My name is Zi WANG, and I am a Ph.D. student at Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology. I work with the Learning and Intelligent Systems Group in the Computer Science and Artificial Intelligence Laboratory. I am advised by Prof. Stefanie Jegelka, Prof. Leslie Pack Kaelbling, and Prof. Tomás Lozano-Pérez.

More about me


Research Interest

Machine learning theory, algorithms and applications in related areas such as robotics, computer vision, data mining, speech, and visualization.


Massachusetts Institute of Technology

2014 - present

Ph.D. student in Electrical Engineering and Computer Science
Advisors: Prof. Stefanie Jegelka, Prof. Leslie Kaelbling, and Prof. Tomás Lozano-Pérez

Massachusetts Institute of Technology

2014 - 2016

S.M. in Electrical Engineering and Computer Science
Master thesis: Optimization as Estimation with Gaussian Process Bandits
Thesis Supervisors: Prof. Stefanie Jegelka and Prof. Leslie Kaelbling

Tsinghua University

2010 - 2014

B.Eng. in Computer Science and Technology
Undergraduate thesis: Fast Dropout Training for Deep Neural Networks (in Chinese)
Thesis advisors: Prof. Fei Sha and Prof. Jun Zhu


My academia adventure

  • Sep. 2014 - Present

    Research Assistant @ CSAIL, MIT

    - Studied Gaussian process optimization, contextual and continuous bandit problems, etc.
    - Proposed and implemented a MAP algorithm for GP optimization with applications to robotics/vision.
    - Derived the theoretical bounds for the new method, and analyzed relations to other methods.

  • Feb. -- May. 2014

    Research Assistant @ Viterbi, USC

    - Researched auto-encoders, deep neural networks and dropout training.
    - Derived and implemented fast training algorithm with regularizer via noise marginalization.
    - Analyzed and compared performance of training with regularizer and training with dropout noise for deep neural networks both on CPU and GPU.

  • Mar. 2013 -- Jul. 2014

    Research Assistant @ State Key Lab of Intelligent Tech. & Sys., Tsinghua

    - Studied topic modeling, variational inference, Gibbs sampling and data augmentation.
    - Implemented the prototype for scalable inference algorithm for correlated and dynamic topic models.
    - Researched different evaluation methods, and compared perplexity results for variational LDA, Gibbs LDA, variational CTM, and our partially collapsed Gibbs sampling algorithm for CTM.
    - Created hierarchical visualizations for 1000 topics learned from New York Times with graphviz, C#, D3.js.

  • Jul. -- Aug. 2013

    Undergraduate Researcher @ Viterbi, USC

    - Researched non-negative matrix factorization (NMF) with sparse coding for speech separation.
    - Derived the algorithm and implemented the prototype for Discriminative NMF.
    - Our method yields a 11.5% improvement on signal-to-noise ratio over traditional sparse NMF.

  • Dec. 2012 -- May 2013

    Research Assistant @ Future Internet Technology Sub-interest Group, Tsinghua

    - Competed in Baidu's Movie Recommendation Algorithm Contest. Ranked top 10.



Focused Model-Learning and Planning for Non-Gaussian Continuous State-Action Systems

Zi Wang, Stefanie Jegelka, Leslie Pack Kaelbling, Tomás Lozano-Pérez

arXiv preprint arXiv:1607.07762

Abstract PDF arXiv Video BibTex

Optimization as Estimation with Gaussian Processes in Bandit Settings

Zi Wang, Bolei Zhou, and Stefanie Jegelka

International Conference on Artificial Intelligence and Statistics (AISTATS), 2016

Oral presentation (6% acceptance rate)

Abstract PDF arXiv Code Project page BibTex


Fast Learning with Noise in Deep Neural Nets

Zhiyun Lu*, Zi Wang*, and Fei Sha

(Spotlight Presentation) NIPS Workshop: Perturbations, Optimization, and Statistics, Quebec, Canada, 2014

Abstract PDF BibTex

Discriminative Non-Negative Matrix Factorization for Single-Channel Speech Separation

Zi Wang and Fei Sha

IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Florence, Italy, 2014

Abstract PDF Project page IEEE Poster BibTex


Scalable Inference for Logistic-Normal Topic Models

Jianfei Chen, Jun Zhu, Zi Wang, Xun Zheng, and Bo Zhang

Advances in Neural Information Processing Systems (NIPS), Lake Tahoe, CA, 2013

Abstract PDF Project page NIPS Poster Code BibTex

Contact Me

ziw 'at' mit.edu