http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
뉴로-심볼릭 방법을 사용한 CCTV 감시의 시각적 질의응답 시스템
부석준(Seok-Jun Bu),조성배(Sung-Bae Cho) 한국정보과학회 2023 정보과학회 컴퓨팅의 실제 논문지 Vol.29 No.5
모니터링에 사용되는 장비가 저렴해지고 정교해짐에 따라 분석할 수 있는 CCTV 영상 데이터의 양이 급속히 증가하고 있다. 이상 상황이 발생 시 관리자가 시스템과 시각적 질의응답을 수행하기 위해서는 객체 및 관계의 인식 기능과 언어로 표현된 질의를 처리하는 기능이 동시에 필요하다. 본 논문에서는 실제 도로에서 촬영되는 행인, 차량, 객체들의 관계를 인식하는 고성능 신경망과 자연어 기반의 복잡한 질의 응답을 위한 추론 알고리즘을 결합한 뉴로-심볼릭 질의응답 방법을 제안한다. 이상상황의 판단과 객체인식을 위해 컨볼루션 신경망을 사용하였고, 인식된 객체와 질의의 맵핑 및 응답의 추론을 위해 심볼릭 연산을 수행하다. 제안하는 방법은 실제 CCTV 환경에서 다섯가지 종류의 질의에 대해 99.71%의 응답 정확도를 달성하였다. 추가적으로, 학습된 질의응답 모형을 웹 기반 데모시스템으로 구축하여 실용성을 평가하였다. As monitoring equipment becomes cheaper and more sophisticated, the volume of CCTV video data that can be analyzed will rapidly increase. For an administrator to perform visual question-and-answering with the system when an abnormal situation occurs, recognizing relationships between objects and processing queries expressed in language need to occur simultaneously. This paper proposes a neuro-symbolic question and answer by combining a high-performance neural network that recognizes the relationship between observed objects and an inference algorithm for question and answer based on natural language. A convolutional neural network was used to determine anomalies and object recognition, and a sym-bolic operation was performed for mapping objects and queries to infer responses. The proposed method achieved a 99.71% response accuracy for five queries in the real-world CCTV environment. We also demonstrated the practicality of the pro-posed method by developing a web-based real-time monitoring system.
피싱 URL 분류를 위한 컨볼루션-순환 트리플렛 신경망 기반 웹주소 특징공간의 학습
부석준(Seok-Jun Bu),김혜정(Hae-Jung Kim) Korean Institute of Information Scientists and Eng 2021 정보과학회논문지 Vol.48 No.2
Automated classification of phishing URLs propagated through hyperlinks is critical in environments reinforcing personal connections due to the explosive growth of social media services. Deep learning models for the classification of phishing URLs based on convolutional-recurrent neural networks yielded the best performance in terms of accuracy by modeling the character-level and word-level features. However, the deep learning-based classifier focused on the fitting of a given task via accumulated URLs is limited due to the class imbalance of the phishing attacks that are generated and discarded immediately. We address the class imbalance issue in terms of deep learning-based URL feature space generation task. We propose a modified triplet network structure that explicitly learns the similarity between URLs based on Euclidean distance to alleviate the limitations of the existing deep phishing classifiers. Experiments investigating the real-world dataset of 60,000 URLs collected from web addresses showed the highest performance among the latest deep learning methods, despite the hostile class imbalance. We also demonstrate that the generated URL feature space from the proposed method improved recall by 45.85% compared to the existing methods.