Interpreting Gaussian Process Models with Integrated Gradients
Author Information
Author(s): Zhang Fan, Ono Naoaki, Kanaya Shigehiko
Primary Institution: Nara Institute of Science and Technology
Hypothesis
Can Integrated Gradients effectively interpret the predictions of Gaussian Process Regression models?
Conclusion
The proposed method using Integrated Gradients allows for a better understanding of the predictive mean and standard deviation in Gaussian Process models.
Supporting Evidence
- The method provides insights into the reliability of predictions by quantifying uncertainty associated with each feature.
- Models trained with the proposed method achieved R2 values higher than 0.7, confirming their predictive accuracy.
- The study utilized a dataset of 1100 organic compounds to validate the interpretability of the models.
Takeaway
This study shows how to understand predictions made by a complex model called Gaussian Process Regression, helping us see which parts of the data are most important.
Methodology
The study combines Gaussian Process Regression with Integrated Gradients to interpret model predictions by assessing the contribution of each feature.
Limitations
The method may struggle with high-dimensional data and requires careful selection of features to ensure interpretability.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website