Pan Zhang

I am working at the Institute of Theoretical Physics, Chinese Academy of Sciences as an associate professor. My research is in the interdisciplinery field between statistical physics, applied mathematics and computer science.

News:  
• At the ITP, CAS we are organizing International workshop on Physics, Inference and Learning during Oct.29  Nov.3, 2018.  
• I am looking for a postdoctoral research fellow in the field of statistical physics and machine learning.  
Research:  
Probably lots of the readers are familiar with the Boltzmann machine which models joint probability distribution of data using Boltzmann distribution. Boltzmann machine is a great contribution from Statistical Physics to Machine Leanring. In this Physical Review X paper we propose a fresh unsupervised machine learnig model borrowed from Quantum Physics, which models the joint distribution of datea using Born's rule. Thus we call it Born Machine. This model connects tensor networks and generative modeling. You can find a tutorial and a jupyter notebook on the topic of tensor network, matrix product state, and generative learning.  
Spectral methods are popular in detecting global structures in the given data that can be represented as a matrix. However when the data matrix is sparse or noisy, classic spectral methods usually fail to work, due to localization of eigenvectors (or singular vectors) induced by the sparsity or noise. In this paper ( NIPS 2016) we propose a general method to solve the localization problem by learning a regularization matrix from the localized eigenvectors. Here is a Demo for the algorihtm "XLaplacian".  
Many realworld networks are dynamic, with nodes changing their connections and affiliations over time in complicated ways. This situation makes community detection more challenging, but correlations across time provide a means to circumvent this issue. In this Physical Review X paper, we derive a precise mathematical limit on our ability to recover the underlying community structure in a dynamic network, which depends only on the strength of the hidden communities and the rate at which nodes change their community membership.  
Maximizing modularity is the most popular
method of detecting communities in networks. However it is
prone to overfitting. In this PNAS paper, with Cris Moore we proposed to solve this overfitting problem using ideas from statistical physics, and developped an efficient algorithm for detecting communites and hierarchies in large networks.  
Spectral algorithms are popular method of clustering data. However they often fail in sparse networks, because of existence of localized eigenvectors. In a paper published in PNAS, with collaborators we gave a new spectral algorithm based on the nonbacktracking operator that is immune to this disease, and works very well in large sparse networks. 

Code:  
Data clustering using message passing: C++ code  
XLaplacian: Demo Preprint  
Spectral clustering using the Nonbacktracking matrix: Matlab code paper (open access)  
Message passing for modulairty: C++ code paper preprint  
Inference of the Stochastic Block Model by Belief Propagation: C++ code paper  
A message passing based complete solver for Quantified Boolean Formulas: C++ code paper  
Inference of the Kinetic Ising model on sparse graphs using dynamic cavity method: code paper  
Contact me:  Last modified: Sep. 26, 2018 