPOD) [11]의 구조를 기반으로 하여 가구의 조립 설명서에 그려진 가구 부품들의 자세를 추정하고자 한다. DPOD [11]는 입력으로 RGB 영상을 받으며, 해당 영상에서 자세를 추정하고자 하는 객체의 영역에 해당하는 픽셀들을 추정하고, 객체의 영역에 해당되는 각 픽셀에서 해당 객체의 3D 모델의 UV map 값을 추정한다. 이렇게 픽셀 개수만큼의 2D - 3D 대응이 생성된 이후에는, RANSAC과 PnP 알고리즘을 통해 RGB 영상에서의 객체와 객체의 3D 모델 간의 변환 관계 행렬이 구해지게 된다. 본 논문에서는 사전에 정해진 24개의 자세 후보들을 기반으로 가구 부품의 3D 모델을 2D에 투영한 RGB 영상들로 인공 신경망을 학습하였으며, 평가 시에는 실제 조립 설명서에서의 가구 부품의 자세를 추정하였다. 실험 결과 IKEA의 Stefan 의자 조립 설명서에 대하여 100%의 ADD score를 얻었으며, 추정 자세가 자세 후보군 중 정답 자세에 가장 근접한 경우를 정답으로 평가했을 때 100%의 정답률을 얻었다. 제안하는 신경망을 사용하였을 때, 가구 조립 설명서에서 가구 부품의 위치를 찾는 객체 검출기(object detection network)와, 각 개체의 종류를 구분하는 객체 리트리벌 네트워크(retrieval network)를 함께 사용하여 최종적으로 가구 부품의 자세를 추정할 수 있다.

• Korean Chemical Engineering Research
• /
• v.61 no.3
• /
• pp.388-393
• /
• 2023

The analysis of engineering data using neural network based on supervised learning has been utilized in various engineering fields such as optimization of chemical engineering process, concentration prediction of particulate matter pollution, prediction of thermodynamic phase equilibria, and prediction of physical properties for transport phenomena system. The supervised learning requires training data, and the performance of the supervised learning is affected by the composition and the configurations of the given training data. Among the frequently observed engineering data, the data is given in log-scale such as length of DNA, concentration of analytes, etc. In this study, for widely distributed log-scaled training data of virtual 100×100 images, available loss functions were quantitatively evaluated in terms of (i) confusion matrix, (ii) maximum relative error and (iii) mean relative error. As a result, the loss functions of mean-absolute-percentage-error and mean-squared-logarithmic-error were the optimal functions for the log-scaled training data. Furthermore, we figured out that uniformly selected training data lead to the best prediction performance. The optimal loss functions and method for how to compose training data studied in this work would be applied to engineering problems such as evaluating DNA length, analyzing biomolecules, predicting concentration of colloidal suspension.

• Journal of Advanced Navigation Technology
• /
• v.12 no.5
• /
• pp.502-508
• /
• 2008

With the increased use of the internet and the tremendous amount of data it transfers, it is more necessary to summarize documents. We propose a new method using the Non-negative Semantic Variable Matrix (NSVM) and the generic important of semantic features obtained by Non-negative Matrix Factorization (NMF) to extract the sentences for automatic generic summarization. The proposed method use non-negative constraints which is more similar to the human's cognition process. As a result, the proposed method selects more meaningful sentences for summarization than the unsupervised method used the Latent Semantic Analysis (LSA) or clustering methods. The experimental results show that the proposed method archives better performance than other methods.

• Journal of Internet Computing and Services
• /
• v.5 no.3
• /
• pp.57-62
• /
• 2004

In this paper, an efficient speaker identification based on robust vector quantizationprincipal component analysis (VQ-PCA) is proposed to solve the problems from outliers and high dimensionality of training feature vectors in speaker identification, Firstly, the proposed method partitions the data space into several disjoint regions by roust VQ based on M-estimation. Secondly, the robust PCA is obtained from the covariance matrix in each region. Finally, our method obtains the Gaussian Mixture model (GMM) for speaker from the transformed feature vectors with reduced dimension by the robust PCA in each region, Compared to the conventional GMM with diagonal covariance matrix, under the same performance, the proposed method gives faster results with less storage and, moreover, shows robust performance to outliers.