Digital Library[ Search Result ]
Analyzing Model Hubs for Effective Composition of Pre-Trained Machine Learning Models
http://doi.org/10.5626/JOK.2025.52.1.42
Deep Neural Network (DNN) models have become prevalent. They are increasingly adopted as components in software systems. Designing and training these DNNs from scratch is not trivial. Designing requires domain expertise and familiarity with DNN frameworks while training necessitates substantial computational resources and large training datasets. Following the philosophy of traditional software engineering, developers often reuse Pre-Trained Models (PTMs) organized in model hubs. However, challenges arise when PTMs that match a developer’s specific requirements are lacking. In this paper, we explored the concept of PTM composition and investigated whether a combination of PTMs could fulfill application requirements without needing fine-tuning or creating a new DNN. We present current challenges in PTM composition through our case study and identified shortcomings of existing model hubs. By drawing parallels between PTM composition and web service composition, we highlighted essential technologies required for successful PTM composition and discussed potential solutions to these issues.
Vision-based Position Deviation Fault Injection Method for Building a Collaborative Robot Motion Fault Dataset
Donghee Yun, Dongyeon Yoo, Jungwon Lee
http://doi.org/10.5626/JOK.2023.50.9.795
The data-based fault detection method, which collects data from internal and external sensors in real-time and predicts fault, is being applied to collaborative robots, which are key facilities in smart factories. The data-based fault detection method requires a large amount of data for learning, and in particular, a large amount of data labeled as a fault state is essential. However, it is difficult to obtain large amounts of actual fault data in industrial settings. Therefore, in this study, the output of the collaborative robot fault state based on a vision sensor was analyzed and compared with the output of the normal state, and a fault injection method was proposed based on the deviation between the analyzed output signals. Collaborative robot data collected in the actual fault state could be replaced with data collected in the proposed fault injection state. The comparison of the performance of the model trained with fault injection data and trained with actual fault data confirmed that there was almost no difference, with an average of 0.97 and 0.98 accuracy, thus verifying the effectiveness of the proposed fault injection method.
Comparative Analysis of Accuracy and Stability of Software Reliability Estimation Models based on Recurrent Neural Networks
Taehyoun Kim, Duksan Ryu, Jongmoon Baik
http://doi.org/10.5626/JOK.2023.50.8.688
Existing studies on software reliability estimation based on recurrent neural networks have used networks to create one model under the same conditions and evaluated the accuracy of the model. However, due to the randomness of artificial neural networks, such recurrent neural networks can generate different training results of models even under the same conditions, which can lead to inaccurate software reliability estimation. Therefore, this paper compares and analyzes which recurrent neural networks could estimate software reliability more stably and accurately. We estimated software reliability in eight real projects using three representative recurrent neural networks and compared and analyzed the performances of these models in terms of accuracy and stability. As a result, Long Short-Term Memory showed the most stable and accurate software reliability estimation performance. A more accurate and stable software reliability estimation model is expected to be selected based on the results of this study.
RESEDA: Software REliability Model SElection using DAta-driven Software Reliability Prediction
Nakwon Lee, Duksan Ryu, Ilhoon Cho, Jeakun Song, Jongmoon Baik
http://doi.org/10.5626/JOK.2022.49.6.443
To solve the model generalization problem, i.e., there is no single best model that fits all types of software failure data, model selection techniques and data-driven reliability prediction techniques have been proposed. However, model selection techniques still wrongly select some failure data, and the reliability metrics that the data-driven techniques can observe are limited. In this paper, we propose a software reliability model selection technique using data-driven reliability prediction to improve the prediction accuracy with obtaining reliability metrics. The proposed approach decides either selection or data-driven for target failure data using a classifier generated from historical failure data sets. If data-driven is chosen, the proposed approach builds an augmented failure data using the prediction results of the data-driven technique and selects a model for the augmented data. The proposed approach shows a 21% lower median value of the mean error of prediction compared to the best technique for comparison. With the improved reliability prediction accuracy using the proposed approach, the higher software reliability is achieved.
A Method for Cancer Prognosis Prediction Using Gene Embedding
http://doi.org/10.5626/JOK.2021.48.7.842
Identifying prognostic genes and using them to predict the prognosis of cancer patients can help provide them with more effective treatments. Many methods have been proposed to identify prognostic genes and predict cancer prognosis, and recent studies have focused on machine learning methods including deep learning. However, applying gene expression data to machine learning methods has the limitations of a small number of samples and a large number of genes. In this study, we additionally use a gene network to generate many random gene paths, which we used for training the model, thereby compensating for the small sample problem. We identified the prognostic genes and predicted the prognosis of patients using the gene expression data and gene networks for five cancer types and confirmed that the proposed method showed better predictive accuracy compared to other existing methods, and good performance on small sample data.
Predicting Chemical Structure of Drugs Using Deep Learning
Soohyun Ko, Chihyun Park, Jaegyoon Ahn
http://doi.org/10.5626/JOK.2021.48.2.234
Numerous computer-based methods have been investigated in attempts to reduce the time and cost of drug development. In particular, with the recent development of deep learning techniques, various generation models for generating the chemical formulas of candidate compounds and reinforcement learning models to generate chemical formulas that satisfy specific conditions have been presented. In this paper, we propose a reinforcement learning model that exploits predicted binding affinity information between specific proteins and generated compounds. More specifically, the generative model used in this paper is Stack-RNN, and reinforcement learning is implemented by using Stack-RNN as a policy to ensure that the generated formula has specific chemical properties and high binding affinity with specific proteins. The proposed model generates paper, we generated the chemical formulas of compounds that are similar to three anti-cancer drugs (Sorafenib, Sunitinib, and Dasatinib) by using the target protein information of these three anti-cancer drugs.
Reducing the Learning Time of Code Change Recommendation System Using Recurrent Neural Network
Byeong-il Bae, Sungwon Kang, Seonah Lee
http://doi.org/10.5626/JOK.2020.47.10.948
Since code change recommendation systems select and recommend files that needing modifications, they help developers save time spent on software system evolution. However, these recommendation systems generally spend a significant amount of time in learning accumulated data and relearning whenever new data are accumulated. This study proposes a method to reduce the time spent on learning when using Code change Recommendation System using Recurrent Neural Network (RNN-CRS), which works by avoiding the learning that is unlikely to contribute to new knowledge. For the five products used in the experimental evaluation, our proposed method reduced the time to relearn data and re-generate a learning model by as much as 49.08%-68.15%, and by 10.66% in the least effective case, compared to the existing method.
The Method Using Reduced Classification Models for Distributed Processing of CNN Models in Multiple Edge Devices
Junyoung Kim, Jongho Jeon, Minkwan Kee, Gi-Ho Park
http://doi.org/10.5626/JOK.2020.47.8.787
Recently, there have been increasing demands for edge computing that processes data at the end of the network wherein data is collected because of various problems such as network load caused by a large amount of data transfer to a cloud server. However, it is difficult for edge devices to use deep learning applications used in cloud servers because most edge devices at the end of the network have limited performance. To overcome these problems, this paper proposes a distributed processing method that uses reduced classification models to jointly perform inferences on multiple edge devices. The reduced classification models have compressed model weights, and perform inferences for some parts of the total classification labels. The experimental results confirmed that the accuracy of the result of the proposed distributed processing method is similar to the accuracy of the result of the original model, even if the proposed reduced classification models had much less parameters than those of the original model.
Image Caption Generation using Object Attention Mechanism
http://doi.org/10.5626/JOK.2019.46.4.369
Explosive increases in image data have led studies investigating the role of image caption generation in image expression of natural language. The current technologies for generating Korean image captions contain errors associated with object concurrence attributed to dataset translation from English datasets. In this paper, we propose a model of image caption generation employing attention as a new loss function using the extracted nouns of image references. The proposed method displayed BLEU1 0.686, BLEU2 0.557, BLEU3 0.456, BLEU4 0.372, which proves that the proposed model facilitates the resolution of high-frequency word-pair errors. We also showed that it enhances the performance compared with previous studies and reduces redundancies in the sentences. As a result, the proposed method can be used to generate a caption corpus effectively.
Diagnostic and Therapeutic Model for Korean Major Depressive Disorder Using Multi-Modal Data
Yonghwa Choi, Aram Kim, Minji Jeon, Sunkyu Kim, Kyu-Man Han, Eunsoo Won, Byung-Joo Ham, Jaewoo Kang
http://doi.org/10.5626/JOK.2019.46.1.71
Depression is one of the most common mental illnesses in the modern society, and it increases the social burden due to repeated recurrences. However, since there are many pre-disposing factors that cause depression, there is need to develop a machine-learning model that examine these factors effectively. In this paper, we propose a model that can diagnose depression and predict the degree of antidepressant response using four multi modal data including basic information, MRI, genetics, and cognitive test. The model achieved 0.923 AUROC score for diagnosis and 0.08 MSE for prediction of antidepressant response. In addition, the results of the proposed model were quantitatively analyzed, and it confirmed that accurate diagnosis and drug response prediction are possible when the patient’s data is added. Qualitative analysis was also conducted to provide new hypotheses as well as findings on the main factors causing depression.
Search
Journal of KIISE
- ISSN : 2383-630X(Print)
- ISSN : 2383-6296(Electronic)
- KCI Accredited Journal
Editorial Office
- Tel. +82-2-588-9240
- Fax. +82-2-521-1352
- E-mail. chwoo@kiise.or.kr