Gaia Grosso

Postdoctoral Associate
Laboratory for Nuclear Science (MIT)
School of Engineering and Applied Sciences (Harvard SEAS)

Learning begins where expectation fails.

I am a physicist working across statistics and machine learning. I build methods that decide — rigorously — when data contain something a model did not predict. I did my PhD at CERN, and after that joined IAIFI as a postdoctoral Fellow, working with Philip Harris (MIT LNS) and Demba Ba (Harvard SEAS).

I am on the 2026–27 faculty job market. I am looking for positions where statistical methodology, machine learning, and the physical sciences meet — get in touch.

Email CV Scholar INSPIRE ORCID GitHub LinkedIn

Research

An anomaly is not always visible one observation at a time. A loaded die looks fair on any single throw and gives itself away only in the frequencies of many — and the same holds for new fundamental interactions distorting the distributions of particle collisions, or for a language model whose behaviour has drifted in ways no single output reveals. Classical tests answer that question rigorously but lose power when deviations are rare or buried among uninformative features. I build tests that scale without giving up the calibrated p-value or the account of what produced it.

News

Oct 2026

New paper An end-to-end framework for uncertainty-aware validation of generative models is on arXiv — led by Giada Badaracco, on propagating the error of a fitted generative model into the goodness-of-fit test itself. arXiv:2610.37398

28 Sep 2026

New paper Covariant contrastive learning for uncertainty-aware anomaly detection is on arXiv — led by Shelley Tong (MIT), on building latent representations that behave predictably under systematic distortions. arXiv:2609.35755

Sep 2026

Invited talk at the LHC Physics Center, Fermilab — Discovery through the lens of AI: scaling anomaly detection with learned representations.

Sep 2026

Started as Postdoctoral Associate at the MIT Laboratory for Nuclear Science, following three years as an IAIFI Fellow.

Aug 2026

New paper VERAIPHY, the opening article of the PHYSTAT series on statistical standards for machine learning in fundamental physics. I also joined the PHYSTAT organizing team and Advisory Committee. arXiv:2608.17724

Jul 2026

Presented Adaptive and interpretable two-sample test as a batch-level tool for AI safety as a poster at the ICML Workshop on Mechanistic Interpretability, Seoul.

10 Jul 2026

New paper Are we ready for AI-driven discovery? — my review for the VERAIPHY series, on when verification is essential in the statistical discovery workflow. arXiv:2607.10039

Feb 2026

Co-organized the VERAIPHY workshop in Amsterdam (16–17 February) — the initiative’s first meeting on validation and evaluation practice for machine learning in fundamental physics.

Nov 2025

New paper SparKer — sparse, self-organizing ensembles of local kernels that detect rare statistical anomalies — is on arXiv, and is covered by a US provisional patent application. arXiv:2511.03095

Jun 2025

Co-recipient, as a member of CMS, of the 2025 Breakthrough Prize in Fundamental Physics awarded to the LHC collaborations.

Contact

Emailgaiag795@mit.edu · ggrosso@seas.harvard.edu

CVcv.pdf · updated October 2026

Based inCambridge, MA

I am glad to hear from students and researchers interested in anomaly detection — whether the data comes from a collider, a telescope, or a language model.