Ben Höltgen

Come talk to me at EAAMO in Pittsburgh next week!

...Fairness Through Unawareness (omitting group attributes) can significantly reduce outcome inequality!
...it is often possible to reduce outcome inequality without reducing accuracy!
...Logistic Regression with group attributes is particularly prone to exacerbating inequality!

Reconsidering Fairness Through Unawareness From the Perspective of Model Multiplicity | Proceedings of the 5th ACM Conference on Equity and Access in Algorithms, Mechanisms, and Optimization

New ACM EAAMO paper out, joint work with @nuriaoliver.com at @ellisalicante.org!
dl.acm.org/doi/10.1145/...
Against common belief (but in line with the emerging multiplicity literature), we show theoretically and empirically that for algorithmic tasks like predicting unemployment...

Theory of XAI Workshop Explainable AI (XAI) is now deployed across a wide range of settings, including high-stakes domains in which misleading explanations can cause real harm. For example, explanations are required by law ...

Interested in provable guarantees and fundamental limitations of XAI? Join us at the "Theory of Explainable AI" workshop Dec 2 in Copenhagen! @ellis.eu @euripsconf.bsky.social

Speakers: @jessicahullman.bsky.social @doloresromerom.bsky.social @tpimentel.bsky.social

Call for Contributions: Oct 15

Bridging Prediction and Intervention Problems in Social Systems Many automated decision systems (ADS) are designed to solve prediction problems -- where the goal is to learn patterns from a sample of the population and apply them to individuals from the same popul...

Very impressive and comprehensive piece of work on the challenges and opportunities of using data-driven algorithms for decision-making in society. Maybe not surprising given the all-star lineup
This will be a key point of reference for many years to come
arxiv.org/abs/2507.05216

If you assume there is a true distribution from which we can draw iid, then with enough data we can approximate it, so there can only be an epsilon-RCP because we are delta-close to the assumed true distribution?

Is the intuition correct that, if we draw iid from a true distribution, then if two models disagree enough, at least one is far from the true model. So we can improve the models and with more data, both models approximate each other and in the limit the true distribution?