The evolution of the Distributed Denial of Service Attack(DDoS Attack) method has increased the difficulty in the detection process. One of the solutions to overcome the problems caused by the limitations of the existing victim-side detection method was the source-side detection technique. However, there was a problem of performance degradation due to network traffic irregularities. In order to solve this problem, research has been conducted to detect attacks using a collaborative network between several nodes based on artificial intelligence. Existing methods have shown limitations, especially in nonlinear traffic environments with high Burstness and jitter. To overcome this problem, this paper presents a collaborative source-side DDoS attack detection technique introduced with an attention mechanism. The proposed method aggregates detection results from multiple sources and assigns weights to each region, and through this, it is possible to effectively detect overall attacks and attacks in specific few areas. In particular, it shows a high detection rate with a low false positive of about 6% and a high detection rate of up to 4.3% in a nonlinear traffic dataset, and it can also confirm improvement in attack detection problems in a small number of regions compared to methods that showed limitations in the existing nonlinear traffic environment.

본 연구에서는 얼굴 동영상에서 입술의 움직임과 음성 간의 동기화 탐지 방법을 제안한다. 기존의 연구에서는 얼굴 탐지 기술로 얼굴 영역의 바운딩 박스를 도출하고, 박스의 하단 절반 영역을 시각 인코더의 입력으로 사용하여 입술-음성 동기화 탐지에 필요한 시각적인 특징을 추출하였다. 본 연구에서는 입술-음성 동기화 탐지 모델이 음성 정보의 발화 영역인 입술에 더 집중할 수 있도록 사전 학습된 시각적 Attention 기반의 인코더 도입을 제안한다. 이를 위해 음성 정보 없이 시각적 정보만으로 발화하는 말을 예측하는 독순술(Lip-Reading)에서 사용된 Visual Transformer Pooling(VTP) 모듈을 인코더로 채택했다. 그리고, 제안 방법이 학습 파라미터 수가 적음에도 불구하고 LRS2 데이터 세트에서 다섯 프레임 기준으로 94.5% 정확도를 보임으로써 최근 모델인 VocaList를 능가하는 것을 실험적으로 증명하였다. 또, 제안 방법은 학습에 사용되지 않은 Acappella 데이터셋에서도 VocaList 모델보다 8% 가량의 성능 향상이 있음을 확인하였다.

In this paper, we explore an emotion classification method through multimodal learning utilizing wav2vec 2.0 and KcELECTRA models. It is known that multimodal learning, which leverages both speech and text data, can significantly enhance emotion classification performance compared to methods that solely rely on speech data. Our study conducts a comparative analysis of BERT and its derivative models, known for their superior performance in the field of natural language processing, to select the optimal model for effective feature extraction from text data for use as the text processing model. The results confirm that the KcELECTRA model exhibits outstanding performance in emotion classification tasks. Furthermore, experiments using datasets made available by AI-Hub demonstrate that the inclusion of text data enables achieving superior performance with less data than when using speech data alone. The experiments show that the use of the KcELECTRA model achieved the highest accuracy of 96.57%. This indicates that multimodal learning can offer meaningful performance improvements in complex natural language processing tasks such as emotion classification.

Traditional profanity detection methods have limitations in identifying intentionally altered profanities. This paper introduces a new method based on Named Entity Recognition, a subfield of Natural Language Processing. We developed a profanity detection technique using sequence labeling, for which we constructed a dataset by labeling some profanities in Korean malicious comments and conducted experiments. Additionally, to enhance the model's performance, we augmented the dataset by labeling parts of a Korean hate speech dataset using one of the large language models, ChatGPT, and conducted training. During this process, we confirmed that filtering the dataset created by the large language model by humans alone could improve performance. This suggests that human oversight is still necessary in the dataset augmentation process.

Lipreading is one of the important parts of speech recognition, and several studies have been conducted to improve the performance of lipreading in lipreading systems for speech recognition. Recent studies have used method to modify the model architecture of lipreading system to improve recognition performance. Unlike previous research that improve recognition performance by modifying model architecture, we aim to improve recognition performance without any change in model architecture. In order to improve the recognition performance without modifying the model architecture, we refer to the cues used in human lipreading and set other regions such as chin and cheeks as regions of interest along with the lip region, which is the existing region of interest of lipreading systems, and compare the recognition rate of each region of interest to propose the highest performing region of interest In addition, assuming that the difference in normalization results caused by the difference in interpolation method during the process of normalizing the size of the region of interest affects the recognition performance, we interpolate the same region of interest using nearest neighbor interpolation, bilinear interpolation, and bicubic interpolation, and compare the recognition rate of each interpolation method to propose the best performing interpolation method. Each region of interest was detected by training an object detection neural network, and dynamic time warping templates were generated by normalizing each region of interest, extracting and combining features, and mapping the dimensionality reduction of the combined features into a low-dimensional space. The recognition rate was evaluated by comparing the distance between the generated dynamic time warping templates and the data mapped to the low-dimensional space. In the comparison of regions of interest, the result of the region of interest containing only the lip region showed an average recognition rate of 97.36%, which is 3.44% higher than the average recognition rate of 93.92% in the previous study, and in the comparison of interpolation methods, the bilinear interpolation method performed 97.36%, which is 14.65% higher than the nearest neighbor interpolation method and 5.55% higher than the bicubic interpolation method. The code used in this study can be found a https://github.com/haraisi2/Lipreading-Systems.

Local reservoirs are crucial sources for agricultural water supply, necessitating stable water level management to prepare for extreme climate conditions such as droughts. Water level prediction is significantly influenced by local climate characteristics, such as localized rainfall, as well as seasonal factors including cropping times, making it essential to understand the correlation between input and output data as much as selecting an appropriate prediction model. In this study, extensive multivariate data from over 400 reservoirs in Jeollabuk-do from 1991 to 2022 was utilized to train and validate a water level prediction model that comprehensively reflects the complex hydrological and climatological environmental factors of each reservoir, and to analyze the impact of each input feature on the prediction performance of water levels. Instead of focusing on improvements in water level performance through neural network structures, the study adopts a basic Feedforward Neural Network composed of fully connected layers, batch normalization, dropout, and activation functions, focusing on the correlation between multivariate input data and prediction performance. Additionally, most existing studies only present short-term prediction performance on a daily basis, which is not suitable for practical environments that require medium to long-term predictions, such as 10 days or a month. Therefore, this study measured the water level prediction performance up to one month ahead through a recursive method that uses daily prediction values as the next input. The experiment identified performance changes according to the prediction period and analyzed the impact of each input feature on the overall performance based on an Ablation study