41530 research outputs found
Sort by
Drilled Displacement Columns as a Method to Mitigate San Francisco \u27Bay Mud\u27
Full text linkThis case study investigates the process of drilled displacement columns as a substitute for traditional pile foundations. The setting for the project is in California\u27s infamous \u27bay mud\u27 which poses in the form of building settlement. The case study begins with a run down of what drilled displacement columns are compared to traditional pile foundations. To gain a diverse perspective several parties were interviewed on the process of drilled displacement columns. The interviewees include members of the general contractor\u27s team, the subcontractor, and the Owner\u27s Representative. The universal positives were time and materials which result in reducing the price for foundations. The case study also explores challenges associated with the drilled displacement column process on this project but are not always present with drilled displacement columns. Ultimately, the case study justifies the use of drilled displacement columns to support substructures. Lastly, future research is proposed to validate the effectiveness of drilled displacement columns over time. Saving money and time are key elements on any project but the value of those returns can only be understood over time as the building performs and responds to its environment. Taking measurements of future settlement can further support the conclusions made from the research
Secure. Contain. Piano.
No full textSecure. Contain. Piano. is a five-movement piano cycle inspired by modern folklore entitled Secure. Contain. Protect., or SCP. This folklore stems from an online encyclopedia documenting fictional monsters referred to as SCPs. The cycle is based on a story about a prisoner escaping a government facility which also contains these SCPs. Each movement portrays a mixture of dread and comedy as the prisoner runs through the prison, desperately searching for the exit and avoiding monsters on the way
PeerProxy: A WebRTC Proxy for HTTP
Full text linkAdvances in networking technologies have empowered individuals to easily self-host digital services such as websites and smart home systems. However, accessing these services externally often requires port forwarding, which requires manual router configuration, technical expertise in networking, and is sometimes restricted by internet service providers. Proxy-based services such as Ngrok and Cloudflare Tunnels simplify external access by using publicly hosted proxy servers, but introduce increased infrastructure costs and privacy concerns due to reliance on third-party servers that can inspect or store traffic.
This thesis presents PeerProxy, a novel framework that simplifies access to self-hosted web services without manual network configuration, privacy risks, or increased infrastructure costs. PeerProxy is built on WebRTC, a set of protocols built into modern web browsers to form end-to-end encrypted peer-to-peer and proxied connections. This solution includes a custom local proxy server, a specialized browser client that loads web applications over WebRTC in unmodified browsers, and a lightweight signaling server for connection management. Additionally, this work introduces a custom packet protocol for efficiently transmitting HTTP messages over WebRTC data channels.
Performance evaluations show that while PeerProxy’s download throughput (2.6 MB/s) is lower than traditional proxies (14.08 MB/s) due to WebRTC limitations, it maintains comparable latency for requests up to 1KB. Future proposed WebRTC improvements, such as RTCQuicTransport, could enhance its performance. Additionally, resource utilization tests confirmed that PeerProxy\u27s browser client adds minimal overhead.
This research contributes a working prototype of PeerProxy, a thorough evaluation of HTTP proxying performance over WebRTC, and insights into the current limitations of high-throughput applications using modern browsers\u27 implementations of WebRTC. It lays a foundation for further advancements in secure and decentralized web service hosting that empower users with secure, private, and efficient external access to self-hosted services
Operating Room Forcing Function
Full text linkThe Operating Room Forcing Function system aims to address the persistent issue of wrong-site, wrong-procedure, and wrong-patient errors (WSPEs) in surgical settings. WSPEs are largely caused by circumvention of the timeout procedure (a planned pause before incision to stop and confirm crucial details of the procedure to be performed known as ‘The Universal Protocol’). The goal set forth by our sponsor, Dr. Robert Turbow, is to develop a lock-out mechanism that requires a timeout to be performed by all appropriate personnel before an operation can begin.
Hospitals are extremely complex systems with competing values, productivity pressures, fast paced work, tight-coupled systems, non-linear interactions, and a punitive culture which can all make room for short cuts and noncompliance. These pressures, along with poor human factors consideration in hospital and device design, allow for circumvention of safety protocols and in some cases, catastrophic failure.
Initial investigation of the problem highlighted that the system needed to be intuitive to learn, provide some sort of visual/audible feedback, and provide multiple levels of verification while still allowing for a total system override in case of emergencies. After defining the problem and key customer requirements, multiple design concepts were created and evaluated in their functionality and ability to address the issue at hand. To create a system that exceeds the Universal Protocol in reliability, we planned to implement a multi-layered design that ensures/forces compliance with the timeout procedure while remaining efficient and unobstructive to normal surgical workflow with a special emphasis on human factors engineering. Interviews with operating room staff and observation of live timeout procedures provided information and feedback throughout the course of the design process.
A final concept was selected, and a prototype system was created. The design proposed in this report entails an audio playback of the timeout protocol where each piece of information must be individually addressed and confirmed. The system includes a set of pedals that operating room staff members must use to simultaneously signal their agreement with each set of information. Only once all information has been confirmed and the timeout checklist is complete, the system switches power on to an external component (Bovie machine, unlocking sharps container, overhead lights, etc.) to provide the lock out functionality. This system ensures that all members are present and attentive during the timeout procedure and that surgery cannot begin until all members are in anonymous agreement of all procedural information.
Tests were performed to ensure the design provided all the desired functionality and met customer requirements initially defined. These tests yielded reliable device functionality, positive user feedback and satisfaction, and indications for reducing the risk of WSPEs and enhancing overall patient safety
Minifying Deep Denoising Networks with Knowledge Distillation
Full text linkHearing loss is a prevalent condition, affecting hundreds of millions globally, with a higher incidence among older adults. While hearing aids are the standard treatment, the majority of those who could benefit from hearing aids choose not to wear them, attributing this decision in large part to their inability to perform well in conversations in large groups and in noisy situations. To date, no denoising systems on commercial hearing aids are able to improve speech intelligibility. Recent advances in artificial intelligence research have shown that large deep-learning models can in fact improve speech intelligibility by removing background noise from audio. However, these models are far too large and computationally expensive to be run in real-time on a small embedded system like a commercial hearing aid.
This work builds upon the success of recent speech enhancement deep-learning models by investigating and measuring the effectiveness of knowledge distillation in creating a smaller and faster model. In using knowledge distillation, we extract the capabilities of a larger, more powerful teacher model into a smaller, more efficient student model, optimizing its suitability for embedded processors with limited computational resources.
The student model’s denoising ability was statistically significantly greater than that of the control model across a wide range of noise conditions to a very high confidence, particularly in the [2.5, 4] pMOS range, where it showed the most substantial improvements. This range is especially important, as it aligns closely with the conditions under which the model is most likely to be useful in real-world scenarios.
These findings demonstrate that knowledge distillation can enhance the performance of compact U-Net speech denoising models, making it a powerful and complementary tool for engineers optimizing audio systems for resource-constrained, real-world applications
A Decision Support System for Conference Session Selection using Natural Language Processing
Full text linkConference attendees are faced with selecting from hundreds to thousands of presentations and sessions in pursuit of new findings and methods relevant to their area of interest, an overwhelming amount of information from which to clearly make a decision. To address this, we developed a decision support system leveraging natural language processing (NLP) techniques such as semantic matching. By creating and matching embeddings of conference presentation abstracts and titles, the application provides improved query matching compared to keyword searching. We introduce Session Scout, a novel conference decision support system built upon a semantic retrieval framework. Session Scout is designed to move beyond keyword-matching by understanding the meaning and context behind an attendee\u27s interests and the content of conference presentations. Users can select from a variety of input types, including keyword selection and natural language query, and receive a collection of presentations with abstracts semantically matching the input. Notably, users can input longer queries (e.g. an abstract of a paper most relevant to their area of interest) and receive a collection of relevant conference presentations. Key contributions of this work include the application of semantic matching to the conference presentation recommendation problem. They also include the evaluation of Session Scout\u27s model on two unsupervised datasets, including the use of LLM as a judge to address this challenge. Additionally, Session Scout has been deployed and tested at a live conference, receiving positive user feedback