• 한국융합학회논문지
• /
• 제9권6호
• /
• pp.33-38
• /
• 2018

본 논문에서는 단일 클래스만을 학습하여 차량에 대한 새로운 공격을 탐지한다. 분류 성능 평가를 위해 Car-Hacking 데이터셋을 사용한다. Car-Hacking 데이터셋은 실제 차량의 OBD-II 포트를 통해 CAN (Controller Area Network) 트래픽을 로깅하여 생성된다. 이 데이터셋에는 네 가지 공격 유형이 포함된다. 실험에 사용한 단일 클래스 분류기법은 정상 클래스만을 학습하여 비정상인 공격 클래스를 분류해내는 비지도 학습이다. 비지도 학습 방법을 사용하는 경우에 훈련 과정에서 네거티브 인스턴스를 사용하지 않기 때문에 고효율의 분류 성능을 내는 것은 어렵다. 하지만, 비지도 학습은 라벨이 없는 새로운 공격 데이터를 분류하는데 적합한 장점이 있다. 본 연구에서는 네트워크 침입탐지 시스템에서 서명기반의 규칙으로 탐지하기 어려운 새로운 공격 유형을 탐지하기 위해 단일 클래스 분류기를 사용한다. 제안 방법은 새로운 공격을 모두 탐지하고 정상데이터에 대해서도 효율적인 분류 성능을 보이는 파라미터 조합을 제시한다.

• 한국융합학회논문지
• /
• 제9권11호
• /
• pp.15-21
• /
• 2018

본 논문에서는 단일 클래스만을 학습하여 네트워크 침입탐지 시스템 상에서 새로운 비정상 행위를 탐지하는 것을 목표로 한다. 분류 성능 평가를 위해 KDD CUP 1999 데이터셋을 사용한다. 단일 클래스 분류는 정상 클래스만을 학습하여 공격 클래스를 분류해내는 비지도 학습 방법 중 하나이다. 비지도 학습의 경우에는 학습에 네거티브 인스턴스를 사용하지 않기 때문에 상대적으로 높은 분류 효율을 내는 것이 어렵다. 하지만, 비지도 학습은 라벨이 없는 데이터를 분류하는데 적합한 장점이 있다. 본 연구에서는 서포트벡터머신 기반의 단일 클래스 분류기와 밀도 추정 기반의 단일 클래스 분류기를 사용한 실험을 통해 기존에 없던 새로운 공격에 대한 탐지를 한다. 밀도 추정 기반의 분류기를 사용한 실험이 상대적으로 더 좋은 성능을 보였고, 신규 공격에 대해 낮은 FPR을 유지하면서도 약 96%의 탐지율을 보인다.

• 대한산업공학회지
• /
• 제38권4호
• /
• pp.258-261
• /
• 2012

In this paper, I propose a new asset allocation framework to cope with the dynamic nature of the financial market. The investment performance can be much improved by protecting the capital from the market crashes, and such crashes can be pre-identified with high probabilities by regime detection analysis via a specialized unsupervised machine learning technique.

• 한국지능시스템학회논문지
• /
• 제25권1호
• /
• pp.97-104
• /
• 2015

An 'associative cube', a class of auto-associative memories, is revisited here, in which training data and hidden orthogonal basis functions such as wavelet packets or Fourier kernels, are combined in the weight cube. This weight cube has hidden units in its depth, represented by a three dimensional cubic structure. We develop an unsupervised incremental learning mechanism based upon the adaptive least squares method. Training data are mapped into orthogonal basis vectors in a least-squares sense by updating the weights which minimize an energy function. Therefore, a prescribed orthogonal kernel is incrementally assigned to an incoming data. Next, we show how a decoding procedure finds the closest one with a competitive network in the hidden layer. As noisy test data are applied to an associative cube, the nearest one among the original training data are restored in an optimal sense. The simulation results confirm robustness of associative cubes even if test data are heavily distorted by various types of noise.

• 품질경영학회지
• /
• 제49권4호
• /
• pp.581-594
• /
• 2021

Purpose: The purpose of this study is to compare machine learning models for anomaly detection of mechanical facility equipment and suggest an anomaly detection system for mechanical facility equipment in subway stations. It helps to predict failures and plan the maintenance of facility. Ultimately it aims to improve the quality of facility equipment. Methods: The data collected from Daejeon Metropolitan Rapid Transit Corporation was used in this experiment. The experiment was performed using Python, Scikit-learn, tensorflow 2.0 for preprocessing and machine learning. Also it was conducted in two failure states of the equipment. We compared and analyzed five unsupervised machine learning models focused on model Long Short-Term Memory Variational Autoencoder(LSTM-VAE). Results: In both experiments, change in vibration and current data was observed when there is a defect. When the rotating body failure was happened, the magnitude of vibration has increased but current has decreased. In situation of axis alignment failure, both of vibration and current have increased. In addition, model LSTM-VAE showed superior accuracy than the other four base-line models. Conclusion: According to the results, model LSTM-VAE showed outstanding performance with more than 97% of accuracy in the experiments. Thus, the quality of mechanical facility equipment will be improved if the proposed anomaly detection system is established with this model used.

• Smart Structures and Systems
• /
• 제31권5호
• /
• pp.485-500
• /
• 2023

Machine learning-based structural health monitoring (ML-based SHM) methods are researched extensively in the recent decade due to the availability of advanced information and sensing technology. ML methods are well-known for their pattern recognition capability for complex problems. However, the main obstacle of ML-based SHM is that it often requires pre-collected historical data for model training. In most actual scenarios, damage presence can be detected using the unsupervised learning method through anomaly detection, but to further identify the damage types would require prior knowledge or historical events as references. This creates the cold-start problem, especially for new and unobserved structures. Modal-based methods identify damages based on the changes in the structural global properties but often require dense measurements for accurate results. Therefore, a two-stage hybrid modal-machine learning damage detection scheme is proposed. The first stage detects damage presence using Principal Component Analysis-Frequency Response Function (PCA-FRF) in an unsupervised manner, whereas the second stage further identifies the damage. To solve the cold-start problem, mode shape assessment using the first mode is initiated when no trained model is available yet in the second stage. The damage identified by the modal-based method would be stored for future training. This work highlights the performance of the scheme in alleviating the cold-start issue as it transitions through different phases, starting from zero damage sample available. Results showed that single and multiple damages can be identified at an acceptable accuracy level even when training samples are limited.

• 대한후두음성언어의학회지
• /
• 제33권3호
• /
• pp.142-155
• /
• 2022

Diagnosis using voice is non-invasive and can be implemented through various voice recording devices; therefore, it can be used as a screening or diagnostic assistant tool for laryngeal voice disease to help clinicians. The development of artificial intelligence algorithms, such as machine learning, led by the latest deep learning technology, began with a binary classification that distinguishes normal and pathological voices; consequently, it has contributed in improving the accuracy of multi-classification to classify various types of pathological voices. However, no conclusions that can be applied in the clinical field have yet been achieved. Most studies on pathological speech classification using speech have used the continuous short vowel /ah/, which is relatively easier than using continuous or running speech. However, continuous speech has the potential to derive more accurate results as additional information can be obtained from the change in the voice signal over time. In this review, explanations of terms related to artificial intelligence research, and the latest trends in machine learning and deep learning algorithms are reviewed; furthermore, the latest research results and limitations are introduced to provide future directions for researchers.

• 센서학회지
• /
• 제32권6호
• /
• pp.378-385
• /
• 2023

Human biosignals provide essential information for diagnosing diseases such as dementia and Parkinson's disease. Owing to the shortcomings of current clinical assessments, noninvasive solutions are required. Machine learning (ML) on wearable sensor data is a promising method for the real-time monitoring and early detection of abnormalities. ML facilitates disease identification, severity measurement, and remote rehabilitation by providing continuous feedback. In the context of wearable sensor technology, ML involves training on observed data for tasks such as classification and regression with applications in clinical metrics. Although supervised ML presents challenges in clinical settings, unsupervised learning, which focuses on tasks such as cluster identification and anomaly detection, has emerged as a useful alternative. This review examines and discusses a variety of ML algorithms such as Support Vector Machines (SVM), Random Forests (RF), Decision Trees (DT), Neural Networks (NN), and Deep Learning for the analysis of complex clinical data.

• Nuclear Engineering and Technology
• /
• 제55권2호
• /
• pp.603-622
• /
• 2023

Human error (HE) is an important concern in safety-critical systems such as nuclear power plants (NPPs). HE has played a role in many accidents and outage incidents in NPPs. Despite the increased automation in NPPs, HE remains unavoidable. Hence, the need for HE detection is as important as HE prevention efforts. In NPPs, HE is rather rare. Hence, anomaly detection, a widely used machine learning technique for detecting rare anomalous instances, can be repurposed to detect potential HE. In this study, we develop an unsupervised anomaly detection technique based on generative adversarial networks (GANs) to detect anomalies in manually collected surveillance data in NPPs. More specifically, our GAN is trained to detect mismatches between automatically recorded sensor data and manually collected surveillance data, and hence, identify anomalous instances that can be attributed to HE. We test our GAN on both a real-world dataset and an external dataset obtained from a testbed, and we benchmark our results against state-of-the-art unsupervised anomaly detection algorithms, including one-class support vector machine and isolation forest. Our results show that the proposed GAN provides improved anomaly detection performance. Our study is promising for the future development of artificial intelligence based HE detection systems.

• 산업기술연구
• /
• 제33권A호
• /
• pp.31-39
• /
• 2013

There are two methods to make a distinction of deterioration of high-speed railway track. One is that an administrator checks for each attribute value of track induction data represented in graph and determines whether maintenance is needed or not. The other is that an administrator checks for monthly trend of attribute value of the corresponding section and determines whether maintenance is needed or not. But these methods have a weak point that it takes longer times to make decisions as the amount of track induction data increases. As a field of artificial intelligence, the method that a computer makes a distinction of deterioration of high-speed railway track automatically is based on machine learning. Types of machine learning algorism are classified into four type: supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning. This research uses supervised learning that analogizes a separating function form training data. The method suggested in this research uses SVM classifier which is a main type of supervised learning and shows higher efficiency binary classification problem. and it grasps the difference between two groups of data and makes a distinction of deterioration of high-speed railway track.