• The Journal of the Acoustical Society of Korea
• /
• v.40 no.5
• /
• pp.479-487
• /
• 2021

In this paper, we propose an Sound Event Detection (SED) model using self-training based on a noisy student model. The proposed SED model consists of two stages. In the first stage, a mean-teacher model based on an Residual Convolutional Recurrent Neural Network (RCRNN) is constructed to provide target labels regarding weakly labeled or unlabeled data. In the second stage, a self-training-based noisy student model is constructed by applying different noise types. That is, feature noises, such as time-frequency shift, mixup, SpecAugment, and dropout-based model noise are used here. In addition, a semi-supervised loss function is applied to train the noisy student model, which acts as label noise injection. The performance of the proposed SED model is evaluated on the validation set of the Detection and Classification of Acoustic Scenes and Events (DCASE) 2020 Challenge Task 4. The experiments show that the single model and ensemble model of the proposed SED based on the noisy student model improve F1-score by 4.6 % and 3.4 % compared to the top-ranked model in DCASE 2020 challenge Task 4, respectively.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.11
• /
• pp.1608-1614
• /
• 2022

As competition between universities intensifies due to the recent decrease in the number of students, it is recognized as an essential task of universities to predict students who are underperforming at an early stage and to make various efforts to prevent dropouts. For this, a high-performance model that accurately predicts student performance is essential. This paper proposes a method to improve prediction performance by removing or amplifying abnormal data in a classification prediction model for identifying underperforming students. Existing anomaly data processing methods have mainly focused on deleting or ignoring data, but this paper presents a criterion to distinguish noise from change indicators, and contributes to improving the performance of predictive models by deleting or amplifying data. In an experiment using open learning performance data for verification of the proposed method, we found a number of cases in which the proposed method can improve classification performance compared to the existing method.