I am a second year Ph.D. student in the Department of Computer Science,University of Colorado at Boulder. My advisor is Prof. Jordan Boyd-Graber.

My research interests lies in the intersection of natural language processing (NLP) and deep learning. Particularly, I'm interested in incorporating knowledge into deep learning methods, including the domain knowledge from linguistics and the human knowledge such as common-sense knowledge from knowledge base.

Before comming here,I received both my B.S and M.S from Xi'an Jiaotong Univesrity. I was also an intern in the system group at Microsoft Research Asia, working with Dr. Lintao Zhang on the research of deep learning algorithms.

And I've received a French Engineering Diploma (Diplôme d'Ingénieur) from Ecole Centrale de Lyon, France under a Master's double-degree program "Sino-French 4+4 program".

Here's my resume.

Research Experience

Research on combining knowledge with representation learning in NLP

Research Assistant at CU Boulder, Advisor: Prof. Jordan Boyd-Graber

  • Multi-sense word embedding with interactive learning
  • I'm currently studying how learning from human can help with learning better multi-sense word embeddings. I developed a soft-attentional skip-gram model for learning multi-sense word embedding and easy word sense disambiguation, and built pipeline for task-specific interactive learning.

    Research on Deep Learning, cooperated with Dr. Lintao Zhang at MSRA

    • Built deep learning platform using C/C++ (CPU/GPU computation supported)
    • The platform is built for the purpose of research. It supports most of the popular network structures including deep neural network and convolutional neural network, as well as various unsupervised feature learning techniques including auto-encoders and restricted boltzmann machine. It also supports some of the useful tricks such as dropout, local response normalization, newbob learning rate schedule etc.

    • Developed and implemented rotated kernels in the convolutional neural networks(CNNs)
    • The CNNs with shared weights can capture translational invariant structure in data. To extract rotational invariant local feature in images, we designed and implemented rotated kernels in CNNs.

    • Research on one-shot/zero-data learning, applied on the Chinese Handwriting Characters

    Research assistant at the Inst. of Integrated Automation, XJTU, cooperated with Prof. Deqiang Han

    • Proposed a novel extension of the k-means algorithm
    • The k-means algorithm and most of its variants use points to model the data. I proposed to use intervals to represent clusters and designed metric to measure the similarity between points and clusters. The advantage lies in three: relaxing the strong assumption of Gaussian distribution, making use of the prior knowledge that natural clusters are compact, and making use of the ordering information of the data which is less sensitive to outliers and hence make the algorithm more robust. The results are accepted by ICTAI'14.


Fenfei Guo, Deqiang Han, ChongZhao Han, k-intervals: a new extension of the k-means algorithm. In Proceedings of 26th IEEE International Conference on Tools with Artificial Intelligence (ICTAI'14), Limassol, Cyprus. November 2014. [pdf]


Univeresity of Colorado, Boulder 09/2015 -- present
  Ph.D. student in Computer Science
Xi'an Jiaotong Univesrity, China 09/2012 -- 06/2015
  M.S. in Control Theory and Control Engineering
  GPA: 91/100 (1/103)
Ecole Centrale de Lyon, France 09/2010 -- 05/2012
  Engineer Diploma (Diplôme d’Ingénieur)
  Master's double-degree program
Xi'an Jiaotong Univesrity, China 09/2008 -- 06/2010
  B.S. in Electronic Engineering and Information Science
  Honored Class name after Hsue-Shen Tsien (68/3800)
  GPA: 88.7/100; Major GAP: 89.1/100