論文發表 

最佳論文

IVA-PBC: A Smart Virtual Assistant Service with CHT-PBC Framework for Cloud Data Centers
  • 會議名稱:
  • IEEE ICKII 2024
  • 論文摘要:
  • This study introduces IVA-PBC, an innovative intelligent virtual assistant designed specifically for cloud data centers. Developed on the CHT-PBC framework, IVA-PBC utilizes a cloud-native architecture. It offers lightweight and highly packaged business functionalities, integrating cohesive and loosely coupled functional modules to achieve agile development goals. Additionally, the modular nature of CHT-PBC allows for seamless integration among PBC instances, thus enabling enhanced capabilities. By integrating IVA-PBC with existing cloud IDC management PBC, it facilitates the development of value-added applications, thereby establishing a cutting-edge 3D intelligent assistant service that provides domain expertise and business support within a cloud IDC management PBC. This study also provides an overview of the CHT-PBC framework. With the successful implementation of the CHT-PBC framework by Chunghwa Telecom, this architecture holds significant potential for adoption across various industries, particularly as businesses actively embrace digital transformation.
FUN-VFI: Flow-Upscaling Network for 4K Video Frame Interpolation
  • 會議名稱:
  • 2024 2nd International Conference on Artificial Intelligence and Power Engineering
  • 論文摘要:
  • 4K video frame interpolation is challenging due to the large motion and unexpected occlusion between two consecutive frames. To address this challenge, we propose the Flow Upscaling Network for 4K Video Frame Interpolation, denoted as FUN-VFI. Specifically, it consists of two main modules, global flow flied estimation and adjustable upscaler. In global flow field estimation, the flow estimator predicts bilateral flow fields between two downscaled consecutive frames. Subsequently, we iteratively enhance these bilateral flow fields using the adjustable upscaler. the core of upscaler is ConvNext block, a pure CNN model that can rival vision transformers while preserving the simple structure and efficient runtime of convolution neural networks. Last, we warp two consecutive input frames using refined flow fields and blend them to generate the reconstructed intermediate frame. Experimental results demonstrate that proposed method achieves outstanding performance across diverse benchmark datasets. Furthermore, compared with existing methods like M2M-PWC and BI-Former, FUN-VFI accelerates inference times by approximately 2-6 times at 4K resolution, while consistently delivering superior outcomes.
Energy Monitoring for Containerized Network Service in Telco Cloud
  • 會議名稱:
  • IEEE International Conference on Electronic Communications, Internet of Things and Big Data Conference 2024 (IEEE ICEIB 2024)
  • 論文摘要:
  • Traditional virtualized services are gradually transitioning to containerized services with telecom shifts towards cloud-native. Additionally, governments have been actively striving to improve energy consumption efficiency to reduce environmental impact in recent years. Therefore, managing the power consumption of containerized network services and achieving energy-saving benefits has attracted attention in telecom, such as containerized network functions. This paper proposes an architecture and method for monitoring the power consumption of containerized network services in the telco cloud, aiming to measure the utilization of telecom containerized network services and quantify power consumption to support telecom service auto-scaling and equipment power dynamic control.
5G Handover Analysis in Real Network
  • 會議名稱:
  • IEEE ICEIB 2023 - International Conference on Electronic Communications, Internet of Things and Big Data
  • 論文摘要:
  • Handover (HO) is the key function to maintaining users’ connections while moving within the coverage of cellular communication networks. During the HO process, it is possible to decrease the data throughput and cause interruptions of time-critical services. In addition, the HO signal processes between the mobile phone and the mobile network increase energy consumption for both of them. These problems are even more complex in the 5G era because of the co-existence of macro- and micro-cell (ultra-dense small cells), and different deployment architectures, such as non-standalone (NSA) and standalone (SA). To investigate HO behaviors in real 5G networks, we collected a rich mobile signal dataset consisting of 2 sets of repeated driving trips under a 5G commercial NSA network in Taoyuan, Taiwan. Based on the dataset, we analyze and compare the HO sequential frequent patterns and the probability of the occurrence of ping-pong HO for the 2 driving routes. The results show the HO frequent patterns with high support can be found, and several ping-pong events occur repeatedly at the same position. Several observations are described to discuss the value of utilizing the dataset and design AI-assisted HO algorithms for decreasing ping-pong effects and unnecessary HO events.

SCI期刊(2022-2024)

Frequency-Aware Axial-ShiftedNet in Generative Adversarial Networks for Visible-to-Infrared Image Translation
  • 期刊名稱:
  • IEEE Access
  • 論文摘要:
  • Infrared imagery is indispensable for capturing temperature data by detecting infrared radiation, particularly in challenging environments characterized by low-light conditions where visual perception is compromised. As a result, there has been considerable interest in the conversion of visible images into their infrared counterparts. In this research, we present the Freq-ShiftedNet model, which employs an adversarial generative network approach for training. By harnessing the power of the Haar wavelet transform, we adeptly preserve frequency information, directing low-frequency features to the Decoder and high-frequency features to the Encoder. Analysis of the KAIST dataset demonstrates that our model outperforms the InfraGAN, achieving a Structural Similarity (SSIM) score of 0.825, marking a 5.4% improvement, and a Learned Perceptual Image Patch Similarity (LPIPS) score of 0.228, indicating a 41.3% decrease. These findings underscore the effectiveness of our proposed generator, the integration of wavelet features into the Freq-ShiftedNet model, and its suitability for real-world applications.
Deep Learning-Based Surgical Treatment Recommendation and Nonsurgical Prognosis Status Classification for Scaphoid Fractures by Automated X-ray Image Recognition
  • 期刊名稱:
  • Biomedicines
  • 論文摘要:
  • Biomedical information retrieval for diagnosis, treatment and prognosis has been studied for a long time. In particular, image recognition using deep learning has been shown to be very effective for cancers and diseases. In these fields, scaphoid fracture recognition is a hot topic because the appearance of scaphoid fractures is not easy to detect. Although there have been a number of recent studies on this topic, no studies focused their attention on surgical treatment recommendations and nonsurgical prognosis status classification. Indeed, a successful treatment recommendation will assist the doctor in selecting an effective treatment, and the prognosis status classification will help a radiologist recognize the image more efficiently. For these purposes, in this paper, we propose potential solutions through a comprehensive empirical study assessing the effectiveness of recent deep learning techniques on surgical treatment recommendation and nonsurgical prognosis status classification. In the proposed system, the scaphoid is firstly segmented from an unknown X-ray image. Next, for surgical treatment recommendation, the fractures are further filtered and recognized. According to the recognition result, the surgical treatment recommendation is generated. Finally, even without sufficient fracture information, the doctor can still make an effective decision to opt for surgery or not. Moreover, for nonsurgical patients, the current prognosis status of avascular necrosis, non-union and union can be classified. The related experimental results made using a real dataset reveal that the surgical treatment recommendation reached 80% and 86% in accuracy and AUC (Area Under the Curve), respectively, while the nonsurgical prognosis status classification reached 91% and 96%, respectively. Further, the methods using transfer learning and data augmentation can bring out obvious improvements, which, on average, reached 21.9%, 28.9% and 5.6%, 7.8% for surgical treatment recommendations and nonsurgical prognosis image classification, respectively. Based on the experimental results, the recommended methods in this paper are DenseNet169 and ResNet50 for surgical treatment recommendation and nonsurgical prognosis status classification, respectively. We believe that this paper can provide an important reference for future research on surgical treatment recommendation and nonsurgical prognosis classification for scaphoid fractures.
Robust Compensation with Adaptive Fuzzy Hermite Neural Networks in Synchronous Reluctance Motors
  • 期刊名稱:
  • Computer Science and Information Systems
  • 論文摘要:
  • In this paper, a robust compensation scheme using adaptive fuzzy Hermite neural networks (RCAFHNN), for use in synchronous reluctance motors (SRMs), is proposed. SRMs have a simple underlying mathematical model and mechanical structure, but are affected by problems related to parameter ariations, external interference, and nonlinear dynamics. In many fields, precise control of motors is required. Although the use of neural network and fuzzy are widespread, such controllers are affected by unbound nonlinear system model. In this study, RCAFHNN, based on an adaptive neural fuzzy interface system (ANFIS), was used to bound motor system model controller algorithm. RCAFHNN can be characterized in three parts. First, RCAFHNN offers fuzzy expert knowledge, a neural network for online estimation, and recursive weight estimation. Second, the replacement of the Gaussian function by the Hermite polynomial in RCAFHNN enables reduced membership function training times. Third, the system convergence and robustness compensation of RCAFHNN were confirmed using Lyapunov stability.
Measurement of the 5S1/2 to 5D5/2 two-photon clock transition frequency of rubidium-85 in high vacuum
  • 期刊名稱:
  • Optics Letters
  • 論文摘要:
  • We present a scheme to precisely resolve the unperturbed line shape of an optical rubidium clock transition in a high vacuum, by which we avoided the systematic errors of “collision shift” and “modulation shift.” The spectral resolution resolved by this scheme is significantly improved such that we can use “Zeeman broadening” to inspect the stray magnetic field, through which we were able to compensate the magnetic field inside the Rb cells to be below 10−3 Gauss. We thus update the absolute frequency of the clock transition and propose a standard operation procedure (SOP) for the clock self-calibration.
Trapezoid-structured LSTM with Segregated Gates and Bridge Joints for Video Frame Inpainting
  • 期刊名稱:
  • Springer - The Visual Computer Journal
  • 論文摘要:
  • This work considers the video frame inpainting problem, where several former and latter frames are given, and the goal is to predict the middle frames. The state-of-the-art solution has applied bidirectional long short-term memory (LSTM) networks, which has a spatial-temporal mismatch problem. In this paper, we propose a trapezoid-structured LSTM architecture called T-LSTM-sbm for video frame inpainting with three designs: (i) segregated spatial-temporal gates, (ii) bridge joints, and (iii) multi-kernel LSTM. To prevent the spatial-temporal mismatch problem, while features are being passed through multi-layered LSTM nodes, the trapezoid structure reduces its number of LSTM nodes by two after each layer. This makes the model converge to the inpainted results more effectively. The separated temporal and spatial gates design can learn better spatial and temporal features by using individual gates. To relieve the information loss problem during the convergence of the trapezoidal layers, we use bridge joints among layers to better preserve useful information. The multiple kernels in LSTM are to enable extracting multi-scale information flows. T-LSTM-sbm is proved to outperform the state-of-the-art solutions in peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) on three common datasets, KTH Action, HMDB-51, and UCF-101.
QoE Sustainability on 5G and Beyond 5G Network
  • 期刊名稱:
  • IEEE Wireless Communications
  • 論文摘要:
  • The explosion of mobile applications and phenomenal adoption of mobile connectivity by end-users has generated an increasing amount of mobile data traffic. Application posing stringent network requirements of high bandwidth and low latency (e.g., immersive videos) and the substantial amount of data traffic has put tremendous pressure on existing network infrastructures. Cognizant of the need of increasing network capacity, the simultaneous use of heterogeneous network technologies (HetNet) has been proposed to address this imperative problem. 5G is expected to further drive the concept of HetNet by allowing the use of huge available bandwidth at milli-meter wave frequencies. While HetNet concentrates on improving network capacity from the data transmission perspective, it overlooks the importance of enhancing the support of user’s Quality of Experience (QoE) for the evolving new services. There is a wide range of factors influencing user’s QoE, such as network performance including delay, jitter and throughput, contextual influence such as personalized content delivery, mobility aware content caching and dissemination, and human impact such as human roles and demographic attributes. This paper initially presents a QoE-centric analysis, and evaluation of multimedia services over 5G networks. Then, to address the shortcomings of existing mobile networks, we propose a framework to enhance the support of QoE, to enable smooth delivery of personalized immersive video environment and personalized interaction with an immersive video, anywhere, anytime and on any device. Finally, we propose new solutions to achieve practically feasible spectrum allocation and personalized content caching and dissemination, to provide uninterrupted multimedia services to end-users.
High-Performance Content-Based Music Retrieval via Automated Navigation and Semantic Features
  • 期刊名稱:
  • Engineering Applications of Artificial Intelligence
  • 論文摘要:
  • Content-based music retrieval has been studied for many years. However, it is not easy to achieve effective and efficient retrieval because two issues such as search strategy and music feature are not considered simultaneously. Therefore, in this paper, we propose an innovative music search method using automated navigations and semantic features to cope with these issues. For automated navigations, it is a novel autonomous-feedback technique that moves the search towards the user interest space effectively and efficiently. For semantic features, the low-level audio features are transformed into high-level semantic features to effectively associate with user concepts. To reveal the performance of the proposed method, we conducted a set of comprehensive evaluations on two real music datasets. In the comparative experiments, semantic features are shown to be more effective than audio features. Additionally, the proposed method is superior to state-of-the-art methods in terms of precision, which indicates the average improvements of 151.67% and 148.02% on two datasets, respectively. Moreover, the subjective evaluation shows that the proposed method can earn the users’ satisfactions in the materialized system. In summary, the proposed automated navigations and semantic features are useful for dealing with issues of search strategy and music feature in content-based music retrieval.
Binary Signal Perfect Recovery from Partial DFT Coefficients
  • 期刊名稱:
  • IEEE Transactions on Signal Processing
  • 論文摘要:
  • How to perfectly recover a binary signal from its discrete Fourier transform (DFT) coefficients is studied. The theoretic lower bound and a practical recovery strategy are derived and developed. The concept of ambiguity pair is introduced. This pair of signals has almost the same DFT coefficients except for some positions. It can prove that when the signal length is N, then at least τ (N) DFT coefficients must be sampled, where τ (N) is the number of factors of the signal length N. A recovery algorithm is proposed and implemented. It can achieve the lowed bound for length N = 2m,m ≤ 6. To overcome the length limitation problem, a more practical recovery method is also proposed and implemented for N = 2m,m 大於 6. We can sample 11% of the total DFT coefficients to perfectly recover the binary signal. We also extend the concept of ambiguity pair to other discrete transforms (DCT andWHT) and two-dimensional DFT cases.