Research Group

Internet Services Research Center (ISRC), Microsoft Research, Redmond.

Education Background

Ph.D. in Electrical and Computer Engineering, University of California, San Diego, 2011.

• Advisor: Prof. Bhaskar D. Rao

B.E. in Computer Science and Technology, Tsinghua University, Beijing, China, 2005.

Research Interests

    Sparse Signal Recovery (Compressed Sensing)

I am strongly interested in the theory, algorithms, and applications of sparse signal recovery. Particularly, I have been focusing on the performance limits of optimal algorithms as well as practical algorithms. These performance limits provide theoretical insights into the theoretical boundaries where successful sparse signal recovery is still possible. To achieve these performance limits, an important analytical tool I developed is an information theoretical framework that are flexible enough to explain the performance limits of many models of sparse signal recovery.

Motivated by the connection between sparse signal recovery and multiple user communication, I realized that the optimal order of reconstructing the nonzero entries of sparse signals is highly related to the capacity region of a correspondng MAC. Taking advantage of the dynamic range of the nonzero entries, I designed the MultiPass Lasso algorithm which is both effective and efficient for sparse signal recovery.

I have been also actively working on applications that can enjoy the power of sparse signal recovery techniques. I found that the problem of robust regression can be naturally cast as a sparse signal problem once the outliers are modeled as a linearly additive component. As a result, many existing techniques for finding sparse solutions can be immediately extended to address robust regression and outlier detection. Another area where I found sparsity helps is adaptive filtering for sparse channels, which are often seen in wireless communication scenarios.

    Semantic Entity Resolution

After joining ISRC, I have been working on the entity related research in the context of Web search and information processing. I found this area very exciting since it provides great opportunities for sparse signal recovery techniques. Understanding the entities mentioned in an unstructured Web document and their relations is a key step to learning deep structures in the huge amount of data available on the Internet. It will enable novel user experiences as well as new applications that can better organize and present information as well as interact with all the participant in the Web eco-system.

    Low-Energy GPS

I am also collaborating with the Energy and Sensing Research Group (SERG) on Low energy GPS sensing. It is another application domain where I found the principles for finding sparse signal naturally apply.

Latest News

    Organizing ICASSP 2013 Special Session on Sparse Signal Techniques for Web Information Processing

The area of sparse signal recovery has received lots of research attention recently. The sparsity nature underlying various signals has proven substantial to the successes of many applications, such as speech coding, adaptive filtering, channel equalization, medical imaging, robust regression, computer vision, etc. On the other hand, the vast amount of information on the Web, for example in the forms of news, commerce, consumer reviews, online forums, blogs, and social networks, etc., present unprecedented challenges for researchers and practitioners in improving the experiences for various participants of a Web-scale ecosystem. The special session aims to inspire future research potentials in this multi-disciplinary topic and broaden the applications of sparse signal processing theories and algorithms.

Co-organizing with Dr. Kuansan Wang.

    Video tutorial on Compressed Sensing (available to MS FTE only)

If you wish to spare 20 minutes to have some fun after a day full of work --- look no further :-)  With many colorful and intuitive illustrations, this video tutorial introduces the problem of compressed sensing and walks you through a series of seemingly different applications which can all benefit from a sparsity formulation. We also peek at several highly active research fronts in this area.

Please feel free to contact me for more information. 

Publications

Sparse Lexical Representations for Semantic Entity Resolution, Yuzhe Jin, Kuansan Wang, and Emre Kiciman, to appear in ICASSP 2013.

Support Recovery of Sparse Signals in the Presence of Multiple Measurement Vectors, Yuzhe Jin, Bhaskar Rao, to appear in IEEE Transactions on Information Theory, 2013.

Sparse Signal Recovery in the Presence of Intra-Vector and Inter-Vector Correlation (Invited paper), Bhaskar D. Rao, Zhilin Zhang, Yuzhe Jin, International Conference on Signal Processing and Communications (SPCOM 2012), Bangalore, India, July, 2012

MultiPass Lasso Algorithms for Sparse Signal Recovery, Yuzhe Jin, Bhaskar D. Rao, ISIT 2011, St. Petersburg, Russia.

Sparse Signal Recovery: A Multiple-User Information Theoretic Viewpoint, Yuzhe Jin, Bhaskar Rao, ICML Structured Sparsity Workshop, 2011.

On the Role of the Properties of the Nonzero Entries on Sparse Signal Recovery (Invited paper), Yuzhe Jin, Bhaskar Rao, 2010 Asilomar Conference on Signals, Systems, and Computers.

Limits on Support Recovery of Sparse Signals via Multiple-Access Communication Techniques, Yuzhe Jin, Young-Han Kim, Bhaskar Rao, IEEE Trans. on Information Theory, Dec. 2011.

Performance Tradeoffs for Exact Support Recovery of Sparse Signals, Yuzhe Jin, Young-Han Kim, Bhaskar Rao, ISIT 2010, Austin, USA.

Algorithms for Robust Linear Regression by Exploiting the Connection to Sparse Signal Recovery, Yuzhe Jin, Bhaskar Rao, ICASSP 2010, Dallas, USA.

Performance Limits of Matching Pursuit Algorithms, Yuzhe Jin, Bhaskar Rao, ISIT 2008, Toronto, Canada.

Insights into the Stable Recovery of Sparse Solutions in Overcomplete Representations using Network Information Theory, Yuzhe Jin, Bhaskar Rao, ICASSP 2008, Las Vegas, USA.

Spectral Estimation of Voiced Speech Using a Family of MVDR Estimates, Rajesh M Hegde, Yuzhe Jin, Bhaskar Rao, ICASSP 2007, Hawaii, USA.