34254 research outputs found
Sort by
A Hierarchical Classification Model for Intrusion Detection System
No full textIntrusion detection systems (IDS) are used to detect abnormal behaviors in networks. Several studies use machine learning (deep learning) methods to develop detection models. This kind of method attempts to enhance the performance of these models by adjusting the model or parameter. However, there are some different types of abnormal behaviors with similar properties, making it difficult to classify them via the single-model. The proposed method learns multiple classifiers for different data types; allows the model to clearly distinguish the difference between the different data types; then stacks multiple classifiers to achieve the hierarchical detection. Data augmentation is an important technique in machine learning that diversifies data to enhance model generalization. But the augmented data cannot be identified whether its meaning was changed; nevertheless, some augmented data with the correct infor- mation can further improve the result and make the model more generalized. The proposed method utilizes the domain adaptation method to inspect data and delete unsuitable augmented data. Experimental results show that our proposed method can outperform other comparative methods in terms of accuracy and recall, including machine learning, ensemble learning, and deep learning methods. It is shown that the proposed method can provide a promising design to detect abnormal behaviors
Compressed sensing architecture for high-speed sampling biomedical signals based on Fast-Fourier-Transform and Gaussian Random matrix
No full textCompressed Sensing (CS) algorithm, also known as sparse sampling, is a signal processing technique for efficiently obtaining and reconstructing signals by using the solution of an uncertain linear system. Based on the above principles, compressed sensing can restore the original signal using far fewer samples than Nyquist-frequency. Usually, in the field of signal processing engineering, it is impossible to restore without measurement. However, CS only needs to use assumptions to perfectly restore the original signal under the condition that the sampled signal meets the conditions of sparsity and non-correlation.
In the CS architecture, it is divided into two major measurement items and a reconstruction algorithm, which are the frequency domain correction circuit and the measurement circuit. The function of the frequency domain correction circuit is to obtain a sparse matrix (Sparse Matrix ), to conform to the sparsity of compressed sensing, discrete cosine transform (Discrete Cosine Transform, DCT), discrete wavelet transform (Discrete Wavelet Transform, DWT), discrete Fourier transform (Discrete Fourier Transform, DFT) or DFT Derived Fast Fourier Transform (Fast Discrete Fourier Transform, FFT).
The function of the measurement circuit is to implement compressed sensing sampling. By controlling the measurement process, the sampling amount can be reduced while ensuring that the effective information in the target signal will not be lost. The original signal can be restored through reconstruction. Gaussian is often used. Random matrix (Random Matrix) and Hadamard matrix (Hadamard matrix).
The function of the reconstruction algorithm is to solve and find the best solution from the previous sampling values. The accuracy and stability of the reconstruction algorithm are the key points in the design. L1 norm normalization (L1 regularization) and regularization are often used. Orthogonal Matching Pursuit (OMP)
PTT Upvotes, Downvotes, and Sentiment Analysis: Examining the Impact of Engagement in Discussion Threads on Movie Reviews on Cinema Attendance in Taipei.
No full textThis study delves into the potential causal relationship between interactive comment threads beneath positive and negative movie reviews on Taiwan's PTT movie board and cinema admissions. The study encompasses 4,247 films shown in Taipei theaters from 2017 to 2022, totaling 654,894 observations. The primary variables in focus are comment labels (upvotes and downvotes) and sentiment analysis scores.
Using a Fixed Effects Model for panel data analysis, the research reaffirms that positive reviews correlate with increased box office revenues. In contrast, neutral and negative comments might be detrimental, with the impact of negative feedback being notably stronger than that of positive. For interactive threads, upvotes and downvotes on positive articles have corresponding positive and negative effects. However, a majority of downvotes or negative sentiments under a positive review could paradoxically boost ticket sales, possibly because downvotes intensify the discussion's engagement. Upvotes and positive sentiments on neutral reviews can mitigate their inherent negative implications, even possibly rendering a positive influence on ticket sales.
Regarding negative reviews, the effects of upvotes and downvotes are not consistently linear. An overwhelming number of upvotes or downvotes on a single article might have negative implications. However, mild disagreement might have positive outcomes. Sentiment analysis further indicates that positive sentiments in an individual negative review might harm ticket sales, but if most negative reviews exhibit positive sentiments, it could be beneficial.
In conclusion, the optimal strategy for word-of-mouth movie promotion is to amplify the upvote interactions under positive reviews. For neutral or negative reviews, increasing upvotes and positive sentiments can be an effective tactic, while a balanced mix of downvotes and positive sentiments under negative reviews can alleviate adverse effects
Prediction of Alzheimer's Disease via Spatial Temporal Graph Convolution
No full textAlzheimer's disease, a progressive neurodegenerative condition, is one of the most prevalent forms of dementia in the elderly. In recent years, early prediction of Alzheimer's disease to prevent disease progression has become a popular research topic. This paper focuses on converting whole-brain images from resting-state functional Magnetic Resonance Imaging (fMRI) into adjacency matrices and time series, which respectively represent the functional connectivity between predefined brain regions and the neural activity signals of each brain region at different time points. Subsequently, graph theory analysis and a Spatio-Temporal Graph Convolutional Network (ST-GCN) are employed to predict and classify individuals as healthy or Alzheimer's patients.
To optimize computational resources, dimensionality reduction is commonly applied to separate the adjacency matrices and time series. However, most existing methods in the field rely on either the adjacency matrix or the time series for prediction and analysis, leading to a compromise in temporal or spatial characteristics. In contrast, this paper introduces a spatiotemporal graph convolutional network that considers both the spatial characteristics of adjacency matrices and the temporal characteristics of time series.
The proposed approach involves preprocessing the functional MRI data by applying a brain template and extracting features from each brain region. These region-specific features are then merged and fed into the ST-GCN for prediction. This method not only allows the model to learn edge weights with physical significance but also achieves a 94% accuracy on the ADNI dataset
Experimental investigation of plunging dam break wave impact on square prism
No full textWhen designing coastal or offshore structures, assessing the impact of waves is crucial. The study simulates the impact of breaking waves on the ocean using plunging dam break wave. We discusse the flow field, impact pressure, and void fraction. An electric linear stage was used to activate the gates, and Particle Image Velocimetry (PIV)and Bubble Image Velocimetry (BIV) were employed to measure the flow field. Four structures are sequentially positioned downstream, starting from the location where the mushroom-like jet impacts the water surface as the first measurement cross-section. The subsequent cross-sections are positioned at intervals of 40 mm, 40 mm, and 50 mm. At each measurement cross-section, starting from the downstream still water level, ten pressure measurement points are vertically spaced at intervals of 12 mm to measure the impact pressure when waves collide with the structures at different elevations. Additionally, a Fiber Optic Reflectometer (FOR) is used to measure the void fraction of the waves.
From the experimental data, it was found that the positions of maximum impact pressure in each section were greater than the still water level. Furthermore, as the sections moved backward, the position of maximum impact pressure gradually increased. Additionally, this study discovered that higher void fractions during wave impact result in greater impact pressures. The concept of Peregrine and Thais' (J Fluid Mech 325:377\ue2397, 1996) filling flow was introduced to analyze the relationship between air and impact pressure. It is found that as the void fraction increases, the proportion of pressure caused by compressed air in the impact pressure also increases. The study also conducted an analysis of the impact coefficient. After considering the correction of fluid density based on void fraction, the average impact coefficient increased from 1.54 to 2.16. Finally, by combining flow field measurements and FOR in empty water tanks measurement, it was found that void fraction measurement is indispensable in capturing the aeration flow field
Fabrication of Back-etched Thin Film Acoustic Wave Filters for 5G n41 Frequency Band
No full textThis research paper investigates the fabrication of a 2.6 GHz back-etched thin-film bulk acoustic wave (BAW) filter using aluminum scandium nitride (AlScN) piezoelectric thin films, aiming to meet the specifications of the 5G n41 frequency band. The study explores the modulation of the filter's bandwidth by varying the thickness of the parallel arm of the top electrodes, and evaluates its relevant filtering performance.
In this study, silicon (Si) is used as the substrate, platinum/titanium (Pt/Ti) is used as the electrode thin film, and aluminum scandium nitride (AlScN) is used as the piezoelectric layer to fabricate the back-etched BAW filter device. Initially, the bottom electrode is deposited using a DC magnetron sputtering system. Then, a reactive RF magnetron sputtering system is employed with an Al0.9Sc0.1 alloy target. By adjusting the nitrogen flow rate, sputtering pressure, sputtering power, and growth temperature, the deposition of the piezoelectric thin film is carried out. Through thin film characterization analysis, optimal sputtering parameters are obtained, resulting in an Al0.92Sc0.08N piezoelectric thin film with a high c-axis (002) orientation. Subsequently, the backside cavity is etched using a two-stage wet etching process with a 30 wt% KOH solution, completing the fabrication of the FBAR filter device. Finally, the passband width of the filter is achieved by modulating the thickness of the parallel arm of the filter with different top electrodes (150 nm, 170 nm, 190 nm).
Ultimately, the designed FBAR filter in this study exhibits optimal performance with a parallel arm thickness of 170 nm. It has a center frequency of 2.60 GHz, a 3 dB bandwidth of 52.89 MHz, an insertion loss of 12.50 dB, a maximum rejection bandwidth of 107.55 MHz, a sidelobe suppression of 26.72 dB, and a shape factor of 2.03, meeting the required performance and bandwidth specifications of the 5G n41 frequency band
Impact of AI-Induced Job Changes on Employee Creativity and Job Insecurity: Mitigating Stress and Enhancing Creativity through Job Crafting Strategies
No full textThis study investigates the different impacts of job changes caused by artificial intelligence (AI) on employee outcomes, with particular attention being placed on job insecurity, creativity, engagement, and job crafting. This study incorporates the Job Demands-Resources (JDR) theory and examines the significance of employee resources and job crafting in alleviating the adverse consequences of changes induced by artificial intelligence (AI). It demonstrates the importance of ICT resources and digital dexterity for assisting employees cope with AI-induced job changes and emphasizes the importance of approach and avoidance strategies in the form of job crafting. The availability of ICT resources and the implementation of measures aimed at mitigating job insecurity are essential factors in enabling employees to effectively navigate AI-driven environments and actively engage in proactive job crafting. This study makes a significant contribution to the current body of literature by offering valuable insights into the impact of artificial intelligence (AI) on employee outcomes. It also guides organizations in effectively implementing AI technologies and fostering creativity in the ever-changing workplace
The effects of teacher self-disclosure on student mindset and mathematics self-efficacy
No full textMathematics is considered the cornerstone of foundational subjects, and students' engagement with math goes beyond solving mathematical problems. It aims to enhance their logical thinking abilities, enabling them to tackle issues in various academic disciplines and real-life scenarios. Schools universally strive to cultivate students' interest and motivation in learning mathematics. Among various influencing factors, mathematical self-efficacy stands out as a critical determinant of mathematical learning. While past research emphasizes the motivating role of role models on students' self-efficacy and their growth/fixed mindsets, STEM studies predominantly focus on renowned scientists as role models, often overlooking the influential role teachers play in students' learning journeys. This study seeks to investigate how teachers, as role model figures, influence students' growth/fixed mindsets and subsequently impact their mathematical self-efficacy through the self-disclosure of their struggles and achievements. Conducted in a private secondary school in the southern region, this research involves 86 first-year junior high school students. Participants are randomly assigned to either the experimental or control group, and a quasi-experimental design is employed to validate the research hypotheses. The findings of this study reveal that teacher self-disclosure significantly affects students' growth/fixed mindsets, while its impact on mathematical self-efficacy is less pronounced. It is postulated that the development of mathematical self-efficacy requires prolonged cultivation, making it less susceptible to short-term experimental manipulation. In conclusion, this study sheds light on the pivotal role of teachers as role models and their influence on students' mindset development, ultimately impacting mathematical self-efficacy. Despite the transient effects observed in this short-term experiment, it underscores the need for long-term efforts to foster students' mathematical self-efficacy
Analysis of Heat Transfer Characteristics of Two-Fluid Nozzle With HFE-7300 Spay on Different Impingement Surfaces
No full textWith the rapid advancement of modern technology, there is a growing demand for efficient heat management and cooling systems across various industries. In many heat exchange applications, two-phase spray cooling has emerged as a widely adopted technique. Notably, HFE-7300, known for its non-flammability, non-oiliness, low toxicity, non-corrosiveness, good material compatibility, and high-temperature stability, coupled with the superior heat dissipation benefits of graphene-coated surfaces, presents an attractive option.
Therefore, this study primarily focuses on a 1.6 mm exit diameter two-fluid nozzle, utilizing the dielectric fluid HFE-7300 and comparing it with DI water for spray cooling. The nozzle is fixed at a distance of 50 mm from the test surface, while the air-liquid ratio is adjusted by varying the air velocity with constant liquid flow rates (DI water = 100 ml/min, HFE-7300 = 60 ml/min). Four different air-liquid ratios were investigated (ALR=0.13, 0.19, 0.21, 0.25), with two types of test surfaces: a smooth copper block surface and a graphene-coated surface (thickness <50 nm) on a copper block. The base was heated to 200\uc2\ub0C using a heating rod, enabling spray cooling while observing changes in temperature distribution and droplet velocity fields.
Temperature measurements were carried out using thermocouples, and experimental parameters such as test surface temperature and heat transfer rate were calculated using extrapolation to construct quenching cooling curves and transient boiling curves. Through calculations and analysis, the optimal Critical Heat Flux (CHF) for the dielectric fluid spray cooling was determined. In terms of flow field analysis, micro Particle Image Velocimetry (\uce\ubcPIV) was employed to capture and analyze droplet velocities, while Interferometric Particle Imaging (IPI) was used to examine variations in droplet sizes.
The experimental results indicate that, with a fixed liquid flow rate, higher air-liquid ratios lead to higher CHF, with the optimal CHF observed at an air-liquid ratio of 0.25. For HFE-7300, the optimal CHF on the smooth surface and graphene-coated surface were 281.17 W/cm2 and 362.04 W/cm2, respectively. Additionally, droplet velocity increases with an increasing air-liquid ratio, following a trend similar to CHF. Conversely, the slip ratio decreases as the air-liquid ratio increases, and the experiments reveal that lower slip ratios are associated with higher cooling rates. Interestingly, at an air-liquid ratio of 0.25, the lowest slip ratio and highest cooling rate were observed at a distance of 40 mm from the nozzle exit, which contrasts with the trend observed for void fraction
A Study on Heat Transfer Enhancement of Methanol Solution in Microchannels with Roughened Surfaces at Low Reynolds Number
No full textNowadays, the usage of electronic products is prevalent, and the heat dissipation mechanism of these products has gradually garnered attention. Direct liquid cooling has emerged as a promising alternative in recent years. Microchannel surfaces in direct contact with the object can significantly reduce the interface thermal resistance during heat dissipation. Compared to traditional methods, only a small amount of fluid is required to achieve efficient cooling. Methanol is an easily transportable liquid and has a high volumetric energy density, which may serve as an alternative to hydrogen as energy carrier. This study delves the impact of different volume concentrations (5%, 10%, 20%, 30%) of methanol solutions on heat transfer enhancement in rough surface microchannels at low Reynolds numbers (10\ue2\ua4Re\ue2\ua430).
To conduct the experiment, rectangular microchannels with three different heights (25\uce\ubcm, 30\uce\ubcm, 35\uce\ubcm) of ribs and three pitches (120\uce\ubcm, 132\uce\ubcm, 144\uce\ubcm) of ribs were fabricated on a silicon wafer using the microelectromechanical systems (MEMS) processes. These microchannels were then replicated using polydimethylsiloxane (PDMS) and bonded to glass to create the experimental setup.
Microparticle image velocimetry (\uce\ubcPIV) was utilized to analyze the velocity distribution within the microchannels. The temperature field was measured using thermocouples, connected to Labview software to calculate the local heat transfer coefficient (hx) and Nusselt number (Nux). The study further examined the impact of height and pitch of ribs in microchannel on heat transfer enhancement.
The results show that the convective heat transfer coefficient and Nusselt number increase with increasing ribs height and pitch in the channel. The highest local convective heat transfer coefficient (hx) of 1097.26 W/m\uc2\ub2K and Nusselt number (Nux) of 0.202 are achieved with a channel having a ribs height of 35\uce\ubcm and pitch of 144\uce\ubcm. The experimental results indicate that, the heat transfer enhancement of microchannels is better when the ribs height is higher, the pitch is larger, and the methanol solution concentration is higher at higher Reynolds number. The channel with ribs height of 35\uce\ubcm has a 24% increase in heat transfer efficiency compared to a smooth channel. The heat transfer efficiency of the channel with a ribs pitch of 144\uce\ubcm is 12% higher than that of the channel with a ribs pitch of 120\uce\ubcm. The heat transfer efficiency of the microchannel at Reynolds number of 30 is 27% higher than that at a Reynolds number of 10. The heat transfer efficiency of a 30% volume concentration methanol solution is 8% higher than that of a 5% volume concentration methanol solution