Siheng Chen, a Ph. D student from Carnegie Mellon University, visited our lab on May 23th. He gave a talk on "Discrete signal processing on graphs: graph signal inpainting" in Room 9-206, Rohm Building at 10:30 and then met with Professor Gu and students in the afternoon.
Fig.1 Siheng Chen is giving the talk
Fig. 2 Siheng Chen is explaining shift in discrete signal processing
Siheng Chen received his B. Eng degree in optoelectronics engineering in 2011 from Beijing Institute of Technology and the M.S degree in electrical and computer engineering in 2012 from Carnegie Mellon University. He is currently a second year Ph. D student in the Department of Electrical and Computer Engineering and the Center for Bioimage Informatics at Carnegie Mellon University. His research interests include graph signal processing, indirect bridge health monitoring and biomedical image analysis.
Discrete Signal Processing on Graphs: Graph Signal Inpainting
Massive data, generated from various networks, such as social, economics, neuroscience, are becoming increasingly larger. Those data are naturally supported on complex and irregular structures that are commonly represented by graphs. The theory of discrete signal processing on graphs provides a novel framework for analyzing high-dimensional data with complex and irregular structures. The framework extends classical signal processing concepts, including signals, filters, spectrum, Fourier transform, frequency response, and filtering from data on regular lattices to data on arbitrary graphs. This theory offers a new methodology for solving data analysis tasks, such as data compression, denoising, classification, and anomaly detection. In this talk, we will focus on the task of inpainting for data on graphs. Our approach is a generalization of the signal inpainting technique from classical signal processing. We formulate corresponding minimization problems and demonstrate that in many cases they have closed-form solutions. We will discuss the relation between the proposed approach and regression, provide an upper bound on the error for the algorithm, and compare with other existing algorithms on real-world datasets.