• Smart Media Journal
• /
• v.11 no.11
• /
• pp.32-39
• /
• 2022

As IoT applications that provide AI services increase, various hardware and software that support autonomous learning and inference are being developed. However, as the characteristics and constraints of each hardware increase difficulties in developing IoT applications, the development of an integrated platform is required. In this paper, we propose a tool for automatically generating components based on artificial neural networks and spiking neural networks as well as IoT technologies to be compatible with open platforms. The proposed component automatic generation tool supports the creation of components considering the characteristics of various hardware devices through the virtual component layer of IoT and AI and enables automatic application to open platforms.

• Journal of Intelligence and Information Systems
• /
• v.9 no.1
• /
• pp.53-69
• /
• 2003

Many process variables in modem manufacturing processes have influence on quality of products with complicated relationships. Therefore, it is necessary to control multiple quality variables in order to monitor abnormal signals in the processes. This study proposes an integrated procedure of self-organizing map (SOM) neural network and case-based reasoning (CBR) for multivariate process control. SOM generates patterns of quality variables. The patterns are compared with the reference patterns in order to decide whether their states are normal or abnormal using the goodness-of-fitness test. For validation, it generates artificial datasets consisting of six patterns, normal and abnormal patterns. Experimental results show that the abnormal patterns can be detected effectively. This study also shows that the CBR procedure enables to keep Type 2 error at very low level and reduce Type 1 error gradually, and then the proposed method can be a solution fur multivariate process control.

• Proceedings of the Korean Society for Bioinformatics Conference
• /
• 2001.10a
• /
• pp.165-196
• /
• 2001

DNA chip 또는 microarray는 다수의 유전자 또는 유전자 조각을 (보통 수천내지 수만 개)칩상에 고정시켜 놓고 DNA hybridization 반응을 이용하여 유전자들의 발현 양상을 분석할 수 있는 기술이다. 이러한 high-throughput기술은 예전에는 생각하지 못했던 여러가지 분자생물학의 문제에 대한 해답을 제시해 줄 수 있을 뿐 만 아니라, 분자수준에서의 질병 진단, 신약 개발, 환경 오염 문제의 해결 등 그 응용 가능성이 무한하다. 이 기술의 실용적인 적용을 위해서는 DNA chip을 제작하기 위한 하드웨어/웻웨어 기술 외에도 이러한 데이터로부터 최대한 유용하고 새로운 지식을 창출하기 위한 bioinformatics 기술이 핵심이라고 할 수 있다. 유전자 발현 패턴을 데이터마이닝하는 문제는 크게 clustering, classification, dependency analysis로 구분할 수 있으며 이러한 기술은 통계학과인공지능 기계학습에 기반을 두고 있다. 주로 사용된 기법으로는 principal component analysis, hierarchical clustering, k-means, self-organizing maps, decision trees, multilayer perceptron neural networks, association rules 등이다. 본 세미나에서는 이러한 기본적인 기계학습 기술 외에 최근에 연구되고 있는 새로운 학습 기술로서 probabilistic graphical model (PGM)을 소개하고 이를 DNA chip 데이터 분석에 응용하는 연구를 살펴본다. PGM은 인공신경망, 그래프 이론, 확률 이론이 결합되어 형성된 기계학습 모델로서 인간 두뇌의 기억과 학습 기작에 기반을 두고 있으며 다른 기계학습 모델과의 큰 차이점 중의 하나는 generative model이라는 것이다. 즉 일단 모델이 만들어지면 이것으로부터 새로운 데이터를 생성할 수 있는 능력이 있어서, 만들어진 모델을 검증하고 이로부터 새로운 사실을 추론해 낼 수 있어 biological data mining 문제에서와 같이 새로운 지식을 발견하는 exploratory analysis에 적합하다. 또한probabilistic graphical model은 기존의 신경망 모델과는 달리 deterministic한의사결정이 아니라 확률에 기반한 soft inference를 하고 학습된 모델로부터 관련된 요인들간의 인과관계(causal relationship) 또는 상호의존관계(dependency)를 분석하기에 적합한 장점이 있다. 군체적인 PGM 모델의 예로서, Bayesian network, nonnegative matrix factorization (NMF), generative topographic mapping (GTM)의 구조와 학습 및 추론알고리즘을소개하고 이를 DNA칩 데이터 분석 평가 대회인 CAMDA-2000과 CAMDA-2001에서 사용된cancer diagnosis 문제와 gene-drug dependency analysis 문제에 적용한 결과를 살펴본다.

• Journal of the Korea Society of Computer and Information
• /
• v.26 no.6
• /
• pp.1-8
• /
• 2021

In this paper, we propose reinforced VGG style network structure for low performance embedded system to classify low resolution infrared image. The combination of reinforced VGG style network structure and global average pooling makes lower computational complexity and higher accuracy. The proposed method classify the synthesize image which have 9 class 3,723,328ea images made from OKTAL-SE tool. The reinforced VGG style network structure composed of 4 filters on input and 16 filters on output from max pooling layer shows about 34% lower computational complexity and about 2.4% higher accuracy then the first parameter minimized network structure made for embedded system composed of 8 filters on input and 8 filters on output from max pooling layer. Finally we get 96.1% accuracy model. Additionally we confirmed the about 31% lower inference lead time in ported C code.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.21 no.1
• /
• pp.93-98
• /
• 2021

Due to the development of deep learning and AI, the scale of the model has grown, and it has been integrated into other fields to blend into our lives. However, in environments with limited resources such as embedded devices, it is exist difficult to apply the model and problems such as power shortages. To solve this, lightweight methods such as clouding or offloading technologies, reducing the number of parameters in the model, or optimising calculations are proposed. In this paper, quantization of learned models is applied to ONNX models used in various framework interchange formats, neural network structure and inference performance are compared with existing models, and various module methods for quantization are analyzed. Experiments show that the size of weight parameter is compressed and the inference time is more optimized than before compared to the original model.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.22 no.6
• /
• pp.305-313
• /
• 2021

The upcoming warfare will be network-centric warfare with the acquiring and sharing of information on the battlefield through the connection of the entire weapon system. Therefore, the amount of information generated increases, but the technology of evaluating the information is insufficient. Threat assessment is a technology that supports a quick decision, but the information has many uncertainties and is difficult to apply to an advanced battlefield. This paper proposes a threat assessment based on artificial intelligence while removing the target uncertainty. The artificial intelligence system used was a fuzzy inference system and a multi-layer perceptron. The target was classified by inputting the unique characteristics of the target into the fuzzy inference system, and the classified target information was input into the multi-layer perceptron to calculate the appropriate threat value. The validity of the proposed technique was verified with the threat value calculated by inputting the uncertain target to the trained artificial neural network.

• The Journal of Korean Institute of Information Technology
• /
• v.17 no.7
• /
• pp.31-45
• /
• 2019

The core technologies for intelligent services today are deep learning, that is neural networks, and parallel and distributed processing technologies such as GPU parallel computing and big data. However, for intelligent services and knowledge sharing services through globally shared ontologies in the future, there is a technology that is better than the neural networks for representing and reasoning knowledge. It is a knowledge representation of IF-THEN in RIF or SWRL, which is the standard rule language of the Semantic Web, and can be inferred efficiently using the rete algorithm. However, when the number of rules processed by the rete algorithm running on a single computer is 100,000, its performance becomes very poor with several tens of minutes, and there is an obvious limitation. Therefore, in this paper, we analyze the past and current studies on parallel and distributed processing of rete algorithm, and examine what aspects should be considered to implement an efficient rete algorithm.

• Journal of Broadcast Engineering
• /
• v.25 no.2
• /
• pp.200-207
• /
• 2020

Recently, deep neural networks such as CNN are showing excellent performance in various fields such as image classification, object recognition, visual quality enhancement, etc. However, as the model size and computational complexity of deep learning models for most applications increases, it is hard to apply neural networks to IoT and mobile environments. Therefore, neural network compression algorithms for reducing the model size while keeping the performance have been being studied. In this paper, we apply few compression methods to CNN models and evaluate their performances in the embedded environment. For evaluate the performance, the classification performance and inference time of the original CNN models and the compressed CNN models on the image inputted by the camera are evaluated in the embedded board equipped with QCS605, which is a customized AI chip. In this paper, a few CNN models of MobileNetV2, ResNet50, and VGG-16 are compressed by applying the methods of pruning and matrix decomposition. The experimental results show that the compressed models give not only the model size reduction of 1.3~11.2 times at a classification performance loss of less than 2% compared to the original model, but also the inference time reduction of 1.2~2.21 times, and the memory reduction of 1.2~3.8 times in the embedded board.

• Journal of Advanced Navigation Technology
• /
• v.13 no.6
• /
• pp.901-906
• /
• 2009

This paper proposes a call admission control(CAC) method for wireless networks, which is based on the upper bound of a possibility distribution of handoff calls dropping rates. The possibility distribution is estimated in a fuzzy inference and a learning algorithm in neural network. The learning algorithm is considered for tuning the membership functions(then parts)of fuzzy rules for the inference. The fuzzy inference method is based on a weighted average of fuzzy sets. The proposed method can avoid estimating excessively large handoff calls dropping rates, and makes possibile self-compensation in real time for the case where the estimated values are smaller than real values. So this method makes secure CAC, thereby guaranteeing the allowed CDR. From simulation studies we show that the estimation performance for the upper bound of call dropping rate is good, and then handoff call dropping rates in CAC are able to be sustained below user's desired value.

• Journal of KIISE
• /
• v.44 no.2
• /
• pp.179-185
• /
• 2017

Document summarization aims to generate a summary that is consistent and contains the highly related sentences in a document. In this study, we implemented for document summarization that extracts highly related sentences from a whole document by considering both similarities and entailment relations between sentences. Accordingly, we proposed a new algorithm, TextRank-NLI, which combines a Recurrent Neural Network based Natural Language Inference model and a Graph-based ranking algorithm used in single document extraction-based summarization task. In order to evaluate the performance of the new algorithm, we conducted experiments using the same datasets as used in TextRank algorithm. The results indicated that TextRank-NLI showed 2.3% improvement in performance, as compared to TextRank.