Responsible AI Development
What is RAI?
Rai is an open-source project that helps AI developers with various aspects of responsible AI development. It consists of a core API and a corresponding web-based dashboard application. RAI can easily be integrated into AI development projects and measures various metrics for an AI project during each phase of AI development, from data quality assessment to model selection based on performance, fairness, and robustness criteria. In addition, it provides interactive tools and visualizations to understand and explain AI models and provides a generic framework to perform various types of analysis including adversarial robustness.
RAI can handle many types of data such as text, images, and tabular data. In addition to providing a comprehensive set of quantitative metrics that can be used to study the performance and bias and other aspects of models, RAI provides a flexible interactive design where new types of analysis and explanations for data and model can be easily integrated as custom Analysis. The rich set of out of the box visualization such as grad-cam can assist developers with the analysis and debugging of ai models. Here is a summary of the main features of RAI:
- Automated measurement of more than 150 different metrics, with additional information regarding each metric’s definition, range, and reference
- Automated testing of AI models by providing an easy to use, logic-based test cases for certifying various qualities of AI model
- An easily extensible Model analysis tool to provide custom visualization and explainability to AI models
This document is a good starting point.
We welcome feedback, questions, and issue reports.
We welcome contributions. We use GitHub issues to track public bugs. Report a bug by opening a new issue.