2,771 research outputs found
Yves-Heng Lim
No full textYves-Heng Lim est enseignant-chercheur au Département d’Etudes de Sécurité et de Criminologie de l’Université Macquarie, Sydney. Il est l’auteur de China’s Naval Power: An Offensive Realist Approach (Ashgate, 2014). Yves-Heng Lim is a lecturer at the Department of Security Studies and Criminology, Macquarie University. He is the author of China’s Naval Power: An Offensive Realist Approach (Ashgate, 2014)
Professional attachment report [with] Chio Lim & Associates.
No full textThis report serves as a summary of the professional attachment. Besides touching on author experiences working with Chio Lim & Associates (CLA), it wil also touch on other issues before, during and after the program
Lim-inf convergence and its compactness
No full textAbstract- We describe the Mizar formalization of the proof of compactness of lim-inf convergence given in [W33] according to [CCL]. Lim-inf convergence formalized in [W28] is a Moore-Smith convergence investigated in [Y6] and involves the concept of nets. The proof is based on the equivalence of two approaches to convergence in topological spaces: filter convergence and Moore-Smith (net) convergence. The equivalence is worked out in [Y19] and different characterizations of compactness are also given there. These efforts are a continuation of the international project of formalizing the theory of continuous lattices headed by the first author
The test-induced warm-up effect on hamstring flexibility tests
No full textBackground: Although the effect of active warm-up (WU) on acute flexibility enhancement is well documented, the test-induced WU effect in muscle length test has not been widely studied. Objective: This study aimed to verify the test-induced WU effect on hamstring flexibility tests. Methods: The active knee extension (AKE) was performed using the right leg, whereas the straight leg raise (SLR) was performed using the left leg. Ten trials of AKE or SLR were performed: two as the pre-intervention trials (Pre); six as the WU intervention; and another two trials as the post-intervention (Post). During WU, subjects in the WO-Hold group performed six trials of the AKE or SLR without hold, and those in the W-Hold group performed six trials of the AKE or SLR with a 5[Formula: see text]s hold. Results: A significant difference was noted between Pre-AKE and Post-AKE, and between Pre-SLR and Post-SLR, respectively, in both the groups. The effect of WU is clear when performing consecutive AKE or SLR without any additional hold. Conclusion: Practitioners should be cautious in interpreting the testing result to avoid overestimation of the treatment effect since the test itself may induce substantial WU effect to the target tissues
SpecAugment for Sound Event Detection in Domestic Environments using Ensemble of Convolutional Recurrent Neural Networks
No full textIn this paper, we present a method to detect sound events in domestic environments using small weakly labeled data, large unlabeled data, and strongly labeled synthetic data as proposed in the Detection and Classification of Acoustic Scenes and Events 2019 Challenge task 4. To solve the problem, we use a convolutional recurrent neural network composed of stacks of convolutional neural networks and bi-directional gated recurrent units. Moreover, we propose various methods such as SpecAugment, event activity detection, multi-median filtering, mean-teacher model, and an ensemble of neural networks to improve performance. By combining the proposed methods, sound event detection performance can be enhanced, compared with the baseline algorithm. Consequently, performance evaluation shows that the proposed method provides detection results of 40.89% for event-based metrics and 66.17% for segment-based metrics. For the evaluation dataset, the performance was 34.4% for event-based metrics and 66.4% for segment-based metrics.12913
멀티모달 심층 신경망을 이용한 시각 정보 기반 공간 오디오 생성 및 향상
No full text학위논문(박사) - 한국과학기술원 : 문화기술대학원, 2025.2,[iv, 81 p. :]Spatial audio is essential for many immersive content serviceshowever, it is challenging to obtain or create it. Therefore, to provide a wide range of spatial audio services, leveraging existing audio content to create new spatial audio content can be an effective solution. To address this requirement, this study proposed a method for generating and enhancing spatial audio based on visual information using multi-modal deep neural networks. Recently, multi-modal based ambisonic audio generation has emerged as a promising approach for spatial audio generation. It combines multiple modalities, such as audio and video, and provides more intuitive control of ambisonic audio generation. Moreover, it leverages the advantages of machine-learning methods to automatically learn the correlation between different features and generate high-quality ambisonic sounds. Herein, we propose a separation- and localization-based spatial audio generation model. First, the network extracts visual features and separates audio into sound sources. Then, it conducts localization by mapping the separated sound sources to the visual features. To overcome the performance limitation of the previous self-supervised source separation approach, we employ a pre-trained source separator with superior performance. To improve the localization performance further, we propose a channel panning loss function between each channel of the ambisonic signal. We use three different types of datasets to train the model experimentally and evaluate the proposed method with four metrics. The results show that the proposed model achieves better spatialization performance than the baseline models. In addition, we present audio dereverberation based on the visually-informed diffusion model (ADVID), a novel method that leverages visual information to enhance the quality and intelligibility of speech in reverberant environments. Traditional audio dereverberation techniques perform poorly in complex acoustic settings because of their reliance solely on audio signals. By incorporating visual cues, ADVID effectively captures the spatial and material properties of an acoustic scene, thereby improving the dereverberation performance. ADVID utilizes pretrained visual encoder architectures to extract detailed visual features from red, green, and blue (RGB) and depth images. These features provide a crucial environmental context, including the geometry and material composition essential for accurate dereverberation. Our approach integrates a diffusion model based on the noise conditional score network architecture, which is enhanced by a multi-resolution U-Net framework and cross-modal attention mechanisms. Thus, the model is enabled to robustly process complex spectrograms and achieve superior dereverberation results. Moreover, to ensure alignment between the visual and audio embeddings, we introduce a contrastive audio-visual matching loss that further enhances the effectiveness of the model. Further, the performance was experimentally evaluated using audio-visual datasets. The results demonstrate that ADVID significantly outperforms various state-of-the-art methods, achieving higher scores on objective metrics. Subjective listening tests also confirmed that they provided superior speech quality and intelligibility compared with the baseline model. In addition, the feasibility of the proposed model for providing spatial audio content was validated by resynthesizing the target spatial environment from the dereverberated audio. Finally, we validated the performance of the proposed model through experiments using a dataset obtained from a real-world environment, confirming its extensibility.한국과학기술원 :문화기술대학원
FARKLI YOĞUNLUKLARDA YAPILAN SUBMAKSİMAL İZOMETRİK KONTRAKSİYONLARIN SIRASINA BAĞLI OLARAK AĞRI ALGISINDA OLUŞAN FARKLILIK
No full textA Review of Motion Capture Systems: Focusing on Clinical Applications and Kinematic Variables
No full textTo solve the pathological problems of the musculoskeletal system based on evidence, a so-phisticated analysis of human motion is required. Traditional optical motion capture systems with high validity and reliability have been utilized in clinical practice for a long time. However, expensive equipment and professional technicians are required to construct optical motion capture systems, hence they are used at a limited capacity in clinical settings despite their advantages. The development of information technology has overcome the existing limit and paved the way for constructing a motion capture system that can be operated at a low cost. Recently, with the development of computer vision-based technology and optical marker-less tracking technology, webcam-based 3D human motion analysis has become possible, in which the intuitive interface increases the user-friendliness to non-specialists. In addition, unlike conventional optical motion capture, with this approach, it is possible to analyze mo-tions of multiple people at simultaneously. In a non-optical motion capture system, an inertial measurement unit is typically used, which is not significantly different from a conventional optical motion capture system in terms of its validity and reliability. With the development of markerless technology and advent of non-optical motion capture systems, it is a great advan-tage that human motion analysis is no longer limited to laboratories
Ultra-Low Power Circuit Design for Miniaturized IoT Platform
No full textThis thesis examines the ultra-low power circuit techniques for mm-scale Internet of Things (IoT) platforms. The IoT devices are known for their small form factors and limited battery capacity and lifespan. So, ultra-low power consumption of always-on blocks is required for the IoT devices that adopt aggressive duty-cycling for high power efficiency and long lifespan. Several
problems need to be addressed regarding IoT device designs, such as ultra-low power circuit design techniques for sleep mode and energy-efficient and fast data rate transmission for active mode communication. Therefore, this thesis highlights the ultra-low power always-on systems, focusing on energy efficient optical transmission in order to miniaturize the IoT systems.
First, this thesis presents a battery-less sub-nW micro-controller for an always-operating system implemented with a newly proposed logic family.
Second, it proposes an always-operating sub-nW light-to-digital converter to measure instant light intensity and cumulative light exposure, which employs the characteristics of this proposed logic family. Third, it presents an ultra-low
standby power optical wake-up receiver with ambient light canceling using dual-mode operation.
Finally, an energy-efficient low power optical transmitter for an implantable IoT device is suggested. Implications for future research are also provided.PhDElectrical EngineeringUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttps://deepblue.lib.umich.edu/bitstream/2027.42/145862/1/imhotep_1.pd
- …
