Skip to main content
Engineering, Data / ML

Enabling Deep Model Explainability with Integrated Gradients at Uber

October 23 / Global
Featured image for Enabling Deep Model Explainability with Integrated Gradients at Uber
Image
Figure 1: Examples of integrated gradients local feature attributions, plotted using SHAP waterfall plots.
Image
Figure 2: The definition, the discrete approximation, and a visualization of the discrete approximation for integrated gradients in order from top to bottom.
Image
Figure 3: High-level explainability architecture in Michelangelo.
Image
Figure 4: Gradient computation for categorical features for integrated gradients.
Image
Figure 5: Flexible IG framework hooked into the evaluator step (which evaluates a trained model) common to both train and evaluation pipelines.
Image
Figure 6: Parallel IG execution with Ray®.
Hugh Chen

Hugh Chen

Hugh Chen is a Software Engineer on Uber’s AI platform team. He’s currently the leading engineer on explainability.

Eric Wang

Eric Wang

Eric Wang is a Senior Staff Software Engineer working on Uber’s Machine Learning Platform, focusing on developing scalable solutions to enhance the entire machine learning lifecycle.

Gaoyuan Huang

Gaoyuan Huang

Gaoyuan Huang is a former Software Engineer on Uber’s AI platform team in the San Francisco Bay Area. He specializes in the end-to-end machine learning model development lifecycle, ensuring AI solutions move seamlessly from research to production.

Howard Yu

Howard Yu

Howard Yu was a Senior Software Engineer on the Uber AI Platform that worked on various ML quality initiatives. He’s now working on Uber’s EngSec AI team focusing on using ML to improve security and privacy in different areas.

Jia Li

Jia Li

Jia Li is a Senior Staff TLM who leads the ML Analytics team for Michelangelo.

Sally Lee

Sally Lee

Sally is a Senior Staff Tech Lead Manager who leads Uber’s Core ML platform, Michelangelo. With over 15 years of experience, she’s passionate about applying machine learning solutions to large-scale business problems.

Posted by Hugh Chen, Eric Wang, Gaoyuan Huang, Howard Yu, Jia Li, Sally Lee