Sílvia Casacuberta

CS PhD student & @KnightHennessy scholar @Stanford. Prev.: @UniofOxford with @rhodes_trust, @Harvard '23, @MPI_IS, @CSatETH, @IBM. Theory CS for Trustworthy AI

[Nari Johnson @ FAccT @narijohnson]

🔬Science can play a critical role in guiding policy interventions. How can we make our research legible to the policymakers who work on AI?

Our tutorial "From FAccT to Policy Impact" explores how we can translate our research in a specific format: the policy memo! #FAccT2026

[3CatInfo @3CatInfo]

S'il·lumina per primera vegada la torre de Jesús de la Sagrada Família https://t.co/LEzdnExy8B

[Sílvia Casacuberta @SiCaPu]

If you would like to know more, here is our recorded talk for FOCS: https://t.co/K8XmzMCjUY.

For an expository presentation of our results, see: https://t.co/9pnyM9bE5j.

[Sílvia Casacuberta @SiCaPu]

In a previous work with Cynthia Dwork and Salil Vadhan (https://t.co/JNZfLu6fLW, presented at STOC 2024), we showed that from multicalibration we get a stronger and more general Hardcore Lemma, from which we derive the original Hardcore Lemma with optimal density 2*delta.

[Sílvia Casacuberta @SiCaPu]

...other multigroup fairness notions) has recently been shown to be powerful in other settings in TCS, including in omniprediction (https://t.co/g3LWwOI5dp), in pseudoentropy characterizations (https://t.co/B4LZyZMhbA), and in robust learning (https://t.co/OJ904AfmP2).

[Sílvia Casacuberta @SiCaPu]

A similar picture emerges in the setting of complexity theory. We know, from the work of Trevisan, Tulsiani, and Vadhan, that from the Regularity Lemma (= multiaccuracy theorem) we can prove Impagliazzo’s Hardcore Lemma, albeit with suboptimal density delta.

[Sílvia Casacuberta @SiCaPu]

That is, in certain cases there is no way to post-process a multiaccurate predictor to get a weak learner, even assuming the best hypothesis in C has high correlation. However, when we also require p to be calibrated, we recover not just weak, but strong agnostic learning.

[Sílvia Casacuberta @SiCaPu]

We show that for many classes C we can construct a predictor p that is perfectly C-multiaccurate, yet p is completely uncorrelated with the labels (even though some concepts in C do have high correlation with the labels).

[Sílvia Casacuberta @SiCaPu]

We know that we can construct multiaccurate and multicalibrated predictors from weak agnostic learning, and that a C-multicalibrated predictor is a strong agnostic learner for C. But what learning guarantees does multiaccuracy (= computational indistinguishability) offer?

[Sílvia Casacuberta @SiCaPu]

It was great to attend FORC at Harvard this past week!

I gave a talk on our paper “How Global Calibration Strengthens Multiaccuracy” (joint work with Parikshit Gopalan, Varun Kanade, and Omer Reingold), which we presented at FOCS this past December.

https://t.co/lMtsYB2XRl

[Ellen Vitercik @vitercik]

Can machine learning improve discrete optimization algorithms without sacrificing theoretical guarantees?

This was the central question of the talk I gave this spring at a few schools (UCSD, UIC, Yale, Penn): Machine Learning for Discrete Optimization: Theoretical Foundations.

[Aaron Roth @Aaroth]

How many samples do you need from an unknown distribution in order to train a model with multicalibration error at most epsilon?

Answer: 1/epsilon^3 samples is both necessary and sufficient.

[Sílvia Casacuberta @SiCaPu]

If you’re interested in prediction vs downstream decision-making, calibration (in the predictive or generative setting), uncertainty quantification, multigroup learning, or any related problems, I would love to chat!

[Sílvia Casacuberta @SiCaPu]

Excited to be attending #ICLR in Rio this week! I will be presenting our paper (joint with Moritz Hardt) on the sample complexity of treatment allocation.

I will be at Pavilion 4 on Saturday 3:15-5:45 pm (poster session 4, #4209).

https://t.co/QOVrXTJ7CP

[Simons Institute for the Theory of Computing @SimonsInstitute]

Next week at the Simons Institute, a workshop on Agency in Collaborative Learning. Join us!

https://t.co/VcDcJA8fWX

[Harvard SEAS @hseas]

Harvard researchers have launched the Differential Privacy Deployments Registry, a public database that catalogs real-world uses of differential privacy by companies and agencies to better protect individuals’ data. https://t.co/obF8y2kQKq

[Siddartha Devic @sid_devic]

Looks like a super useful tool for practitioners looking to apply multi calibration techniques!!

[Boaz Barak @boazbaraktcs]

New award: Trevisan Award for Expository Work. Love the idea of awarding work that explains and illuminates, and it is so fitting way to honor Luca, who was a master explainer.

https://t.co/WTCu7vk2Cf