scikit-learn-inria-fondation/RemoteInterpretabilityFebruary2021

The remote Interpretability workshop and sprint (February 2021)

Interpretability Workshop and Sprint

The event will follow the agenda below (in Central Europe Timezone), note that the timeslots include presentation and discussion:

February 8th:

• 2PM - 5PM : Interpretability workshop (1st session)

  • 2PM - 2.15PM "Welcome and Introduction" (Gaël Varoquaux, Inria, scikit-learn Consortium)

  • 2.15PM - 3.15PM "Model inspection in scikit-learn: current status, limitations, and ongoing work" (Guillaume Lemaitre, Inria, scikit-learn Consortium)

  • 3.15PM - 3.30PM Break

  • 3.30PM - 4.30PM "FACET: a new open-source library for global model explanations" (Jan Ittner, BCG)

    FACET introduces a new algorithm to quantify dependencies and interactions between features in ML models. It is closely integrated with scikit-learn and adds a new, global perspective to the observation-level explanations provided by the popular SHAP approach.

  • 4.30PM - 5PM "eXplainable AI Interfaces : a user perspective" (Clara Bove, AXA)

    Machine Learning has provided new business opportunities in the insurance industry, but its adoption is for now limited by the difficulty to explain the rationale behind the prediction provided. Yet, limited work has been done on explanations for non-expert users (in AI, statistics, applied domain). In our research, we focus on how we can build and present understandable and useful explanations from various interpretability methods for non-expert users. More recently, we explore how to enhance local feature importance explanations and propose design principles to contextualize these explanations with additional information about the Machine Learning system, the domain and external factors that may influence the prediction.

February 9th:

• 9.30AM - 12.30PM : Interpretability workshop (2nd session)

  • 9.30AM - 10.30AM "ML Interpretability vs. the 'Real Word'" (Xavier Renard, Thibault Laugiel, AXA)

  • 10.30AM - 11AM "Explainability of ML in the financial sector" (Laurent Dupont, ACPR - Banque de France)

    Outline:

    • 1. Use cases observed among financial actors
    • 2. Current and future use cases for the supervision authority
    • 3. Proposed development and governance principles for ML in finance

  • 11AM - 11.15AM Break

  • 11.15AM - 12.15PM "A Causal Perspective On Interpretability Methods" (Léo Dreyfus-Schmidt, Samuel Ronsin, Dataiku)

  • 12.15PM Summary and Conclusions

Sprint

Resources to prepare the sprint are available at this page. Addressable issues are listed here.

February 9th

• 2PM - 5PM : Development Sprint on Interpretability issues

February 10th:

Development Sprint for Members with broader scope than Interpretability