Computer Science
A general approach for determining applicability domain of machine learning models
L. E. Schultz, Y. Wang, et al.
Discover a new, general method to determine where machine-learning predictions are trustworthy by measuring feature-space distance with kernel density estimation. This approach identifies chemically dissimilar groups, links high dissimilarity to large prediction errors and unreliable uncertainty estimates, and includes automated tools to set dissimilarity thresholds for in-domain versus out-of-domain decisions. Research conducted by Lane E. Schultz, Yiqi Wang, Ryan Jacobs, and Dane Morgan.
~3 min • Beginner • English
Related Publications
Explore these studies to deepen your understanding of the subject.
