34254 research outputs found
Sort by
A Brief Review of Deep Learning Optimizers
No full textIn recent years, deep learning technology has attracted the attention of all walks of life. Not only has the related research grown rapidly, but it has also been widely used in business. Optimization plays a key role in deep learning. As deep learning becomes the mainstream of machine learning, the amount of data and the size of neural networks increase sharply, researchers thus propose all kinds of algorithms to deal with these large models. This work briefly tidies up the optimization methods that have been applied in deep learning fields, from the origin of deep learning to the state of the art nowadays. In this thesis, we define machine learning procedure as an optimization problem first, then describe the relationships with different fields of machine learning. After that, we classify the classic optimizers from the perspective of whether they use gradients as main information or not and introduce some famous research. Furthermore, we extend to the mainstream research direction of optimization in modern neural networks and the main problems that need to be resolved and introduce the works that have made significant achievements. At the end of the thesis, we also put forward some issues and challenges that researchers may face in the future in the field of optimization of deep neural networks
Pulse-to-Pulse Coherent Doppler Sonar Signal Processing and Implementation
No full textThe Acoustic Doppler Current Profiler (ADCP) is a crucial technology used in ocean acoustics for measuring water currents. The research theme of this thesis aims to implement the velocity profiling function of the Acoustic Doppler Current Profiler using hardware design based on Pulse-to-Pulse Coherent theory.
This thesis mainly consists of two parts: Digital signal processing and frequency analysis. Digital signal processing plays a crucial role in ocean velocity measurement, involving tasks such as in-phase and quadrature demodulation, and sampling rate modulation. Due to the small values of ocean velocities, the Doppler frequency shift of echo signals differs significantly from the sonar carrier frequency. While sampling rate modulation improves frequency resolution accuracy, simple modulation can lead to aliasing and errors. Therefore, In-Phase and Quadrature modulation techniques are used to perform frequency modulation on the echoes, allowing for a more accurate capture of velocity variations.
Regarding frequency calculations, the Pulse-to-Pulse Coherent theory detects the main frequency of echoes by computing the spectral centroid of energy spectrum. By integrating the Wiener\ue2Khinchin theorem, eliminating the need for extensive computational resources, while ensuring the accuracy of the results.
Furthermore, this thesis discusses the hardware design and implementation, utilizing the Cyclone IV E FPGA and a multi-clock pipeline design. with operating clock frequencies of 1 time and 10 times the sampling rate, the system offers high throughput and adaptability, effectively realizing the ADCP
Fabrication of Gold Nanoparticle Polydimethylsiloxane Film
No full textThe production of composite materials containing nanoparticles involves two main methods, and for this experiment, the chosen approach is to disperse metal salts in a polymer matrix through a reduction process. Preparing nano gold particles is accomplished in a liquid environment by first synthesizing nano gold particles and then incorporating them into a liquid polymer. We utilized a chemical reduction method to synthesize nano gold particles (AuNP), using a strong reducing agent to reduce gold ions from tetrachloroauric acid into nano gold particles as a seed solution. The seed solution is further developed into nano gold rods (AuNRS) through a seed-growth method. These synthesized nano gold rods are then integrated into a reusable multi-composite material.The optical properties of AuNRS in distilled water and curing agent suspensions were analyzed using UV-visible spectrophotometry (UV-VIS). A peak is observed around 500 and 700 nm, corresponding to the long and short axes of AuNRS, respectively. Subsequently, the synthesized nano gold rods are mixed into polydimethylsiloxane (PDMS) to create a thin film. Fourier-transform infrared spectroscopy (FTIR) is employed for analysis, monitoring the composition and morphology of the resulting composite material. Peaks associated with gold are identified at 1945 cm-1, 2156 cm-1, 3280 cm-1, and 3372 cm-1 through peak analysis of the FTIR spectrum. The proposed method not only simplifies the preparation process but also provides insights into the material characteristics of nano gold and its synthesis methods
Research on the User Experiences of Mobile Payment Platforms in Taiwan
No full text\ue3\ue3In recent years, the rapid development of mobile payment in Taiwan can be attributed to several key factors. Firstly, technological advancements have brought innovative solutions and a more convenient user experience to mobile payments. The widespread availability of smartphones and high-speed internet has made it easier for more consumers to access and use mobile payment services.
\ue3\ue3Secondly, the outbreak of COVID-19 has also accelerated the adoption of mobile payments to some extent. The pandemic-induced social distancing and the increased demand for contactless transactions have led many consumers to seek safe and convenient payment methods, making mobile payments an ideal choice.
\ue3\ue3The Taiwanese government has been actively promoting mobile payments, aiming to achieve a 90% adoption rate by 2025. This reflects the global trend's impact on Taiwan and shows Taiwan's efforts in advancing a cashless society. The government has created a favorable environment for the development of the mobile payment market through the establishment of friendly policies and regulations. Government support is not limited to policy but also includes financial subsidies and technical assistance, which have significantly promoted the innovation and application of mobile payment technologies.
\ue3\ue3This study focuses on user reviews of Line Pay, Taiwan Pay, PXPay Plus, and Jkopay on Google Play, employing six major analytical categories, frequency analysis, and thematic analysis as tools to delve into the real feelings, usage habits, and preferences of users towards these apps. These direct user comments not only reflect the intuitive experience of users with mobile payment applications but also reveal specific needs and expectations when using these apps.
\ue3\ue3The research results show that Line Pay exhibits maturity in app development, with user feedback mainly focused on user experience, especially regarding adaptation issues between old and new versions. For Taiwan Pay, user comments are primarily centered on technical issues, indicating shortcomings in app optimization. PXPay Plus faces challenges in payment system integration and has relatively fewer usage scenarios, highlighting its limitations in market breadth. Lastly, the user experience of Jkopay is affected by its inadequate customer service functions and the requirement for real-name authentication. Although Jkopay performs better in technical stability, user feedback reveals shortcomings in customer service and the negative impact of real-name authentication policies on user experience
Practical Analysis of the Impact of TWQR Development on the Penetration Rate of Mobile Payments
No full textTo achieve the goal of a 90% cashless payment ratio, the Taiwanese government has actively promoted the development of mobile payments through inter-departmental cooperation. Although the rate of cashless payments has been steadily increasing, it has yet to meet the expected targets. In this context, the Financial Information Service Co. introduced the TWQR, which is a standard QR Code system, aimed at enhancing the application scenarios for electronic payments and the integration of QR Code payment methods. This initiative allows consumers to conduct transactions at any TWQR-affiliated merchant using any electronic payment platform. Taiwan Pay, another main policy from the government to popularize mobile payments, primarily operates through financial institutions. However, the promotion teams have encountered various obstacles in the process.
This study, through unstructured interviews with personnel from financial institutions, aims to gain an understanding of the challenges faced during the promotion of mobile payments and to explore the potential impact of TWQR on future promotional activities. We employ the PESTEL model for external environment analysis and utilize the SWOT analysis to thoroughly examine the strengths, weaknesses, opportunities, and threats of TWQR. The research finds that the introduction of TWQR not only boosts confidence among consumers and merchants in electronic payments, yielding positive effects during promotion, but also faces significant challenges. These challenges include operational barriers in the system, issues with transaction fees, concerns over taxation, and delays in fund settlement. These issues need to be addressed through complementary policies and diversified promotional strategies
Automatic Detection Method for Rapid Eye Movement Sleep Developed for Ear-worn Wearable Devices
No full textSleep stage analysis constitutes a pivotal technology within the field of sleep medicine research, where understanding the variations in different sleep stages becomes critically important. This is especially relevant as these data often exhibit anomalies in the context of sleep disorders and psychosomatic illnesses. Amongst these, the Rapid Eye Movement (REM) phase is of paramount significance. Conventionally, the assessment of sleep has been reliant on polysomnography (PSG) conducted in hospital settings. This method is not only time-consuming but also typically inconvenient and uncomfortable for individuals undergoing the procedure.
This study, in collaboration with C Hospital and M Company, aims to collect RR Interval (RRI) and six-axis accelerometer data using ear-worn wearable devices. However, aligning the signals collected by the wearable devices with hospital PSG data presents a timing challenge. To address this, we utilized both hospital ECG signals and wearable device data, applying window correlation methods to correct timing discrepancies. Additionally, due to potential RRI signal loss during data collection by the wearable devices, we used a Multilayer Perceptron (MLP) for data imputation to complete the missing data. Subsequently, we employed Dilation Convolution combined with Long Short-Term Memory (LSTM) models for preliminary sleep stage determination. A majority vote was then used to combine each second's decision results within a 30-second sleep epoch into a single sleep stage outcome. Lastly, through residual network (Resnet) analysis, a comprehensive sleep analysis was conducted. The model demonstrated a Positive Predictive Value of 0.8534 and a Matthews Correlation Coefficient of 0.7407 on training data, while on test data, it showed a PPV of 0.6970 and an MCC of 0.4925. These results indicate the potential of our method in sleep stage discrimination.
This method not only simplifies the traditional manual process of sleep stage identification but also makes sleep analysis more convenient. Utilizing wearable devices, people can comfortably and conveniently conduct sleep monitoring at home, greatly enhancing the acceptability and practicality of the test and enabling real-time sleep state monitoring
The Decision-Making Process of Corporate Investment in Banking Product
No full textThe banking industry used to focus its wealth management development on individual clients. However, faced with product competition among domestic and international banks and challenges in financial innovation services, the banking industry should now emphasize the development of Taiwan's small and medium-sized enterprise (SME) clientele. This study interviewed with decision-makers from 7 SMEs to explore the impact of personal investment experience on enterprise investment decisions, factors considered in choosing banking products for investment, and key considerations in pre- and post-investment operations. The findings reveal that decision-makers' personal investment experiences directly influence enterprise investment decisions. When investing, they also consider the relationship, expertise, and product conditions of the banks involved. The results of this study can aid in the internal training of sales, helping them develop and maintain a focus on investment banking products for corporate clients, aiming for a win-win situation between the banking and corporate sectors
An Analysis of the Impact of ESG Ratings on Fund Performance during the COVID-19 Pandemic
No full textThe purpose of this study is to examine the relationship between ESG ratings and fund performance, specifically during the COVID-19 pandemic. In recent years, ESG investing has shifted from being just a trend to a core factor in investors' decision-making process. One of the appeals of ESG investing is its ability to seemingly allow funds to outperform more conventional peers during periods of market turbulence. The COVID-19 pandemic had significant effects on financial markets worldwide, and provides a valuable opportunity to compare performance of ESG and conventional funds during market crises. The results of this study will hopefully assist investors in deciding whether or not to factor in ESG considerations in their investment decisions
Asset Allocation Analysis for Pension Shortfalls under the New Labor Pension Framework
No full textIn recent years, with the changing demographic structure and evolution of the labor retirement system, significant changes are occurring in retirement planning. According to statistics from the National Development Council (2023), Taiwan is expected to enter the super-aged society in 2025. Along with the issue of retirement income gap caused by inflation and longevity risk, how to fill this gap is the problem the researcher would like to study. However, due to the unpredictable nature of the financial market and the occasional occurrence of "black swan" events causing stock market crashes, people are hesitant to invest. How to use asset allocation to create replicable and executable methods is a crucial question the researcher is eager to explore and answer. In summary, this study employs non-intrusive research, secondary data research, literature analysis, and historical research methods. Through case studies, the aim is to analyze the asset allocation that workers with different investment attributes should make. By reducing risks, the goal is to achieve returns that surpass inflation, thereby bridging the retirement income gap. Only through such measures can we ensure a secure elderly life
Performance of ESG Funds and Its Effect on Bank Customers' Purchase Intentions
No full textThe global trend of sustainable investment has been rising, not only on the moral level, but also on the global industrial chain. However, the craze for sustainable investment funds reached its peak in 2021. After 2022, there was a downward trend in popularity of these funds. In order to explore why the boom in sustainable investment funds has subsided, this article compared the performance of ESG funds and non-ESG funds sold in Taiwan, and then used a survey to explore whether ESG funds can meet investors\ue2 expectations. This study first took 26 ESG funds listed by Taiwan Depository & Clearing Corporation as a sample, and then took 26 non-ESG funds in the TEJ database with the same investment targets and regions as the ESG funds selected. These two samples of funds were matched and compared. Results showed that the overall risk and performance of ESG funds were significantly better than non-ESG funds, but there were different results among investment targets. The results also suggested that the people\ue2s willingness to invest in ESG funds mainly comes from investment performance. These results are congruent with those of PwC (2023), which found that nearly half of investors were unwilling to sacrifice return on investment for ESG concerns