http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Autoencoder-based Anomaly Detection
Khoa Anh Ngo(노안콰),Junhan Kim(김준한),Jiseob Kim(김지섭),Jaseong Koo(구자성),Seungjae Baeck(백승재),Byonghyo Shim(심병효) 한국통신학회 2021 한국통신학회 학술대회논문집 Vol.2021 No.2
Recently, much research has been devoted to developing deep learning-based anomaly detection techniques. If there is a huge amount of data samples, then deep learning models can be trained to provide an end-to-end solution for anomaly detection. However, in many realistic scenarios, the number of anomalies that can be used for training deep learning models is very limited. In this case, deep learning models trained in the manner of supervised learning performs poor, especially when detecting anomalies in the test phase. In this paper, to overcome this limitation, we put forth an autoencoder-based anomaly detection technique. Through the simulations based on the dataset provided by Samsung display, we show that the proposed technique can detect anomalies well.
Khoa D. Nguyen,Son H. Doan,Anh N.V. Ngo,Tung T. Nguyen,Nam T.S. Phan 한국공업화학회 2016 Journal of Industrial and Engineering Chemistry Vol.44 No.-
A metal–organic framework Fe3O(BPDC)3 was synthesized, and used as a productive heterogeneouscatalyst for the direct C–N coupling of azoles with ethers via oxidative C–H activation to produce azolederivatives. The MOF-based catalyst displayed higher catalytic efficiency than many homogeneouscatalysts as well as several MOFs in the transformation. The MOF-based catalyst could be reused manytimes for the synthesis of azole derivatives by the direct C–N coupling of azoles with ethers without anoteworthy deterioration in catalytic efficiency. To the best of our knowledge, this direct C–N couplingreaction was not previously performed in the presence of heterogeneous catalysts.
현대 딥러닝 네트워크의 과신뢰 문제 및 캘리브레이션 기법 연구
김지영(Ji Young Kim),김승년(Seungnyun Kim),김지섭(Jiseob Kim),김진홍(Jinhong Kim),김상태(Sangtae Kim),Khoa Anh Ngo,심병효(Byonghyo Shim) 한국통신학회 2022 한국통신학회 학술대회논문집 Vol.2022 No.2
딥러닝(deep learning) 모델이 사용하는 학습 데이터와 시험 데이터의 도메인(domain)은 일반적으로 같다. 그러나 현실세계(real world)에서는 데이터 수집에 한계가 있으므로 모델은 모든 도메인의 데이터를 학습할 수 없다. 따라서 모델은 학습 데이터와 동일한 도메인의 시험 데이터(familiar data)는 물론, 학습 때 보지 못했던 도메인의 시험 데이터(unfamiliar data)에도 강인해야 하며, 신뢰도 추정이 올바르게 이루어져야 한다. 본 연구진은 familiar/unfamiliar 상황에서의 성별분류(gender recognition) 실험을 진행하고, T-SNE을 분석하였다.