Project: Data Representation: Learning Kernels from Noisy Data and Uncertain Information
(ARO
56976-NS)
PI: Rong Jin, Department of CSE, Michigan State University
Co-PI: Anil K. Jain,
Abstract:
Identifying appropriate data representation is critical to many problems in pattern recognition, data mining, and machine learning. Compared to the vector-based representation, kernel-based data representation is more flexible and is particularly suitable for complex objects like trees and graphs that are difficult to be captured by vector-based representation. In this project, we focus on the problem of automatically learning kernel-based data representation from noisy data and uncertain information. This is contrast to most current studies on kernel learning that assume an ideal observation or sensing of objects without any noise. The objective of this project is to develop efficient computational frameworks for learning a robust combination of multiple kernel data representations from noisy data observations and uncertain supervised information. In the proposed research, we aim to develop the following approaches to address the key challenges in kernel learning with noisy data and uncertain information : (1) develop an efficient computational framework for multiple kernel learning that is resilient to the noise in data observation, and (2)
develop an efficient computational framework for multiple kernel learning that is robust to the uncertainty in class assignmentStudents