Research: My two main research interests are:

• Flow (ODE)-based generative neural networks.
• Conformal prediction for uncertainty quantification on data with dependency.

See Google Scholar for a complete list of works.

My email is cxu310@gatech.edu

Education:

• Current: 4th-year Operations Research (OR) Ph.D. student at Georgia Tech School of Industrial and Systems Engineering (ISyE).
  • Advisor: Professor Yao Xie
• Former: joint BS/MS degree during 2016-2020 from the University of Chicago.
  • MS degree: Statistics.
    Advisor: Professor Rina Foygel Barber
    Thesis: Efficient predictive inference with Jackknife+ under ensemble learning link 
  • BS degree: double majors in (1) Computational and Applied Mathematics (CAM) (2) Economics (specialized in data science).

Updates:

• Oct 2023-Jan 2024:
  1) Excited to announce that I will be joining Toyota Research Institute, Robotics as a Research Scientist Intern for Summer 2024.
  2) Flow-based Distributionally Robust Optimization is accepted by IEEE Journal on Selected Areas in Information Theory, 2024.
  3) Normalizing flow neural networks by JKO scheme won the best poster award at Data Science Week 2023, organized by Purdue University Fort Wayne
• June-Sep 2023:
  1) Conformal prediction for time-series is officially published at TPAMI, 2023. Vol. 45, No. 10. pp. 11575 – 11587.
  2) Normalizing flow neural networks by JKO scheme is accepted as a spotlight paper at Thirty-seventh Conference on Neural Information Processing Systems (NeurIPS 2023).
  3) Computing high-dimensional optimal transport by flow neural networks is available as a preprint.
• March-May 2023:
  1) Our paper Conformal prediction for time-series has been accepted at IEEE Transactions on Pattern Analysis and Machine Intelligence (IF: 24.3).
  2) Our paper Sequential Predictive Conformal Inference for Time Series has been accepted by ICML 2023.
  3) Our paper Spatio-Temporal Wildfire Prediction using Multi-Modal Data has been accepted at IEEE Journal on Selected Areas in Information Theory.
  4) Our paper Generalized generalized linear models: Convex estimation and online bounds is available as a preprint
  5) Our paper Optimal transport flow and infinitesimal density ratio estimation is available as a preprint.
• Jan-Feb 2023:
  1) Our paper Conformal prediction for dynamic time-series has been further revised. It is also implemented in the AWS package Fortune.
  2) Our paper Sequential predictive conformal inference for time series is revised
  3) Our paper Spatio-temporal wildfire prediction using multi-modal data is revised.
  4) Talk slides for invertible neural networks as generative models are available, based on works IGNN and JKO-iFlow.
• Oct-Dec 2022:
  1) Our paper Invertible Neural Networks for Graph Prediction has been accepted by IEEE Journal on Selected Areas in Information Theory.
  2) Our paper Spatio-temporal wildfire prediction using multi-modal data is available and currently under review
  3) Our paper Sequential predictive conformal inference for time series is available.
  4) Our paper Invertible normalizing flow neural networks by JKO scheme is available.
• Aug-Sep 2022:
  1) Our paper Invertible Normalizing Flow Neural Networks by JKO scheme is currently under review
  2) An alternative approach to train neural networks using monotone variational inequality will be presented at DeepMath 2022 conference. One-page abstract available.
• July 2022:
  1) Our paper Conformal prediction for time-series is updated to include non-stationary & heteroskedastic time-series and the new Sequential Conformal Predictive Inference (SPCI) idea.
  Glad to soon be supported by Meta to develop conformal change-point detection algorithms for the Kats package.
  2) Shorter versions of iGNN, solar ramping event modeling, and wildfire modeling are available.
• June 2022:
  1) Glad to announce that the scitkit-learn compatible module MAPIE has implemented our work Conformal prediction for dynamic time-series.
  2) Our work Conformal prediction set for time-series is strongly accepted by the Workshop on Distribution-Free Uncertainty Quantification at ICML 2022.
  3) Updated Solar radiation ramping events modeling using spatio-temporal point processes, currently under review.
  
• March-May 2022:
  1) Completed Invertible Neural Networks for Graph Prediction.
  2) Revised An alternative approach to train neural networks using monotone variational inequality.
• February 2022:
  1) Perform wildfire modeling using the “generalized” generalized linear model (GGLM), a collaboration with Prof. Yao Xie and Prof. Arkadi Nemirovski.
  2) Start the collaboration with the Centers for Disease Control and Prevention (CDC) on modeling HIV patient trajectory using latent-class GGLM.
• January 2022: Completed the work Training neural networks using monotone variational inequality with prediction guarantee.
• November 2021: (Postponed due to personal reasons) To work as a full-time intern for Alibaba DAMO Academy–the Decision Intelligence Lab in Summer 2022, under the supervision of Prof. Wotao Yin and Dr. Qingsong Wen.
• October 2021: Conformal Prediction for Dynamic Time-Series significantly revises and extends the ICML 2021 work and is now available.
• September 2021: Passed the Ph.D. Comprehensive Exam in Operations Research.
• August 2021: To work as a full-time intern for Healthcare Xplorers @ Roche Diagnostics on “Developing machine learning-based methods for early anomaly detection in diagnostics instrument and assay data” in Spring 2022.
• July 2021: To work as a full-time intern in Fall 2021 for the Energy Systems @ Argonne National Laboratory and I am supervised by Dr. Feng Qiu. The ongoing work is titled Wildfire modeling with Hawkes point process and Conformal uncertainty quantification.
• June 2021: Conformal Anomaly Detection on Spatio-Temporal Observations with Missing Data is accepted by ICML 2021 workshop on Distribution-free Uncertainty Quantification.
• May 2021: Conformal Prediction Interval for Dynamic Time-Series is accepted by ICML 2021 as a long talk/oral presentation (3% of all submissions).
• May 2021: Become a TA for Prof. Yao Xie’s OMSA ISyE 6740: Computational Data Analysis during Summer 2021, a highly well-regarded machine learning course by past students.
• April 2021: Invited to be a presenter in the INFORMS 2021 Annual Meeting, General Session entitled Adaptive online learning of high-dimensional data in October 2021. I will talk about Online Prediction For High-dimensional Discrete Event Data (Slides available).
• December 2020: Completed the work Solar radiation anomaly events modeling using spatio-temporal mutually interactive processes.
• September 2020: Predictive inference is free with the jackknife+-after-bootstrap is accepted by NeurIPS 2020 as a poster presentation.
• June 2020: Graduated from the University of Chicago as a joint BS/MS student.