Research

I am dedicated to solving some real-world important problems. I have broad interests, including data mining, machine learning, artificial intelligence, information retrieval, social networks. I have publications in top tier Data Mining and Artificial Intelligence Conferences.

Patent

A provisional patent (Z.Kang and Q.Cheng) on Top-N Recommender System has been granted in 2015.

Projects

Unified Clustering Frameworks, June 2016

Many clustering methods are based on several separate steps. This could lead to information loss and suboptimal solutions. I try to develop some unified frameworks.

Top-N Recommender Systems, June 2015

Based on data mining and machine learning techniques, several novel recommendation algorithms are proposed. They show superior performance.

Low Rank Representation based on Novel Ranks Approximations, November 2014

Matrix rank minimization problem is in general NP-hard. The nuclear norm is used to substitute the rank function in many recent studies. Nevertheless, the approximation error may depend heavily on the magnitudes of singular values, which might restrict its capability in dealing with many practical problems. We propose some tighter approximations to the rank function and show their applications in subspace clustering, robust PCA, and matrix completion.

Use of Unsupervised Machine Learning to Identify Patient Subgroups with Advanced Heart Failure and Reduced Ejection Fraction Most Likely to Derive a Survival Benefit from Bucindolol, September 2015

The results of randomized clinical trials represent average effects of an investigational treatment or placebo on a heterogeneous patient population. Thus the overall effect of an intervention represents the sum of its effects on different patient subgroups. We hypothesized that unsupervised machine learning techniques would identify subgroups of patients with heart failure and reduced ejection fraction (HFrEF) treated with bucindolol with substantially reduced mortality compared to placebo treatment.

Nonnegative Matrix Factorization, September 2014

Nonnegative matrix factorization is a useful technique for data representation in many data mining and machine learning tasks. We propose graph regularized NMF method capable of feature learning, and apply it to clustering. Unlike existing NMF methods that treat all features in the original feature space equally, our method distinguishes features by incorporating a feature-sparse approximation error matrix in the formulation.

Deep Learning in Quantum Representation, August 2013

Network structure is crucial to the performance of deep learning algorithm. We propose to automatically learn the number of nodes and weights from the data. This method borrows the idea from physics and uses quantum representation. It assigns a probability to each node, which indicates the existence chance of a node. The probability is updated during the optimization process.

Image Denoising with A Convex Framework, March 2014

Using a high-pass filter and the Schatten-p norm, we constrain the rank of the filtered image, while preserving the global features of the image.