Zirui Tao

1415 Engineering Dr · Madison, WI 53715 ztao23[at]wisc[dot]edu
Curriculum Vitae [pdf]

I am an ECE master student at University of Wisconsin - Madison.

Education

University of Wisconsin Madison

Master of Science
Electrical Engineering
Relevent Course: Advanced Computer Architecure, Statistical Learning Theory, Convex Optimization, Probability Theory, Linear Algebra, Advanced Algotirhms, Mathematical foundation of Machine Learning, Signal Processing

GPA: 3.842/4.00

August 2018 - Dec 2019

University of Wisconsin Madison

Bachelor of Science
Computer Engineering, Computer Science
Relevent Course: Computer Architecture, Database, Algorithm, Data Structures, Embeded System Programing, Machine Organization and Programming, Mobile Laboratory, Electronic Circuits Design

GPA: 3.61/4.00

August 2014 - May 2018

Research

Research Assistant - Biostatistics, Medical Imaging

Medical Science Center, Madison, WI

Conducted research projects regarding medical imaging and computer vision with Professor Vikas Singh.

Published a paper abstract and coauthored one paper abstract on Alzheimer's Association International Conference (AAIC).

Coauthored one paper to Computer Vision and Pattern Recognition [CVPR] 2018 and one paper to International Conference on Computer Vision [ICCV] 2019.

March 2017 - May 2019

Research Assistant - Head Neck Cancer

Wisconsin Institute of Medical Research Carbone Cancer Center, Madison, WI

Compared a variety of inference models including AdaboostM1, ctree, random forest and associated parameter settings and performed hyperparameter search on each model, for the prediction of Head and Neck patients' overall survivals from imaging and genomic data.

Selected the ideal model that produced the highest ROC AUC results.

Fine-tuned model's parameters using cross-validation and achieved under the precision-recall rate at 75 under the limited, highly-skewed medical data.

April 2017 - February 2018

Lab Assistant

Parks Laboratory, Madison, WI

Designed the Epistasis Analysis pipeline for million-scale gene interdependence hypothesis testing for the Center for High Throughput Computing at UW-Madison and reduced the computational time by a factor of 100.

Led the complete lifecycle of the analytical application of query analysis and rich interactive graphics for the visualization of large structured biological datasets.

Tested pipelines on the computational cluster and populated the generated results into SQL Server using ODBC with Python and R.

September 2016 - September 2017

Publication

Seong Jae Hwang, Zirui Tao, Won Hwa Kim, Vikas Singh, "Conditional Recurrent Flow: Conditional Generation of Longitudinal Samples with Applications to Neuromaging", International Conference on Computer Vision (ICCV), 2019. [Acceptance rate: 25%][pdf]

Seong Jae Hwang, Sathya N. Ravi, Zirui Tao, Hyunwoo J. Kim, Maxwell D. Collins, Vikas Singh, "Tensorize, Factorize and Regularize: Robust Visual Relationship Learning", Computer Vision and Pattern Recognition (CVPR), 2018. (Acceptance rate: 29.7%)[pdf][code][poster]

Workshop

The 1st Workshop on Statistical Deep Learning in Computer Vision 2019 (Oral)

Presentation

Zirui Tao, Ronak R. Mehta, Seong Jae Hwang, Rebecca L. Koscik, Erin Jonaitis, Sterling C. Johnson, Vikas Singh, "A Normative Modeling Based Analysis of Amyloid, Cognition, and Tau in Preclinical Alzheimer's Disease", Alzheimer's Association International Conference (AAIC), 2019. [poster]

Seong Jae Hwang, Rebecca L. Koscik, Tobey J. Betthauser, Zirui Tao, Won Hwa Kim, Sterling C. Johnson, Vikas Singh, "Predicting amyloid accumulation trajectories in a risk-enriched Alzheimer's disease cohort with Deep Conditional Neural Networks", Alzheimer's Association International Conference (AAIC), 2019.

Industry

Summer Software Engineering Intern (Manager: Benoit Steiner)

System ML Team, Facebook AI Research, Menlo Park, CA

Improved the cost model accuracy by 50 percent, used the paper "Learning to Optimize Halide with Tree Search and Random Programs", by Adams et al. , that largely boost the tree search efficiency on auto-generated Halide schedulings and pipelines

Designed the ONNX model generation pipelines and co-designed Caffe2 to ONNX conversion pipelines

Augmented the Halide pipeline scheduling vs actual runtime data by expanding ONNX to Halide conversion from opset to object detection architectures

May 2019 - August 2019

Projects

Skills

Programming Languages & Tools
Applications & Frameworks

Services, Awards

Services
Awards

Miscellaneous

Interests

When I am not at researching & programming, I enjoy working out and chilling out.