• KIPS Transactions on Software and Data Engineering
• /
• v.12 no.7
• /
• pp.303-314
• /
• 2023

Voice conversion, a technology that allows an individual's speech data to be regenerated with the acoustic properties(tone, cadence, gender) of another, has countless applications in education, communication, and entertainment. This paper proposes an approach based on the StarGAN-VC model that generates realistic-sounding speech without requiring parallel utterances. To overcome the constraints of the existing StarGAN-VC model that utilizes one-hot vectors of original and target speaker information, this paper extracts feature vectors of target speakers using a pre-trained version of Rawnet3. This results in a latent space where voice conversion can be performed without direct speaker-to-speaker mappings, enabling an any-to-any structure. In addition to the loss terms used in the original StarGAN-VC model, Wasserstein distance is used as a loss term to ensure that generated voice segments match the acoustic properties of the target voice. Two Time-Scale Update Rule (TTUR) is also used to facilitate stable training. Experimental results show that the proposed method outperforms previous methods, including the StarGAN-VC network on which it was based.

• Journal of Internet Computing and Services
• /
• v.25 no.4
• /
• pp.47-56
• /
• 2024

This paper proposes a method for fine-tuning BERT-based natural language processing models to perform sentiment analysis on Korean review data. By varying the input sequence length during this process and comparing the performance, we aim to explore the optimal performance according to the input sequence length. For this purpose, text review data collected from the clothing shopping platform M was utilized. Through web scraping, review data was collected. During the data preprocessing stage, positive and negative satisfaction scores were recalibrated to improve the accuracy of the analysis. Specifically, the GPT-4 API was used to reset the labels to reflect the actual sentiment of the review texts, and data imbalance issues were addressed by adjusting the data to 6:4 ratio. The reviews on the clothing shopping platform averaged about 12 tokens in length, and to provide the optimal model suitable for this, five BERT-based pre-trained models were used in the modeling stage, focusing on input sequence length and memory usage for performance comparison. The experimental results indicated that an input sequence length of 64 generally exhibited the most appropriate performance and memory usage. In particular, the KcELECTRA model showed optimal performance and memory usage at an input sequence length of 64, achieving higher than 92% accuracy and reliability in sentiment analysis of Korean review data. Furthermore, by utilizing BERTopic, we provide a Korean review sentiment analysis process that classifies new incoming review data by category and extracts sentiment scores for each category using the final constructed model.

• KOREAN JOURNAL OF CROP SCIENCE
• /
• v.66 no.2
• /
• pp.105-111
• /
• 2021

Rice lodging is an annual occurrence caused by typhoons accompanied by strong winds and strong rainfall, resulting in damage relating to pre-harvest sprouting during the ripening period. Thus, rapid estimations of the area of lodged rice are necessary to enable timely responses to damage. To this end, we obtained images related to rice lodging using a drone in Gimje, Buan, and Gunsan, which were converted to 128 × 128 pixels images. A convolutional neural network (CNN) model, a deep learning model based on these images, was used to predict rice lodging, which was classified into two types (lodging and non-lodging), and the images were divided in a 8:2 ratio into a training set and a validation set. The CNN model was layered and trained using three optimizers (Adam, Rmsprop, and SGD). The area of rice lodging was evaluated for the three fields using the obtained data, with the exception of the training set and validation set. The images were combined to give composites images of the entire fields using Metashape, and these images were divided into 128 × 128 pixels. Lodging in the divided images was predicted using the trained CNN model, and the extent of lodging was calculated by multiplying the ratio of the total number of field images by the number of lodging images by the area of the entire field. The results for the training and validation sets showed that accuracy increased with a progression in learning and eventually reached a level greater than 0.919. The results obtained for each of the three fields showed high accuracy with respect to all optimizers, among which, Adam showed the highest accuracy (normalized root mean square error: 2.73%). On the basis of the findings of this study, it is anticipated that the area of lodged rice can be rapidly predicted using deep learning.

• Journal of Internet Computing and Services
• /
• v.23 no.3
• /
• pp.13-20
• /
• 2022

IEEE 802.11ay Wi-Fi is the next generation wireless technology and operates in mmWave band. It supports the MU-MIMO (Multiple User Multiple Input Multiple Output) transmission in which an AP (Access Point) can transmit multiple data streams simultaneously to multiple STAs (Stations). To this end, the AP should perform MU-MIMO beamforming training with the STAs. For efficient MU-MIMO beamforming training, it is important for the AP to estimate signal strength measured at each STA at which multiple beams are used simultaneously. Therefore, in the paper, we propose a deep learning-based link quality estimation scheme. Our proposed scheme estimates the signal strength with high accuracy by utilizing a deep learning model pre-trained for a certain indoor or outdoor propagation scenario. Specifically, to estimate the signal strength of the multiple concurrent beams, our scheme uses the signal strengths of the respective single beams, which can be obtained without additional signaling overhead, as the input of the deep learning model. For performance evaluation, we utilized a Q-D (Quasi-Deterministic) Channel Realization open source software and extensive channel measurement campaigns were conducted with NIST (National Institute of Standards and Technology) to implement the millimeter wave (mmWave) channel. Our simulation results demonstrate that our proposed scheme outperforms comparison schemes in terms of the accuracy of the signal strength estimation.

• Journal of Intelligence and Information Systems
• /
• v.29 no.3
• /
• pp.145-165
• /
• 2023

Recommendation system plays a significant role on relieving difficulties of selecting information among rapidly increasing amount of information caused by the development of the Internet and on efficiently displaying information that fits individual personal interest. In particular, without the help of recommendation system, E-commerce and OTT companies cannot overcome the long-tail phenomenon, a phenomenon in which only popular products are consumed, as the number of products and contents are rapidly increasing. Therefore, the research on recommendation systems is being actively conducted to overcome the phenomenon and to provide information or contents that are aligned with users' individual interests, in order to induce customers to consume various products or contents. Usually, collaborative filtering which utilizes users' historical behavioral data shows better performance than contents-based filtering which utilizes users' preferred contents. However, collaborative filtering can suffer from cold-start problem which occurs when there is lack of users' historical behavioral data. In this paper, hybrid music recommendation system, which can solve cold-start problem, is proposed based on the playlist data of Melon music streaming service that is given by Kakao Arena for music playlist continuation competition. The goal of this research is to use music tracks, that are included in the playlists, and metadata of music tracks and playlists in order to predict other music tracks when the half or whole of the tracks are masked. Therefore, two different recommendation procedures were conducted depending on the two different situations. When music tracks are included in the playlist, LightFM is used in order to utilize the music track list of the playlists and metadata of each music tracks. Then, the result of Item2Vec model, which uses vector embeddings of music tracks, tags and titles for recommendation, is combined with the result of LightFM model to create final recommendation list. When there are no music tracks available in the playlists but only playlists' tags and titles are available, recommendation was made by finding similar playlists based on playlists vectors which was made by the aggregation of FastText pre-trained embedding vectors of tags and titles of each playlists. As a result, not only cold-start problem can be resolved, but also achieved better performance than ALS, BPR and Item2Vec by using the metadata of both music tracks and playlists. In addition, it was found that the LightFM model, which uses only artist information as an item feature, shows the best performance compared to other LightFM models which use other item features of music tracks.

• Journal of Korea Water Resources Association
• /
• v.57 no.7
• /
• pp.437-449
• /
• 2024

Rainfall-runoff prediction studies using deep learning while considering catchment attributes have been gaining attention. In this study, we selected two models: the Transformer model, which is suitable for large-scale data training through the self-attention mechanism, and the LSTM-based multi-state-vector sequence-to-sequence (LSTM-MSV-S2S) model with an encoder-decoder structure. These models were constructed to incorporate catchment attributes and predict the inflow of 10 multi-purpose dam watersheds in South Korea. The experimental design consisted of three training methods: Single-basin Training (ST), Pretraining (PT), and Pretraining-Finetuning (PT-FT). The input data for the models included 10 selected watershed attributes along with meteorological data. The inflow prediction performance was compared based on the training methods. The results showed that the Transformer model outperformed the LSTM-MSV-S2S model when using the PT and PT-FT methods, with the PT-FT method yielding the highest performance. The LSTM-MSV-S2S model showed better performance than the Transformer when using the ST method; however, it showed lower performance when using the PT and PT-FT methods. Additionally, the embedding layer activation vectors and raw catchment attributes were used to cluster watersheds and analyze whether the models learned the similarities between them. The Transformer model demonstrated improved performance among watersheds with similar activation vectors, proving that utilizing information from other pre-trained watersheds enhances the prediction performance. This study compared the suitable models and training methods for each multi-purpose dam and highlighted the necessity of constructing deep learning models using PT and PT-FT methods for domestic watersheds. Furthermore, the results confirmed that the Transformer model outperforms the LSTM-MSV-S2S model when applying PT and PT-FT methods.

• Journal of Bio-Environment Control
• /
• v.32 no.4
• /
• pp.434-441
• /
• 2023

Determining the size or area of a plant's leaves is an important factor in predicting plant growth and improving the productivity of indoor farms. In this study, we developed a convolutional neural network (CNN)-based model to accurately predict the length and width of lettuce leaves using photographs of the leaves. A callback function was applied to overcome data limitations and overfitting problems, and K-fold cross-validation was used to improve the generalization ability of the model. In addition, ImageDataGenerator function was used to increase the diversity of training data through data augmentation. To compare model performance, we evaluated pre-trained models such as VGG16, Resnet152, and NASNetMobile. As a result, NASNetMobile showed the highest performance, especially in width prediction, with an R_squared value of 0.9436, and RMSE of 0.5659. In length prediction, the R_squared value was 0.9537, and RMSE of 0.8713. The optimized model adopted the NASNetMobile architecture, the RMSprop optimization tool, the MSE loss functions, and the ELU activation functions. The training time of the model averaged 73 minutes per Epoch, and it took the model an average of 0.29 seconds to process a single lettuce leaf photo. In this study, we developed a CNN-based model to predict the leaf length and leaf width of plants in indoor farms, which is expected to enable rapid and accurate assessment of plant growth status by simply taking images. It is also expected to contribute to increasing the productivity and resource efficiency of farms by taking appropriate agricultural measures such as adjusting nutrient solution in real time.

• Physical Therapy Korea
• /
• v.14 no.3
• /
• pp.48-56
• /
• 2007

Environmental Enrichment (EE) alone is not capable of enhancing the fine digit and the forelimb functions. Therefore, we applied modified constraint-induced movement therapy (mCIMT) under the influence of EE to assess its effect on promoting improved forelimb sensorimotor functions. Focal ischemic brain injury was produced in Sprague-Dawley rats (60 rats, $250{\pm}50$ g) through middle cerebral artery occlusion (MCAO). Before MCAO induction, all rats were trained in modified limb placing tests and reaching tasks for 1 week. Then they were randomly divided into three groups: Group I: application of standard environment (SE) after MCAO induction (n=20), Group II: application of EE after MCAO induction (n=20), Group III: MCAO+EE, mCIMT and task-oriented training that was initiated at 10th day after MCAO induction (n=20). We also applied mCIMT (between 9 AM and 5 PM/daily) which included restraining the forelimb ipsilateral to the lesion using the 'Jones & Schallert' method. We assessed the change of modified limb placing, single pellet reaching test and the immunoreactivity of BDNF by immunohistochemistry (pre, 1st, 5th, 10th and 20th day). Group I showed no improved outcome, whereas group II and III significantly improved on the use of the forelimb and the immunoreactivity. The qualitative analysis of the skilled reaching test, of group III showed the greatest improvement in the fine digit and the forelimb function. These results suggest that EE combined with mCIMT is more functional in promoting enhanced fine digit and forelimb functional movements.

• KIPS Transactions on Software and Data Engineering
• /
• v.10 no.6
• /
• pp.235-242
• /
• 2021

In this work, we develop a Korean dependency parser based on a stack-pointer network that consists of a pointer network and an internal stack. The parser has an encoder and decoder and builds a dependency tree for an input sentence in a depth-first manner. The encoder of the parser encodes an input sentence, and the decoder selects a child for the word at the top of the stack at each step. Since the parser has the internal stack where a search path is stored, the parser can utilize information of previously derived subtrees when selecting a child node. Previous studies used only a grandparent and the most recently visited sibling without considering a subtree structure. In this paper, we introduce graph attention networks that can represent a previously derived subtree. Then we modify our parser based on the stack-pointer network to utilize subtree information produced by the graph attention networks. After training the dependency parser using Sejong and Everyone's corpus, we evaluate the parser's performance. Experimental results show that the proposed parser achieves better performance than the previous approaches at sentence-level accuracies when adopting 2-depth graph attention networks.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.40 no.3
• /
• pp.239-247
• /
• 2022

Cities are becoming more complex due to rapid industrialization and population growth in modern times. In particular, urban areas are rapidly changing due to housing site development, reconstruction, and demolition. Thus accurate road information is necessary for various purposes, such as High Definition Map for autonomous car driving. In the case of the Republic of Korea, accurate spatial information can be generated by making a map through the existing map production process. However, targeting a large area is limited due to time and money. Road, one of the map elements, is a hub and essential means of transportation that provides many different resources for human civilization. Therefore, it is essential to update road information accurately and quickly. This study uses Semantic Segmentation algorithms Such as LinkNet, D-LinkNet, and NL-LinkNet to extract roads from drone images and then apply hyperparameter optimization to models with the highest performance. As a result, the LinkNet model using pre-trained ResNet-34 as the encoder achieved 85.125 mIoU. Subsequent studies should focus on comparing the results of this study with those of studies using state-of-the-art object detection algorithms or semi-supervised learning-based Semantic Segmentation techniques. The results of this study can be applied to improve the speed of the existing map update process.