Wanyu Du

I am a graduate student in Computer Science at University of Virginia. I am currently working under the supervision of Professor Yangfeng Ji in the UVA NLP group.

My general research interest is Deep Learning for Natural Language Processing. In particular, I am interested in deep latent-variable models for probabilistic text processing and understanding, efficient learning algorithms for NLP models, and controllable text generation.

I am going to pursue a PhD degree, and do advanced research in Machine Learning and Natural Language Processing. If you are interested in my works, please contact me by mail: wd5jq@virginia.edu

Publications

An Empirical Comparison on Imitation Learning and Reinforcement Learning for Paraphrase Generation   [code | slide]
Wanyu Du and Yangfeng Ji
Empirical Methods in Natural Language Processing. EMNLP 2019.

An Effective Optimization Algorithm for Application Mapping in Network-on-Chip Designs
Xinyu Wang, Tsan-Ming Choi, Xiaohang Yue, Mengji Zhang, Wanyu Du
IEEE Transactions on Industrial Electronics. 2019.

Research Projects

Improving Variational Autoencoders for Text Modeling

  • Analyzed the local geometry of latent representations of text data in variational autoencoders.
  • Proposed a non-isotropic Gaussian prior that can automatically learn low-dimensional and multi-modal latent representations from real-world text data.

Efficient Learning Algorithms for Paraphrase Generation

  • Developed a unified framework for different learning algorithms (e.g. REINFORCE, DAgger, MLE) in a sequence-to-sequence model.
  • Proposed some variant learning algorithms, and conducted empirical comparisions on the paraphrase generation task.

Named Entity Recognition for Medical Records

  • Implemented a named entity recognition model (BLSTM+CRF) to extract the medical entities from original unstructured medical records, and achieved an overall F1 score of 0.91.
  • Adopted a semantic matching model (CNN+LSTM+NN) to discriminate similar diagnosis records, and achieved an accuracy of 0.87.

Heuristic Algorithm Optimization

  • Implemented the Tabu Search algorithm for Network-on-Chip Design under dynamic bandwidth (NP-hard Problem and Min-Max Problem).
  • Employed the Discrete Particle Swarm Optimization algorithm for Network-on-Chip Design under static bandwidth (NP-hard Problem).
  • Honors

    Professional Activities