Author | Lightnews

Fiona K. Ewald

@fionaewald.bsky.social

470 followers 150 following 8 posts

PhD Student @ LMU Munich Munich Center for Machine Learning (MCML) Research in Interpretable ML / Explainable AI

Posts Media Videos Starter Packs

Fiona K. Ewald @fionaewald.bsky.social · Jul 7

Great experience presenting my work in progress on #RashomonSets for #interpretability and #performance analysis at MCML Munich AI Day last week!

A fantastic chance to connect, learn, and share ideas - big thanks to the @munichcenterml.bsky.social for organizing.

#AI #MachineLearning #IML #xAI

Fiona K. Ewald @fionaewald.bsky.social · May 24

Thank you so much @daiichisankyo.bsky.social for welcoming me in your office in Munich!

Munich Center for Machine Learning @munichcenterml.bsky.social · May 13

Great visit to Daiichi Sankyo in Munich! Thanks to Felix Just & Roger Garriga for the warm welcome. We shared exciting work from Simon Schallmoser, Shanshan Bai & Fiona Ewald on treatment effects, synthetic datasets & feature importance. Looking forward to more collabs! #MCML

Reposted by Fiona K. Ewald

Chair of Statistical Learning and Data Science at the LMU Munich @slds-lmu.bsky.social · May 12

Feature importance measures can clarify or mislead. PFI, LOCO, and SAGE each answer a different question.
Understand how to pick the right tool and avoid spurious conclusions: mcml.ai/news/2025-03...
@fionaewald.bsky.social @ludwig-bothmann.bsky.social @giuseppe88.bsky.social @gunnark.bsky.social

mcml.ai

5 7

Fiona K. Ewald @fionaewald.bsky.social · Mar 28

Thank you for sharing! 🙏🏻

Fiona K. Ewald @fionaewald.bsky.social · Mar 28

The @munichcenterml.bsky.social recently shared our work: mcml.ai/news/2025-03... 🙏🏻

#mcml #ai #interpretability #iml #xAI #blockpost

Beyond the Black Box: Choosing the Right Feature Importance Method

The team of Bernd Bischl created a clear guide to feature importance methods, helping researchers and practitioners interpret AI models effectively.

mcml.ai

Fiona K. Ewald @fionaewald.bsky.social · Dec 1

I was thinking of using the implementation from the CPI package (github.com/bips-hb/cpi/) by Watson D. S. & Wright, M. N. (2021) "Testing conditional independence in supervised learning algorithms" ML.

GitHub - bips-hb/cpi: CPI: Conditional Predictive Impact

CPI: Conditional Predictive Impact. Contribute to bips-hb/cpi development by creating an account on GitHub.

github.com

Fiona K. Ewald @fionaewald.bsky.social · Dec 1

Need an implementation of a conditional sampler, as required, e.g., in Conditional Feature Importance, for a project in R. Since I don't think it's efficient for everyone to implement their own, I'll ask around: Do you know a good implementation?

1 1

Reposted by Fiona K. Ewald

Munich Center for Machine Learning @munichcenterml.bsky.social · Nov 26

What makes AI research excellent?
Who are the researchers behind MCML?
And what drives us at MCML?

Get to know us in our image film ⬇

MCML Imagefilm

YouTube video by MCML_Munich Center for Machine Learning

www.youtube.com

1 9

Fiona K. Ewald @fionaewald.bsky.social · Nov 26

Would be great if you could add me, too. Thank you!

Fiona K. Ewald @fionaewald.bsky.social · Nov 24

Excited to be part of this platform!
My research focuses on global model-agnostic feature importance. Please check our recent paper: "A Guide to Feature Importance Methods for Scientific Inference" (link.springer.com/chapter/10.1...).

A Guide to Feature Importance Methods for Scientific Inference

While machine learning (ML) models are increasingly used due to their high predictive power, their use in understanding the data-generating process (DGP) is limited. Understanding the DGP requires ins...

link.springer.com