• The Korean Journal of Applied Statistics
• /
• v.34 no.3
• /
• pp.329-345
• /
• 2021

Recommender systems use data from customers to suggest personalized products. The recommender systems can be categorized into three cases; collaborative filtering, contents-based filtering, and hybrid recommender system that combines the first two filtering methods. In this work, we introduce and compare deep learning-based recommender system using autoencoder. Autoencoder is an unsupervised deep learning that can effective solve the problem of sparsity in the data matrix. Five versions of autoencoder-based deep learning models are compared via three real data sets. The first three methods are collaborative filtering and the others are hybrid methods. The data sets are composed of customers' ratings having integer values from one to five. The three data sets are sparse data matrix with many zeroes due to non-responses.

• ETRI Journal
• /
• v.34 no.4
• /
• pp.645-648
• /
• 2012

An incremental subspace learning scheme to recover lost speech segments online is presented. Our contributions in this work are twofold. First, the recovery problem is transformed into an interpolation problem of the time-varying gains via nonnegative matrix factorization. Second, incremental nonnegative matrix factorization is employed to allow online processing and track the evolution of speech statistics. The effectiveness of the proposed scheme is confirmed by the experiment results.

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• v.3 no.1
• /
• pp.100-104
• /
• 2003

The SVDD(support vector data description) is one of the most well-known one-class support vector learning methods, in which one tries the strategy of utilizing balls defined on the kernel feature space in order to distinguish a set of normal data from all other possible abnormal objects. The major concern of this paper is to consider the problem of modifying the SVDD into the direction of utilizing ellipsoids instead of balls in order to enable better classification performance. After a brief review about the original SVDD method, this paper establishes a new method utilizing ellipsoids in feature space, and presents a solution in the form of SDP(semi-definite programming) which is an optimization problem based on linear matrix inequalities.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.12 no.6
• /
• pp.2806-2825
• /
• 2018

Unsupervised learning has shown good performance on image, video and audio classification tasks, and much progress has been made so far. It studies how systems can learn to represent particular input patterns in a way that reflects the statistical structure of the overall collection of input patterns. Many promising deep learning systems are commonly trained by the greedy layerwise unsupervised learning manner. The performance of these deep learning architectures benefits from the unsupervised learning ability to disentangling the abstractions and picking out the useful features. However, the existing unsupervised learning algorithms are often difficult to train partly because of the requirement of extensive hyperparameters. The tuning of these hyperparameters is a laborious task that requires expert knowledge, rules of thumb or extensive search. In this paper, we propose a simple and effective unsupervised feature learning algorithm for image classification, which exploits an explicit optimizing way for population and lifetime sparsity. Firstly, a sparse target matrix is built by the competitive rules. Then, the sparse features are optimized by means of minimizing the Euclidean norm ($L_2$) error between the sparse target and the competitive layer outputs. Finally, a classifier is trained using the obtained sparse features. Experimental results show that the proposed method achieves good performance for image classification, and provides discriminative features that generalize well.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.19 no.6
• /
• pp.118-133
• /
• 2020

Research interest in the Mobility-as-a-Service (MaaS) concept for enhancing users' mobility experience is increasing. In particular, dynamic pricing techniques based on reinforcement learning have emerged since adjusting prices based on the demand is expected to help mobility services, such as taxi and car-sharing services, to gain more profit. This paper provides a simulation framework that considers more practical factors, such as demand density per location, preferred prices, the distance between users and drivers, and distance to the destination that critically affect the probability of matching between the users and the mobility service providers (e.g., drivers). The aforementioned new practical features are reflected on a data structure referred to as the Matching Matrix. Using an efficient algorithm of computing the probability of matching between the users and drivers and given a set of precisely identified high-demand locations using HDBSCAN, this study developed a better reward function that can gear the reinforcement learning process towards finding more realistic dynamic pricing policies.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.12 no.9
• /
• pp.1591-1598
• /
• 2008

Z. Cao had proposed NFRM(new fuzzy reasoning method) which infers in detail using relation matrix. In spite of the small inference rules, it shows good performance than mamdani's fuzzy inference method. In this paper, we propose Z. Cao's fuzzy inference method with learning ability which is used a gradient descent method in order to improve the performances. It is hard to determine the relation matrix elements by trial and error method. Because this method is needed many hours and effort. Simulation results are applied nonlinear systems show that the proposed inference method using a gradient descent method has good performances.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.31B no.5
• /
• pp.134-140
• /
• 1994

In this paper, an algorithm of CBAM(Combination Bidirectional Associative Memory) model proposes, analyzes and tests CBAM model `s performancess by simulating with recalls and recognitions of patterns. In learning-procedure each correlation matrix of training patterns is obtained. As each correlation matrix's some elements correspond to juxtaposition, all correlation matrices are merged into one matrix (Combination Correlation Matrix, CCM). In recall-procedure, CCM is decomposed into a number of correlation matrices by spiliting its elements into the number of elements corresponding to all training patterns. Recalled patterns are obtained by multiplying input pattern with all correlation matrices and selecting a pattern which has the smallest value of energy function. By using a CBAM model, we have some advantages. First, all pattern having less than 20% of noise can be recalled. Second, memory capacity of CBAM model, can be further increased to include English alphabets or patterns. Third, learning time of CBAM model can be reduced greatly because of operation to make CCM.

• International Journal of Contents
• /
• v.12 no.4
• /
• pp.31-44
• /
• 2016

Knowledge creation and management are regarded as critical success factors for an organization's survival in the knowledge era. As a process of knowledge acquisition and sharing, organizational learning mechanisms (OLMs) guide the learning function of organizations represented by its different learning activities. We examined a variety of learning processes that constitute OLMs. In this study, we aimed to capture the process and framework of OLMs and knowledge sharing and acquisition. Factors facilitating OLMs were investigated at three levels: individual, group, and organizational. The concept of an OLM has received some attention in the field of organizational learning, however, the relationship among the factors generating OLMs has not been empirically tested. As part of the ongoing discussion, we attempted a systemic approach for OLMs. OLMs can be represented by factors that are inherent to the organization's system; therefore, prior to empirically testing the OLM generating factor(s), evaluation of its organizational integration is required to determine effective treatment of each factor. Thus, we developed a framework to manage knowledge and proposed a method to numerically evaluate factors influencing the OLMs. Specifically, composite importance (CI) of the Decision-Making Trial and Evaluation Laboratory (DEMATEL) method was applied to explore the interaction effect of these factors based on systemic approach. The augmented matrix thus generated is expected to serve as a stochastic matrix of an absorbing Markov chain.

• Journal of Korea Society of Digital Industry and Information Management
• /
• v.17 no.4
• /
• pp.53-62
• /
• 2021

Deep learning models such as convolutional neural networks and recurrent neual networks process a huge amounts of data, so they require a lot of storage and consume a lot of time and power due to memory access. Recently, research is being conducted to reduce memory usage and access by compressing data using the feature that many of deep learning data are highly sparse and localized. In this paper, we propose a compression-decompression method of storing only the non-zero data and the location information of the non-zero data excluding zero data. In order to make the location information of non-zero data, the matrix data is divided into sections uniformly. And whether there is non-zero data in the corresponding section is indicated. In this case, section division is not executed only once, but repeatedly executed, and location information is stored in each step. Therefore, it can be properly compressed according to the ratio and distribution of zero data. In addition, we propose a hardware structure that enables compression and decompression without complex operations. It was designed and verified with Verilog, and it was confirmed that it can be used in hardware deep learning accelerators.

• Journal of Sensor Science and Technology
• /
• v.28 no.3
• /
• pp.182-186
• /
• 2019

When a camera is employed for 3D sensing, accurate camera calibration is vital as it is a prerequisite for the subsequent steps of the sensing process. Camera calibration is usually performed by complex mathematical modeling and geometric analysis. On the other contrary, data learning using an artificial neural network can establish a transformation relation between the 3D space and the 2D camera image without explicit camera modeling. However, a neural network requires a large amount of accurate data for its learning. A significantly large amount of time and work using a precise system setup is needed to collect extensive data accurately in practice. In this study, we propose a two-step neural calibration method that is effective when only a small amount of learning data is available. In the first step, the camera projection transformation matrix is determined using the limited available data. In the second step, the transformation matrix is used for generating a large amount of synthetic data, and the neural network is trained using the generated data. Results of simulation study have shown that the proposed method as valid and effective.