Due to advances in technology, data is available at rates never before seen. For example, wearable devices allow us to have data on health metrics such as heart rate, number of steps per day, and even blood O2 levels virtually at every minute of every day. However, it's not often the case that data is always accessible or complete. In fact, many real world applications of data-intensive processes require methods to treat data sets that have missing data. For example, if one wanted a user's daily step count and the user failed to wear their wearable device that day, then data would be missing from their data set. Standard treatments of missing data ask that a practitioner impute some value to replace the missing data or discard it all together. In addition, oftentimes there are assumptions that the data is missing completely at random. In this poster, we present special structures of missing data and how one can accomplish solving a linear system with missing data without any pre-treatment of the data itself.
Approaches for Incomplete Large-Scale Data
Anna Ma, UC Irvine
2022 AWM Research Symposium