Jianbo Ye

Ph.D. student, since 2013 fall
jxy198 [at] ist.psu.edu

Room 310, IST Building
Penn State University
University Park, PA 16802

Bio: Jianbo Ye is now a Ph.D. candidate at College of Information Science and Technology, The Pennsylvania State University. He works with Prof. James Z. Wang and Prof. Jia Li. He works on machine learning, optimization methods and computational statistics with an emphasis on their connections to real-world. His thesis has been focused on developing scalable and robust numerical algorithms that apply optimal transport theory and Wasserstein geometry to machine learning models. He received the B.Sc. degree in Mathematics from University of Science and Technology of China (USTC). He has worked as a research intern at Intel (2013) and Adobe (2017).

Publications
Google Scholar

Keywords: all selected optimization learning pattern recognition optimal transport graphics

Preprints

Oops, stay tuned!

Peer-reviewed Papers

ICML'17 A Simulated Annealing based Inexact Oracle for Wasserstein Loss Minimization
Jianbo Ye, James Z. Wang and Jia Li, International Conference on Machine Learning, Sydney, Australia, August 2017 (arXiv:1608.03859 [stat.CO], August 2016)
g-scholar

ACL'17 Determining Gains Acquired from Word Embedding Quantitatively Using Discrete Distribution Clustering
Jianbo Ye, Yanran Li, Zhaohui Wu, James Z. Wang, Wenjie Li, Jia Li, Proceedings of The Annual Meeting of the Association for Computational Linguistics, Vancouver, Canada, July 2017. Long paper.
pdf code dataset

TAFFC Probabilistic Multigraph Modeling for Improving the Quality of Crowdsourced Affective Data
Jianbo Ye, Jia Li, Michelle G. Newman, Reginald B. Adams, Jr., James Z. Wang, IEEE Transactions on Affective Computing, 2017 (arXiv:1701.01096 [stat.ML], January 2017)
pdf code dataset g-scholar

This project also develops a scalable data analytic tool, called accelerated D2-clustering, to process large scale distribution data. It could potentially leverage hundreds of CPUs with a very decent scaling efficiency (say, 70-80%).

If you are a government agency, an education institution, or a non-profit organization, we may offer you a FREE academic license of the C/MPI package to run on clusters. Please contact authors by email to discuss details. If you are commercial and would like to use our software, let us know and we will try to arrange to let you use.

TSP Fast Discrete Distribution Clustering Using Wasserstein Barycenter with Sparse Support
Jianbo Ye, Panruo Wu, James Z. Wang and Jia Li, IEEE Transactions on Signal Processing, January 2017 (arXiv:1510.00012 [stat.CO], September 2015)
pdf g-scholar code

EAAI'17 Recovering Concept Prerequisite Relations From University Course Dependencies
Chen Liang, Jianbo Ye, Zhaohui Wu, Bart Pursel, C. Lee Giles, The AAAI 7th Symposium on Educational Advances in Artifical Intelligence, San Francisco, California, 2017
pdf g-scholar dataset preprint

ECCV'16 A Distance for HMMs based on Aggregated Wasserstein Metric and State Registration
Yukun Chen, Jianbo Ye, Jia Li, European Conference on Computer Vision, Amsterdam, Netherlands, October 2016
Spotlight Presentation pdf g-scholar code


R-BiHDM(-s)[YeYY13], which has been used for SHREC14 human track, is a simple, unsupervised, data independent method. (rank among the best 3 in scanned human dataset)
IJCV Shape Retrieval of Non-Rigid 3D Human Models
D. Pickup, et al., International Journal on Computer Vision, April 2016, Springer
pdf g-scholar (benchmark paper, earlier version appeared in EG Workshop 3DOR'14)

TVC A Fast Modal Space Transform For Robust Nonrigid Shape Retrieval
Jianbo Ye, and Yizhou Yu, The Visual Computer, March 2015, Springer
pdf g-scholar (extended version of the ICMR'13 paper)

ICIP'14 Scaling Up Discrete Distribution Clustering Using ADMM
Jianbo Ye and Jia Li, IEEE International Conference on Image Processing, Paris, France, October 2014
pdf g-scholar


State-of-the-art unsupervised method to obtain a global nonrigid shape signature for shape retrieval and comparison.
ICMR'13 Fast Nonrigid 3D Retrieval Using Modal Space Transform
Jianbo Ye, Zhicheng Yan, and Yizhou Yu, ACM International Conference on Multimedia Retrieval, Dallas, April 2013 (Oral, 17.7% acceptance rate)
pdf g-scholar project page dataset slides

Working Papers


Yet another model reduction technique for deformable meshes based on approximation quality controllable subspace.
On the Approximation Theory of Linear Variational Subspace Design
Jianbo Ye and Zhixin Yan, arXiv:1506.08459 [cs.GR], June 2015
pdf g-scholar gitxiv software executable demo: linux-x86_64 video (40M)

Projects

[2015-2016] AD2-Clustering: a parallel clustering algorithm for discrete distributions, including normalized histogram as a special case, under the Wasserstein metric. software (to appear)
C MPI

[2014-2015] neuron: I re-implemented a full-fledged Scala library for composing and training neural network of complex topologies with parameter sharing, supporting different activations, metrics, regularization, and optimization methods. It also includes different variants of multilayer perceptron and auto-encoders. project page
Scala breeze

[2014-2015] dmfCramer: Discrete martrix factorizatin with Cramer risk minimization, which is yet another probabilistic matrix factorization method with loss function based on large deviations theory rather than conventional MLE/MAP framework. project page
Scala breeze

[2013] Pocket Avatar (intern project at Intel): I developed an efficient data-driven framework for real-time facial expression retargeting. demo
C++ MATLAB OpenCV

[2012-2013] iMeshDeform: A C++ mesh deformation framework based on linear variational subspace. project page video
C++ MKL PETSc OpenGL

[2011-2012] R-BiHDM: state-of-the-art, simple and fast signature for nonrigid shapes, which have been tested upon multiple benchmarks. project page
C++ CGAL OpenGL