SURF

Thomas Yerxa

Learning Programs with Hyper Dimensional Computing

Computers learn to solve problems by slowly and continuously changing their strategy, this makes solving problems that involve discrete steps problematic. I will be researching the use of hyper dimensional computing to implement differentiable memory structures that interface with neural networks. Many data structures are inherently discrete which makes them difficult to incorporate into schemes because of the strategy utilized by computers during the training process. Recent work has shown that neural networks that have the ability to learn how to use external data structures outperform traditional networks in an important subset of tasks that are typically tackled with machine learning. This implies they may be an important tool in solving complex problems such as machine translation. I am interested in using hyper dimensional computing to develop a new version of this tool. Hyper dimensional computing involves leveraging the unique properties of vectors with many dimensions to develop non-traditional computing paradigms. Computers learn to solve problems by slowly and continuously changing their strategy, this makes solving problems that involve discrete steps problematic. I will be researching the use of hyper dimensional computing to implement differentiable memory structures that interface with neural networks. Many data structures are inherently discrete which makes them difficult to incorporate into schemes because of the strategy utilized by computers during the training process. Recent work has shown that neural networks that have the ability to learn how to use external data structures outperform traditional networks in an important subset of tasks that are typically tackled with machine learning. This implies they may be an important tool in solving complex problems such as machine translation. I am interested in using hyper dimensional computing to develop a new version of this tool. Hyper dimensional computing involves leveraging the unique properties of vectors with many dimensions to develop non-traditional computing paradigms.

Message to Sponsor

Thank you so much for supporting my research this summer. Being able to work full time allowed me to partake in an immersive research experience that has confirmed by desire to pursue a graduate degree. Working on my own project was an excellent way to learn about how research is conducted, and the obstacles that come along with that. During the course of my work this summer I met new faculty members that I will continue working with throughout the year, and gave me the opportunity to focus on developing skills that will be critical throughout my research career.
  • Major: Physics
  • Sponsor: Guthrie Fund
  • Mentor: Michael Deweese