IDEA

Raw data is usually dirty , e.g., missing or inconsistent information, and needs substantial amounts of resources to clean. Data cleaning is a major obstacle in ML and inference on large data. In this project, we propose efficient methods to learn accurate models or infer accurate results over dirty data without cleaning.

Publications

• Learning Accurate Models on Incomplete Data with Minimal Imputation [Code+Data] [BibTex] 
  Cheng Zhen, Nischal Arya, Arash Termehchy, Prayoga, Garrett Biwer, and Sankalp Patil
  arXiv:2503.13921 [cs.LG], March 2025
• Certain and Approximately Certain Models for Statistical Learning [Slides]   [Code+Data] [BibTex] 
  Cheng Zhen, Nischal Aryal, Arash Termehchy and Amandeep Singh Chabada
  The Proceedings of the ACM on Management of Data (SIGMOD), Article 126, 2024
• When Can We Ignore Missing Data in Model Training?    [Slides][Code+Data] 
  Cheng Zhen, Amandeep Singh Chabada, Arash Termehchy
  In Proceedings of SIGMOD Workshop on Data Management for End-to-End Machine Learning (DEEM), June 2023.
• Learning Over Dirty Data Without Cleaning    [Slides][Code+Data] 
  Jose Picado, John Davis, Arash Termehchy, and Claire Lee
  The Proceedings of SIGMOD, 2020.
  Technical report with proofs
• Learning Efficiently Over Heterogenous Databases    [Poster]
  Jose Picado, Sudhanshu Pathak, and Arash Termehchy
  The Proceedings of the VLDB Endowment (Demonstration Track) , August 2018.

People

• Arash Termehchy
• Cheng Zhen
• Nischal Aryal
• Jose Picado
• Amandeep Singh Chabada
• John Davis
• Claire Lee