Excited to announce our workshop on Causality in Science at #NeurIPS2025!. See you in San Diego 🌴🇺🇸.

RT @FannyYangETH: Register now (first-come first-served) for the "Math of Trustworthy ML workshop" at #LagoMaggiore, Switzerland, Oct 12-16….

RAMEN:

@JavierAbadM @yixinwang_ @FannyYangETH @riccardocadeii @ilkerdemirel_ @FrancescoLocat8 Causal Lifting:

Landed in Singapore for #ICLR—excited to see old & new friends! I’ll be presenting:. 📌 RAMEN @ Main Conference on Saturday 10 am (@JavierAbadM @yixinwang_ @FannyYangETH).📌 Causal Lifting @ XAI4Science Workshop on Sunday (@riccardocadeii @ilkerdemirel_ @FrancescoLocat8 ).

RT @riccardocadeii: Foundational models’ predictions (🦙♊️🦖) can propagate biases in causal downstream tasks, posing a significant risk in A….

This is joint work with amazing collaborators: @JavierAbadM @Guanbo17 @DonhauserKonst @RayDuch @FannyYangETH and Issa Dahabreh. Code:

We also find that model scale matters for efficiency gains. Larger models yield more accurate outcome predictions, leading to smaller variance for H-AIPW (and thus smaller confidence intervals).

We evaluate our estimator across several randomized experiments. Results show that H-AIPW consistently achieves the same statistical precision (i.e. confidence interval width) as the standard AIPW estimator, but with significantly smaller sample sizes.

Key advantages of H-AIPW:.(1) Valid statistical inference – even if the foundation models are biased! .(2) Smaller confidence intervals. (3) No extra assumptions beyond those needed for valid inference in standard randomized experiments.

Looking for a more efficient way to estimate treatment effects in your randomized experiment?. We introduce H-AIPW: a novel estimator that combines predictions from multiple foundation models with real experimental data.

RT @CAUSALab: Sweater weather is better with #causal methodology!. @CAUSALab kicked off the Fall semester with a welcome back social. The e….

RT @patrickc: Mario Draghi's new report on EU competitiveness doesn't mince words. "Across different metrics, a wide gap in GDP has opened….

RT @AmartyaSanyal: The call for this position is now public. This will be jointly supervised with Prof. Amir Yehudayoff at @DIKU_Institut….

RT @SimonsFdn: It is with great sadness that the Simons Foundation announces the death of its co-founder and chair emeritus, James Harris S….

Come to our AISTATS poster (#96) this afternoon (5-7pm) to learn more about hidden confounding!.

RT @alexmeterez: working with Antonio is the most fun I've ever had while also doing amazing research, go apply!.

We validate our approach on various datasets, ranging from synthetic to real-world data (WHI). Code is available here: (7/7).

We propose an alternative strategy: quantify the true confounding strength and discard the study only if the strength is large enough to affect conclusions. Our strategy (1) does not require parametric assumptions and (2) does not discard studies with small confounding. (6/7)

Recent works have proposed 2 strategies:. #1: Combine estimators from the two studies -> requires assumptions to extrapolate. #2: Test for no unobserved confounding -> can discard "good" studies since some degree of confounding is likely present in real-world data. (5/7).