Architecting Computer Vision Workloads on the Cloud; The Importance of Effective Governance, Accountability, and Organizational Responsibility in Cloud Computing

Fernandes, Dylan, School of Engineering and Applied Science, University of Virginia
Shen, Haiying, EN-Comp Science Dept, University of Virginia
Wayland, Kent, EN-Engineering and Society, University of Virginia

Cloud computing has become an incredibly popular technology since Amazon released the first public cloud in 2002. Businesses that choose to use the cloud can take advantage of nearly infinite server capacity, auto-scaling to match the demand for software applications, and an advanced software engineering toolkit that makes development easier. These features offer a competitive advantage to businesses compared to traditional data centers, and as such, adoption of the cloud has increased tremendously, generating over 400 billion dollars in revenue in 2021. Both the technical project and sociotechnical thesis contained within this portfolio are an exploration of cloud computing, including potential applications, risks, and governance of this technology.

The cost of surgery is tremendously high, and a large contributor to that high cost is single-use sterile tools. If a single-use sterile tool is opened, it must be thrown away at the end of every surgery, regardless of whether the tool was used or not. Surgeons are not sure of what inventory to bring into the operating room, so they often tend to request more than they need. To fully understand what tools are needed, we realized that we could track what tools were being used during surgery, and provide recommendations for what tools to use for specific surgeries over a longer period of time. The technical aspect of this portfolio is a system designed to take operating room footage as input and produce a list of disposable items and their locations in each frame as output using machine learning, computer vision, and cloud computing. The system was constructed to be event-driven and scalable, designed to handle one customer or hundreds simultaneously. To reduce wait times for processing, the system breaks the process into stages and runs jobs in parallel. The system was also built using infrastructure as code and developer operations best practices. If the system goes down, a copy of the system can be redeployed in minutes and can receive new jobs while the old one is repaired. Storage is also practically infinite as well, stored in several different locations with the highest level of durability available. While testing of our system is incomplete as of this time, we estimate that two hours of operating room footage can be analyzed within as little as half an hour. The infrastructure was built using the AWS cloud, diligently following best practices to ensure that our system was highly available. The code itself was written using a combination of Python and Typescript.

The sociotechnical research paper focuses more on the risks of using cloud computing, especially in relation to data privacy. The paper argues that there are inherent risks of entrusting a third party with sensitive data, including the loss of control and ownership over data, varying definitions of privacy for different organizations (especially those across borders), and the ‘problem of many hands’ that makes proper accountability challenging. The exposure of consumer data because of these risks can ruin livelihoods and destroy businesses, so it is important to ensure that this data is protected. Since organizations - including governments, cloud service providers, and businesses that use the cloud - are the entities with the most power within the network of cloud computing, it is their duty to protect society’s data, especially the customers that they serve. The paper then offers a look into the literature, combining multiple sources to provide recommendations from advocates and experts for organizations to implement. These recommendations range from accountability frameworks for businesses and service providers to standards for regulations for governments.

I was incredibly proud of the work I produced for my capstone project and my thesis. I originally started out in a completely different direction for my thesis, but after pivoting a few times, I was really happy with the end result. Writing my thesis showed me that technological advances can be used to do a lot of good, but if not regulated and used properly, can have unintended consequences that can do more harm than good. My partner and I accomplished more with the technical project than I thought possible, and we learned so much while developing this very intricate system on the clouds. We enjoyed working on the project so much that we plan on continuing to work on this project throughout the summer. In terms of the next steps, we’re planning on passing the output produced by the machine learning model to an Extended Kalman Filter to track the objects on the scrub table and determine what disposable items are used during surgery.

BS (Bachelor of Science)
Cloud computing, Operating Room, Cloud

School of Engineering and Applied Science
Bachelor of Science in Computer Science
Technical Advisor: Haiying Shen
STS Advisor: Kent Wayland
Technical Team Members: Arvind Anand

All rights reserved (no additional license for public reuse)
Issued Date: