Avatar

Zhimeng Luo

PhD Student

University of Pittsburgh

Biography

I am a PhD student of Information Science at University of Pittsburgh. My research interests include Natural Language Processing, Machine Learning, Knowledge Representation, and their applications in Health Informatics. My advisor is Daqing He, the director of Information Retrieval, Integration and Synthesis (iRiS) lab.

Interests

  • Natural Language Processing
  • Knowledge Representation
  • Machine Learning in Health Informatics

Education

  • PhD in Information Science, Est. 2024

    University of Pittsburgh

  • MS in Information Science, 2019

    University of Pittsburgh

  • BS in Statistics, 2017

    Huazhong University of Science and Technology

Recent Publications

Click here to discover more publications.
Lei Zhang, Zhimeng Luo, Ruimei Chai, Dooman Arefan, Jules Sumkin, Shandong Wu (2019). Deep-learning method for tumor segmentation in breast DCE-MRI. SPIE: Medical Imaging 2019.

DOI Link

Lei Zhang, Ruimei Chai, Aly A Mohamed, Bingjie Zheng, Zhimeng Luo, Shandong Wu (2019). Automated deep-learning method for whole-breast segmentation in diffusion-weighted breast MRI. SPIE: Medical Imaging 2019.

DOI Link

Recent Posts

Week 2 talk summary

Duolingo: Improving Language Learning and Assessment with AI
Last updated on Sep 17, 2019 2 min read talk summary

Experience

 
 
 
 
 
Sep 2018 – Present
Pittsburgh, PA, USA

Research Assistant

iRiS Lab, University of Pittsburgh

  • Researched on clinical abbreviation disambiguation for large scale clinical notes (10-100 millions of clinical notes from UPMC)
  • For annotation part: proposed a scalable clustering algorithm (based on DBSCAN) on abbreviation instances (word embedding of context words) for annotation
  • For model part: Modified fastText model to abbreviation disambiguation as a classification problem (one model for all abbreviations).
 
 
 
 
 
Aug 2018 – Feb 2019
Pittsburgh, PA, USA

Research Assistant

ICCI Lab, University of Pittsburgh

  • Applied 3D U-net on MRI breast tumor segmentation

  • Modified the original 3D U-net from several aspects:

    • Use residual connections within each convolutional module; Use Group Normalization to replace Batch Normalization, which shows better performance when batch size is small
    • Never downsample feature maps along the slice dimension, in order to keep the information along the that dimension
    • Due to the limitation of training dataset size, add an additional Variational Autoencoder (VAE) branch to regularize the encoder

  • two papers related to apply U-net and 3D U-net on MRI have been accepted by Medical Imaging 2019, SPIE. My contribution is applying 3D U-net on breast tumor MRI and whole breast MRI, and comparing with 2D method

 
 
 
 
 
Feb 2017 – Jun 2017
Wuhan, Hubei, China

Research Assistant

Huazhong University of Science and Technology

  • Applied state-of-the-art Convolutional Neural Networks (CNN) in Hyperspectral Image (HSI) Classification and implemented them by Keras and TensorFlow
  • Designed a HSI classification method based on Fully Convolutional Networks (FCN) and U-Net
  • Proposed a training method according to the absence of large number of labeled samples, resulting in faster training speed, and making full use of the global information of HSI for feature extraction
  • This undergraduate thesis won the outstanding undergraduate thesis award (1st/91) in School of Mathematics and Statistics of HUST (2017).

Contact

  • 135 N Bellefield Ave, Pittsburgh, PA, 15213, United States