A new paper "Stochastic blockmodel approximation of a graphon: Theory and consistent estimation" has been presented in NIPS 2013.
Traditionally, network analysis has been focusing on parametric models. That is, given a set of observed data, find a model (using a few parameters) that best describes the data. Approaching from an opposite perspective, we consider a non-parametric perspective of network analysis. For a class of exchangeable random graphs, we show that it is possible to find a non-parametric model as the limiting object of a sequence of convergent graphs. This limiting object is called graphon in the statistics literature. We propose a polynomial time algorithm for estimating the graphs. Our method is the first frequentist approach in the literature with consistency guarantee.
Paper is available at http://arxiv.org/abs/1311.1731
Code is available at https://github.com/airoldilab/SBA
I am a post-doctoral research fellow at Harvard School of Engineering and Applied Science, under the support of the Croucher Foundation Post-Doctoral Research Fellowship.
My current research focuses on statistical signal processing for large-scale data, with long term efforts in developing theories and algorithms to recover latent structures and processes leveraging the data. Topics of interest include signal reconstruction, sampling theory, network analysis and numerical optimization. In particular, I work on the following projects:
- (1) Random Sampling for Image Filtering
- (2) Single-Photon Imaging Sensors
- Network Analysis:
- (1) Nonparametric Graphon Estimation by Stochastic Block Approximation
- (2) Total Variation based Graphon Estimation Algorithms
You are welcome to visit my old website at UC San Diego: http://videoprocessing.ucsd.edu/~stanleychan
|A shorter version of our NIPS paper has been presented in the 1st IEEE Global conference on Signal and Information Processing (GlobalSIP '13), Austin, TX, December, 2013.|
|A new paper "Adaptive Non-local Means for Multiview Image Denoising - Searching for the Right Patches via a Statistical Approach" is presented in IEEE International Conference on Image Processing (ICIP 2013). In this paper, me and my colleagues at UC San Diego propose an adaptive denoising scheme for multiview imaging systems. Different from the classical non-local means where the number of patches is fixed, we propose to adaptively choose the best patches for denoising. For more details of this project, please visit http://videoprocessing.ucsd.edu/~eluo/projects/multiview_denoising/|
|A new paper "Fast non-local filtering by random sampling" is presented in IEEE International Conference on Acoustics Speech and Signal Processing (ICASSP 2013). In this paper, Yue and I applied large deviation principles (LDP) to study a very popular class of image denoising algorithms known as the non-local filtering. We showed that with extremely high probability, the computationally intensive non-local filtering operations can be approximated using a Monte Carlo random sampling scheme. In addition, we found that with one additional cost-less step by normalizing the columns of the graph adjacency matrix, the asymmetric non-local filter becomes symmetrized according to Sinkhorn-Knopp. This symmetrized filter further boosts up denoising quality by a big margin.|