• ETRI Journal
• /
• v.44 no.3
• /
• pp.476-490
• /
• 2022

With the recent advances in technology, automatic speech recognition (ASR) has been widely used in real-world applications. The efficiency of converting large amounts of speech into text accurately with limited resources has become more vital than ever. In this study, we propose a method to rapidly recognize a large speech database via a transformer-based end-to-end model. Transformers have improved the state-of-the-art performance in many fields. However, they are not easy to use for long sequences. In this study, various techniques to accelerate the recognition of real-world speeches are proposed and tested, including decoding via multiple-utterance-batched beam search, detecting end of speech based on a connectionist temporal classification (CTC), restricting the CTC-prefix score, and splitting long speeches into short segments. Experiments are conducted with the Librispeech dataset and the real-world Korean ASR tasks to verify the proposed methods. From the experiments, the proposed system can convert 8 h of speeches spoken at real-world meetings into text in less than 3 min with a 10.73% character error rate, which is 27.1% relatively lower than that of conventional systems.

• International Journal of Computer Science & Network Security
• /
• v.23 no.5
• /
• pp.73-88
• /
• 2023

Brain tumors can also be an abnormal collection or accumulation of cells in the brain that can be life-threatening due to their ability to invade and metastasize to nearby tissues. Accurate diagnosis is critical to the success of treatment planning, and resonant imaging is the primary diagnostic imaging method used to diagnose brain tumors and their extent. Deep learning methods for computer vision applications have shown significant improvements in recent years, primarily due to the undeniable fact that there is a large amount of data on the market to teach models. Therefore, improvements within the model architecture perform better approximations in the monitored configuration. Tumor classification using these deep learning techniques has made great strides by providing reliable, annotated open data sets. Reduce computational effort and learn specific spatial and temporal relationships. This white paper describes transfer models such as the MobileNet model, VGG19 model, InceptionResNetV2 model, Inception model, and DenseNet201 model. The model uses three different optimizers, Adam, SGD, and RMSprop. Finally, the pre-trained MobileNet with RMSprop optimizer is the best model in this paper, with 0.995 accuracies, 0.99 sensitivity, and 1.00 specificity, while at the same time having the lowest computational cost.

• Biomedical Science Letters
• /
• v.29 no.2
• /
• pp.53-57
• /
• 2023

The main objective of pathologists is to achieve accurate lesion diagnoses, which has become increasingly challenging due to the growing number of pathological slides that need to be examined. However, using digital technology has made it easier to complete this task compared to older methods. Digital pathology is a specialized field that manages data from digitized specimen slides, utilizing image processing technology to automate and improve analysis. It aims to enhance the precision, reproducibility, and standardization of pathology-based researches, preclinical, and clinical trials through the sophisticated techniques it employs. The advent of whole slide imaging (WSI) technology is revolutionizing the pathology field by replacing glass slides as the primary method of pathology evaluation. Image processing technology that utilizes WSI is being implemented to automate and enhance analysis. Artificial intelligence (AI) algorithms are being developed to assist pathologic diagnosis and detection and segmentation of specific objects. Application of AI-based digital pathology in biomedical researches is classified into four areas: diagnosis and rapid peer review, quantification, prognosis prediction, and education. AI-based digital pathology can result in a higher accuracy rate for lesion diagnosis than using either a pathologist or AI alone. Combining AI with pathologists can enhance and standardize pathology-based investigations, reducing the time and cost required for pathologists to screen tissue slides for abnormalities. And AI-based digital pathology can identify and quantify structures in tissues. Lastly, it can help predict and monitor disease progression and response to therapy, contributing to personalized medicine.

• Korean Journal of Artificial Intelligence
• /
• v.12 no.1
• /
• pp.17-24
• /
• 2024

This paper introduces a novel approach to time-series estimation for energy load forecasting within Virtual Power Plant (VPP) systems, leveraging advanced artificial intelligence (AI) algorithms, namely Long Short-Term Memory (LSTM) and Seasonal Autoregressive Integrated Moving Average (SARIMA). Virtual power plants, which integrate diverse microgrids managed by Energy Management Systems (EMS), require precise forecasting techniques to balance energy supply and demand efficiently. The paper introduces a hybrid-method forecasting model combining a parametric-based statistical technique and an AI algorithm. The LSTM algorithm is particularly employed to discern pattern correlations over fixed intervals, crucial for predicting accurate future energy loads. SARIMA is applied to generate time-series forecasts, accounting for non-stationary and seasonal variations. The forecasting model incorporates a broad spectrum of distributed energy resources, including renewable energy sources and conventional power plants. Data spanning a decade, sourced from the Korea Power Exchange (KPX) Electrical Power Statistical Information System (EPSIS), were utilized to validate the model. The proposed hybrid LSTM-SARIMA model with parameter sets (1, 1, 1, 12) and (2, 1, 1, 12) demonstrated a high fidelity to the actual observed data. Thus, it is concluded that the optimized system notably surpasses traditional forecasting methods, indicating that this model offers a viable solution for EMS to enhance short-term load forecasting.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.24 no.3
• /
• pp.63-67
• /
• 2024

Recently, artificial intelligence (AI) workloads encompassing various industries such as smart logistics, FinTech, and entertainment are being executed on the cloud. In this paper, we address the scheduling issues of various AI workloads on a multi-tenant cloud system composed of heterogeneous GPU clusters. Traditional scheduling decreases GPU utilization in such environments, degrading system performance significantly. To resolve these issues, we present a new scheduling approach utilizing genetic algorithm-based optimization techniques, implemented within a process-based event simulation framework. Trace driven simulations with diverse AI workload traces collected from Alibaba's MLaaS cluster demonstrate that the proposed scheduling improves GPU utilization compared to conventional scheduling significantly.

• The Journal of the Convergence on Culture Technology
• /
• v.10 no.4
• /
• pp.751-756
• /
• 2024

The accuracy of the model and the explanation of the results are important factors that should be considered simultaneously Recently, applications of explainable artificial intelligence are increasing, and it is especially widely applied in the financial field where interpretation of results is important. In this paper, we compare the performance of open API credit evaluation data using various machine learning techniques. In addition, existing financial logic is verified through explainable artificial intelligence technologies, SHAP and LIME. Accordingly, it is expected to demonstrate the applicability of machine learning in the financial market.

• Smart Structures and Systems
• /
• v.33 no.5
• /
• pp.333-347
• /
• 2024

There has been an increasing interest in the construction of smart buildings that can actively monitor and react to their surroundings. The capacity of these intelligent structures to precisely predict and respond to deflection is a crucial feature that guarantees both their structural soundness and efficiency. Conventional techniques for determining deflection often depend on intricate mathematical models and computational simulations, which may be time- and resource-consuming. Artificial intelligence (AI) algorithms have become a potent tool for anticipating and controlling deflection in intelligent structures in response to these difficulties. The term "deflection-aware smart structures" in this sense refers to constructions that have AI algorithms installed that continually monitor and analyses deflection data in order to proactively detect any problems and take appropriate action. These structures anticipate deflection across a range of operating circumstances and environmental factors by using cutting-edge AI approaches including deep learning, reinforcement learning, and neural networks. AI systems are able to predict real-time deflection with high accuracy by using data from embedded sensors and actuators. This capability enables the systems to identify intricate patterns and linkages. Intelligent buildings have the potential to self-correct in order to reduce deflection and maximize performance. In conclusion, the development of deflection-aware smart structures is a major stride forward for structural engineering and has enormous potential to enhance the performance, safety, and dependability of designed systems in a variety of industries.

• Journal of Intelligence and Information Systems
• /
• v.1 no.1
• /
• pp.23-41
• /
• 1995

Business researchers have traditionally used statistical techniques for classification. In late 1980's, inductive learning started to be used for business classification. Recently, neural network began to be a, pp.ied for business classification. This study reviews the business classification studies, identifies a neural network a, pp.oach as the most powerful classification tool, and discusses the problems and issues in neural network a, pp.ications.

• Journal of Information Processing Systems
• /
• v.13 no.4
• /
• pp.653-667
• /
• 2017

The significant advances in information and communication technologies are changing the process of how information is accessed. The internet is a very important source of information and it influences the development of other media. Furthermore, the growth of digital content is a big problem for academic digital libraries, so that similar tools can be applied in this scope to provide users with access to the information. Given the importance of this, we have reviewed and analyzed several proposals that improve the processes of disseminating information in these university digital libraries and that promote access to information of interest. These proposals manage to adapt a user's access to information according to his or her needs and preferences. As seen in the literature one of the techniques with the best results, is the application of recommender systems. These are tools whose objective is to evaluate and filter the vast amount of digital information that is accessible online in order to help users in their processes of accessing information. In particular, we are focused on the analysis of the fuzzy linguistic recommender systems (i.e., recommender systems that use fuzzy linguistic modeling tools to manage the user's preferences and the uncertainty of the system in a qualitative way). Thus, in this work, we analyzed some proposals based on fuzzy linguistic recommender systems to help researchers, students, and teachers access resources of interest and thus, improve and complement the services provided by academic digital libraries.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.21 no.3
• /
• pp.43-50
• /
• 2021

Recently, researches and projects of companies based on artificial intelligence have been actively carried out. Various services and systems are being grafted with artificial intelligence technology. They become more intelligent. Accordingly, interest in deep learning, one of the techniques of artificial intelligence, and people who want to learn it have increased. In order to learn deep learning, deep learning theory with a lot of knowledge such as computer programming and mathematics is required. That is a high barrier to entry to beginners. Therefore, in this study, we designed and implemented a web-based deep learning platform called DeepBlock, which enables beginners to implement basic models of deep learning such as DNN and CNN without considering programming and mathematics. The proposed DeepBlock can be used for the education of students or beginners interested in deep learning.