Shuiwang Ji BDA 224BB, BY 584 Computer Science and Engineering Center for Evolutionary Functional Genomics, The Biodesign Institute Arizona State University Tempe, AZ 85287 U.S.A. E-mail: shuiwang.ji@asu.edu

Shuiwang Ji is a Ph.D. candidate in the Department of Computer Science and Engineering, Arizona State University, advised by Prof. Jieping Ye. He also works closely with Prof. Sudhir Kumar in the Center for Evolutionary Functional Genomics of The Biodesign Institute on the FlyExpress project. His research interests are computational biology, machine learning, and computer vision.

News:

• Jun Liu, Shuiwang Ji, and Jieping Ye will give a tutorial entitled "Mining Sparse Representations: Theory, Algorithms, and Applications" to The 2010 SIAM International Conference on Data Mining.
   
• Shuiwang Ji has been invited to serve on the program committee of The 27th International Conference on Machine Learning (ICML 2010).
   
• Shuiwang Ji has been invited to serve on the program committee of The 16th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD 2010).
   
• Shuiwang Ji has been invited to serve on the program committee of The 35th International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2010).
   
• Our human action recognition system achieved the best performance on all three participated tasks in the TRECVID video surveillance evaluation (2009) by NIST.
   
• Version 2.0 of the sparse learning package, SLEP, was released on October 10, 2009.

Research Projects:

Spatiotemporal Genetic Regulatory Network Modeling in Drosophila Embryogenesis

Development of embryo is achieved by the concerted efforts of a large number of genes that regulate each other spatially and temporally. Therefore, accurate modeling of global genetic regulatory networks should account for the gene expression dynamics in both space and time. In this project, we made explicit efforts towards documenting both forms of dynamics by timing the developing embryo and positioning the gene expression in each embryo using computational approaches. Based on the obtained spatial and temporal information of gene expression, we developed a system to recover the genetic regulatory networks underlying Drosophila embryogenesis. Collaborators: Timothy Karr, Sudhir Kumar, Jun Liu, Jieping Ye

Body Part Keywords Annotation of Drosophila Embryos

In this project, we develop computational methods to annotate Drosophila gene expression pattern images with body part keywords, which are essential for keywords-based image comparison systems such as FlyBase. Selected publications: [1] S. Ji, Y.-X. Li, Z.-H. Zhou, S. Kumar and J. Ye: A Bag-of-words approach for Drosophila gene expression pattern annotation, BMC Bioinformatics, 10(1):119, 2009. [2] S. Ji, L. Sun, R. Jin, S. Kumar and J. Ye: Automated annotation of Drosophila gene expression patterns using a controlled vocabulary, Bioinformatics, 24(17):1881-1888, 2008.

Developmental Time Annotation of Drosophila Embryos

We obtain a set of gene expression pattern images with precise stage information from expert biologists and develop computational methods to predict the developmental stages of all currently available embryos. Meanwhile, important developmental landmarks are discovered to characterize the major developmental events during Drosophila embryogenesis. Collaborators: Timothy Karr, Charlotte Konikoff, Sudhir Kumar, Stuart Newfeld, Jieping Ye, Lei Yuan

Learning from Multiple Tasks

In this project, we develop efficient algorithms to solve multi-task formulations and apply them to the Drosophila gene expression pattern annotation problem. Selected publications: [1] S. Ji, L. Yuan, Y.-X. Li, Z.-H. Zhou, S. Kumar and J. Ye: Drosophila gene expression pattern annotation using sparse features and term-term interactions, The 15th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 407-416, 2009. [2] T. K. Pong, P. Tseng, S. Ji and J. Ye: Trace norm regularization: Reformulations, algorithms, and multi-task Learning, Submitted to SIAM Journal on Optimization, 2009. [3] S. Ji and J. Ye: An accelerated gradient method for trace norm minimization, The 26th International Conference on Machine Learning, 457-464, 2009.

Linear and Kernel Discriminant Analysis

In this project, we study the theoretical properties of generalized discriminant analysis for high-dimensional data analysis and propose novel kernel learning formulations for them. Selected publications: [1] S. Ji and J. Ye: Generalized linear discriminant analysis: A unified framework and efficient model selection, IEEE Transactions on Neural Networks, 19(10):1768-1782, 2008. [2] S. Ji and J. Ye: Kernel uncorrelated and regularized discriminant analysis: A theoretical and computational study, IEEE Transactions on Knowledge and Data Engineering, 20(10):1311-1321, 2008. [3] J. Ye, S. Ji and J. Chen: Multi-class discriminant kernel learning via convex programming, Journal of Machine Learning Research, 9:719-758, 2008.

Deep Learning for Visual Object and Action Recognition

Deep models are fully trainable systems consisting of multiple layers of abstractions from input to output. In this project, we extend traditional deep convolutional neural networks to extract spatiotemporal features and apply them to human action recognition problems. The resulting system achieved the best performance in all three participated tasks on the TRECVID video surveillance evaluation (2009) by National Institute of Standards and Technology. Selected publications: [1] Ming Yang, Shuiwang Ji, Wei Xu, Jinjun Wang, Fengjun Lv, Kai Yu, Yihong Gong, Mert Dikmen, Dennis J. Lin, and Thomas S. Huang. Detecting Human Actions in Surveillance Videos. Notebook paper, TREC Video Retrieval Evaluation Workshop, 2009.