7903 research outputs found
Sort by
Image Generation Task Based on Attention Mechanism and Diffusion Model
No full textIn recent years, artificial intelligence technology has made remarkable achievements, especially in the field of AIGC, such as the text generation model GPT DeepSeek, Image generation models such as stable diffusion and Midjourney benefit from large models pretrained on a large amount of data, with a large number of parameters and phenomena such as knowledge emergence, enabling their effects to reach commercial application levels and be similar to those created by humans. But their achievements may benefit from a large amount of data and parameters, as well as powerful computing power. However, their network model structure may not be optimal, such as the stunning stable diffusion model, which relies solely on traditional CNNs, ordinary up and down sampling modules, residual connections, and simple feature fusion. Due to their high training costs, it is almost impossible to improve the structure and retrain it. At present, there is a lack of research on the impact of network structure on the performance of diffusion models. Therefore, this study aims to conduct research on a small amount of data to explore the influence of different network structures (especially the introduction of attention mechanisms) on the performance of diffusion models. This study selects a simple image generation task based on diffusion models, introduces channel attention mechanisms and temporal attention mechanisms to process feature maps, and uses attention mechanisms to fuse different feature maps, and explores their impact on the quality of image generation. Research has shown that the introduction of attention mechanism significantly improves the image generation quality of diffusion models, while the Natural Image Quality Evaluator (NIQE) and Blind/Referential Image Spatial Quality Evaluator (BRISQUE) are significantly reduced. Especially, the operation of using Vit row feature fusion has the best generation performance. Secondly, Vit can significantly improve the spatial and structural rationality of generated images, which may be attributed to its excellent spatial feature extraction ability; Attention mechanisms such as CA and human CBAM can significantly improve the naturalness of generated images, but their ability to represent details of the generated image targets is poor
Codelympics: An Educational Game
No full textThis thesis addresses the growth of Computer Science and the increasing demand for resources that educate people about programming. The central problem is that existing resources either fall short in engaging the user, or are ineffective at communicating the basics of programming. Through a year of development and testing, we developed an application to serve as a bridge between games and educational content, making programming skills fun and accessible for younger students with no prior experience.
This thesis details the development of Codelympics, an educational game designed to teach the basics of computer programming, including control flow, functions, variables, and problem-solving skills. Codelympics prepares aspiring programmers for real-world coding environments, through open-ended problems that challenge users and promote creative thinking. Codelympics is developed with the theme of sport-based minigames, designed to mirror programming concepts and make them engaging for younger students.
Results from both alpha and beta testing show that video games can be an effective medium for teaching students programming. Results revealed an increase in knowledge about control flow and logic after completing both minigames. Additional feedback revolved around patching potential bugs in the game. Future work includes the development of additional minigames, improved visual and auditory assets, and introducing new topics such as arrays, objects, classes, and other object-oriented programming concepts
EMT Vision
No full textAugmented Reality (AR) has demonstrated considerable promise for future mobile technologies, offering the ability to overlay crucial information within a user’s vision while they can still maintain awareness of the surrounding environment. Similarly, Artificial Intelligence (AI) is an increasingly influential technology with significant potential to revolutionize the medical field. Its ability to rapidly learn and adapt to specific tasks makes it particularly promising for supporting paramedics during emergency calls. AI can efficiently analyze real-time data and present it in a concise, actionable format, enhancing decision making in critical situations.
Given the potential of these technologies, we have developed a smart AR headset designed to leverage AI for the benefit of first responders. This innovative device transcribes and analyzes real-time dialogue between patients and paramedics, extracting and displaying key information. This allows paramedics to review essential details during patient care and when completing necessary documentation, ultimately improving the quality and efficiency of emergency medical services. Furthermore, our headset is capable of providing 3D holographic visuals that display procedures and checklists for patient treatment, as well as using cloud computing algorithms that interpret conversations to automate the collection of patient data for scenario reports
BOOST-EVX: Balancing Battery State-of-Charge for Extended EV Range
No full textClimate change continues to threaten human existence, continuing pushes for sustainable alternatives for current systems. Electric Vehicle (EV) technologies are rapidly changing as their popularity increases. This project addresses the need for additional driving range for electric vehicles by proposing a highly-adaptable and effective battery management system. The battery management system would allow the use of any DC-DC converter and is compatible with varying electric vehicle battery architectures. The specific deliverable and results of this project is a Simulink test bench environment intended to be used by students, researchers, and EV engineers for further EV battery technology development. Future work can explore further increasing the accuracy and variability of our modeling system in progress to create a physical implementation
Veggielution Solar
No full textVeggielution is a nonprofit community farm located in East San Jose that supports food justice and education programs. However, its buildings face limitations due to a lack of reliable and sustainable power. With no funding allocated for infrastructure upgrades, Veggielution sought a cost-effective energy solution.
The goal of this project was to design a solar-powered microgrid that could supply electricity for lighting, ventilation, and hot water to Veggielution. The system needed to be low-cost, easy to install, and maintain by community members.
Our team developed a complete deployment package that includes a cost estimate, DIY installation instructions, component sourcing recommendations, and a system maintenance plan featuring load-shedding strategies. The system design incorporates eight 300W solar panels mounted at a 30.3° southwest angle on a newly proposed structure.
Although the full system could not be implemented on-site due to budget constraints, a scaled lab prototype was constructed using funding from the School of Engineering to validate the performance of our design. The final deliverable, a comprehensive deployment manual, equips Veggielution with the guidance needed to install the system when future funding becomes available, ensuring long-term sustainability and energy independence
Self Driving Robot Car
No full textIn the real world, vehicular crashes are the result of human error and inadequate reaction time and are often deadly. Due to the issue of accidents being frequent occurrences, using machine learning to drive vehicles has quickly become a relevant topic as it could potentially become a way to minimize fatalities and damages. Thus, the project aims to approach self-automation with the focus on maximizing the vehicle’s performance and the machine’s ability to make the best decisions within that time.
Today’s self-driving cars use radar or cameras as well as digital signal processing algorithms to sense the environment while using machine learning algorithms to process that data to make informed decisions. Currently, the bulk of these algorithms run in software. This implementation is simpler than in hardware and is often the chosen method due to the already complex nature of the software stack. However, it incurs more overhead than hardware implementations in terms of performance which is an especially pressing issue in the design of these time and safety-critical systems.
This project aims to implement vehicle lane detection into hardware using the Xilinx PYNQ-Z2 FPGA board in order to produce an autonomous robot. To gather data from the environment, a camera onboard the robot captures and transfers frames as input data to the FPGA board. Image processing algorithms are then used to process the information and make decisions based on the results. At this stage, the usage of hardware acceleration maximizes the performance speed of these algorithms in which the computation load is transferred to the FPGA fabric from the CPU. Finally, based on the decision made, the board controls the motors driving power to the wheels in order to direct the movement of the robot
Active Suspension System
No full textThis thesis presents the design, development, and prototyping of an active suspension system intended for high-performance racing applications, with a focus on karting. The goal of the project was to create a geometrically active suspension capable of dynamically adjusting key parameters such as camber, caster, toe, and track width in real-time to improve grip, cornering stability, and overall vehicle performance.
The team conducted extensive research into suspension theory, customer needs, and existing industry solutions before developing a mechanical platform based on a Short-Long Arm (SLA) double wishbone configuration. Using linear actuators to introduce five degrees of freedom, the system was designed to actively modify suspension geometry during various phases of cornering, braking, and acceleration.
Due to time and resource constraints, the prototype focused on mechanical design, integration, and preliminary testing. While the control system remains a future development goal, the foundation for real-time actuation and sensor feedback has been established. Results from mechanical simulations and basic functional tests confirm that the system is capable of achieving meaningful geometric adjustments without compromising structural integrity.
This project demonstrates the feasibility of a mechanically complex active suspension system for racing use and lays the groundwork for future enhancements, including control logic development, predictive capabilities, and machine learning integration
Radicalization and Exodus: Analyzing the Impact of Governmental Extremism on Migration Patterns in Venezuela
No full textVenezuela’s political and economic crisis has led to one of the largest migratory movements in Latin American history. Over the past two decades, millions of Venezuelans have fled due to political repression, economic collapse, and humanitarian emergencies. This project examines general migration patterns from Venezuela between 1999 and the present. Using data from the United Nations Department of Economic and Social Affairs (UNDESA) and World Population Prospects, the study analyzes how political radicalization and economic instability have influenced migration trends over time. The findings provide crucial insights into the scale and nature of this mass displacement and its implications for both migrants and host nations.
The findings indicate a significant correlation between governmental radicalization, economic collapse, and mass migration. The research categorizes migration trends into three political phases: the Chávez era (1999-2013), Maduro’s early years (2013-2018), and the ongoing humanitarian crisis (2018-present). Data analysis reveals that migration patterns directly correlate with major political events, including oil price fluctuations, hyperinflation, and governmental crackdowns on opposition. Furthermore, the analysis reveals that migration has transformed from a gradual economic outflow to an urgent survival mechanism, with entire families fleeing due to food insecurity, hyperinflation, and violence. Additionally, the project highlights gendered migration trends, with a consistently higher proportion of female migrants, emphasizing the importance of gender-sensitive migration policies. It also suggests that women face unique socio-economic vulnerabilities and pressures in Venezuela that drive their displacement
Children’s Merchandising
No full textThis paper explores how gender is constructed and reinforced through children\u27s merchandising at Target, focusing on the toy, clothing, and greeting card departments. Using qualitative observational methods, I documented the ways in which colors, themes, and product placements subtly but powerfully separate items by gender. My findings reveal that gendered marketing is deeply embedded in retail environments and reflects broader societal expectations surrounding masculinity and femininity. These early messages shape children\u27s self-perceptions, behaviors, and even long-term aspirations. Drawing on sociological theories of gender socialization and primary framing, as well as existing literature from Lips, Wade, and Ferree, this study situates retail merchandising as a key site of cultural transmission. The analysis also considers the role of intersectionality and the limitations of inclusivity in consumer spaces. Ultimately, the paper argues that while some families may attempt to resist gender norms, the pervasive binary marketing in major retail spaces continues to shape children’s understanding of what roles are appropriate for boys and girls, reinforcing systemic inequality from a young age