Date: 13 March 2023
Time: 16:00 (CET)
Title: Responsible AI Mitigations and Tracker: New open-source tools for guiding mitigations in Responsible AI
Abstract: Responsible AI Toolbox is an open-source effort at Microsoft for accelerating and operationalizing Responsible AI via a set of interoperable tools, libraries, and customizable dashboards. Recently, we released two new tools as part of the toolbox:
Responsible AI Mitigations: Python library for implementing and exploring mitigations for Responsible AI.
Responsible AI Tracker: JupyterLab extension for tracking, managing, and comparing Responsible AI mitigations and experiments.
Both tools contribute to supporting a systematic and targeted process of model improvement by identifying, diagnosing, mitigating, and comparing failure modes. During the presentation, we will delve into the new tooling additions and will illustrate their functionalities through a hands-on demonstration. In particular, we will show how insights extracted via the Responsible AI Dashboard through the identification and diagnosis stage can then be used to ideate and implement different mitigation techniques via the Responsible AI Mitigations library. Finally, we will also demonstrate how different mitigation techniques can be compared and validated through Responsible AI Tracker directly on Jupyter Lab.
Speaker Biography: Besmira Nushi is a Principal Researcher in the Adaptive Systems and Interaction group at Microsoft Research. Her interests lie at the intersection of human and machine intelligence focusing on Reliable Machine Learning and Human-AI Collaboration. She has made practical and scientific contributions on implementing and deploying Responsible AI tools for debugging and troubleshooting ML systems. Currently, Besmira is leading the research behind several Responsible AI tools for debugging ML systems, such as Responsible AI Toolbox, Error Analysis, and BackwardCompatibilityML.