Browsing by Author "Karbasi, Seyedeh Paniz, 1986-"
Now showing 1 - 3 of 3
- Results Per Page
- Sort Options
Item A fast seeding technique for k-means algorithm.(2014-11-06) Karbasi, Seyedeh Paniz, 1986-; Hamerly, Gregory James, 1977-The k-means algorithm is one of the most popular clustering techniques because of its speed and simplicity. This algorithm is very simple and easy to understand and implement. The first step of this algorithm is choosing k initial cluster centers. The way that this set of initial cluster centers are chosen, have a great effect on speed and quality of k-means. One of the most popular seeding techniques is k-means++ initialization, but this method needs k passes over the dataset. The goal of this thesis is to propose a new seeding technique which chooses the initial centers much faster than k-means++.Item Recommendations Made Easy(2014-06-23) Guinness, Darren, 1990-; Karbasi, Seyedeh Paniz, 1986-; Nazarov, Rovshen; Speegle, Gregory David.Fueled by ever-growing data, the need to provide recommendations for consumers, and the considerable domain knowledge required to implement distributed large scale graph solutions we sought to provide recommendations for users with minimal required knowledge. For this reason in this paper we implement a generalizable 'API-like' access to collaborative filtering. Three algorithms are introduced with three execution plans in order to accomplish the collaborative filtering functionality. Execution is based on memory constraints for scalability and our initial tests show promising results. We believe this method of large-scale generalized 'API-like' graph computation provides not only good trade-off between performance and required knowledge, but also the future of distributed graph computation.Item Robust and efficient methods for proton computed tomography.(2018-07-23) Karbasi, Seyedeh Paniz, 1986-; Schubert, Keith Evan.Proton computed tomography (pCT) is a recent promising imaging modality with the goal of generating accurate 3D maps of relative stopping power (RSP) with respect to water. Since the early developments of this imaging technique in 1970's, there have been significant improvements regarding the reconstruction of accurate RSP which makes pCT a reliable alternative to X-ray CT for planning proton therapy treatments. There are several conditions in pCT that can negatively affect the accuracy of pCT images. The goal of this dissertation is developing efficient image reconstruction methods generating accurate RSP values under both normal and critical conditions.