Nicholas Navaroli
NDSEG Fellow
4059 Donald Bren Hall
University of California, Irvine
| Lab: | UCI Datalab |
| Adviser: | Padhraic Smyth |
| E-mail: | nnavarol@uci.edu (nnavarol [at] uci [dot] edu) |
Nicholas NavaroliNDSEG Fellow4059 Donald Bren Hall
|
|
I am a second year graduate student at University of California, Irvine. My adviser is Padhraic Smyth and I am a part of the UCI Datalab group, a group whose research interests involve modeling and analyzing datasets. I received my B.S. degree in Computer Science and minor in Mathematics at California State University San Bernardino in 2009, and am currently working towards both my M.S. degree in Computer Science and Ph.D. in Machine Learning.
Click here to view my CV.
I am interested in modeling large datasets that contain noisy and/or missing data. Datasets generated by large sensor networks (traffic sensors, weather sensors, etc) tend to have these characteristics. Models that handle such datasets must be able to both accurately and efficiently describe the underlying relationships of the dataset. For my current research I am investigating how to efficiently model these datasets using Markov Random Fields and data augmentation algorithms.
I am also interested in clustering, image analysis, supervised learning and probabilistic modeling.
J. R. Foulds, N. Navaroli, P. Smyth, A. Ihler. Revisiting MAP Estimation, Message Passing and Perfect Graphs. Proceedings of the 14th International Conference on AI and Statistics (AI Stats), April 2011.
N. Navaroli, D. Turner, A. Concepcion, R. Lynch. Performance Comparison of ADRS and PCA as a Preprocessor to ANN for Data Mining. The 8th International Conference on Intelligent Systems Design and Applications (ISDA 2008), Nov 2008, Kaohsuing, Taiwan.
I worked with Arturo Concepcion and David Turner at CSUSB on benchmarking the performance of Bayesian Data Reduction Algorithm (BDRA), developed by Robert S. Lynch. BDRA is a probabilistic classifier that eliminates "irrelevant" features from a labeled training dataset in order to improve the accuracy of classifying test datasets. We developed experiments to compare classification errors between BDRA and an Artificial Neural Network, and showed that in many cases BDRA can acheive smaller classification errors.
In Summer 2008, I worked with Rolland Trapp in the Mathematics department at CSUSB on optimizing the structure of "knots", arrays of 3-dimensional vertices that represent loopy strings. The vertices and angles between edges define various energies, such as curvature energy (how curved the string is). We developed algorithms that manipulated the locations of vertices in order to minimize these energies, while maintaining overall structure.
In Summer 2008, I worked with Haiyan Qiao at CSUSB on developing numerous clustering algorithms. I evaluated the performance of these algorithms using multiple metrics, including the Dunn Index and Davies-Bouldin Index.
In Summer 2007, I participated in the Bio-Image Internship at UCSB, an internship that focused on developing programs and tools for biologists to use. I worked with Luca Bertelli on implementing several existing image segmentation algorithms (such as normalized graph cuts and the fast marching method). We also developed a competitive algorithm similar to fast marching method, using only pixel colors. We applied the algorithms on images of cat retina to detect photoreceptors and describe their characteristics, and to images of breast cancer cells in order to detect various nuclei and their patterns.