Data Efficient Active Machine Learning
Abstract
Data-efficient machine learning (DEML) is critical to AF/DoD operations for the following reasons. First, training machine learning algorithms generally requires a large and completely labeled training dataset. Human labeling of raw data is an expensive and time-consuming process, especially with a limited pool of expert analysts. Therefore, machine learning algorithms must produce accurate predictive models from limited labeled training data. Moreover, mission environments and objectives can be varied and rapidly changing, and so machine learning models must be quickly adaptable to the situation at hand. The quality of the raw data available to a machine learning system (and to human analysts) is also often unpredictable. It may often happen that not all of the desired features for making predictions and decisions are available. Therefore, machine learning algorithms must be robust to missing or partially unobserved data.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 01, 2022
- Accession Number
- AD1158854
Entities
People
- Robert D. Nowak
Organizations
- University of Wisconsin–Madison