research | Shailen K. Smith

I am broadly interested in AI safety, security, and trustworthiness. I’ve most recently been focusing on intersections between the following fields:

Training data privacy – model inversion, data extraction from generative models
Robustness – adversarial examples/training, semantic meaning of features
Interpretability – feature attribution, representation inversion, concept-based approaches

publications

2026

ICLR ’26

Reducing information dependency does not cause training data privacy. Adversarially non-robust features do.

Rasmus Torp^*, Shailen Smith^*, and Adam Breuer

In The Fourteenth International Conference on Learning Representations, 2026

Bib PDF

@inproceedings{torpreducing,
  title = {Reducing information dependency does not cause training data privacy. Adversarially non-robust features do.},
  author = {Torp, Rasmus and Smith, Shailen and Breuer, Adam},
  booktitle = {The Fourteenth International Conference on Learning Representations},
  year = {2026},
}

2022

URTC ’22

Are Fair Learning To Rank Models Really Fair? An Analysis Using Inferred Gender

Alexander Pietrick, Alyssa Romportl, Shailen Smith, and 3 more authors

In IEEE MIT Undergraduate Research Technology Conference (URTC), 2022

Bib PDF

@inproceedings{10002233,
  author = {Pietrick, Alexander and Romportl, Alyssa and Smith, Shailen and Olulana, Oluseun and Cachel, Kathleen and Rundensteiner, Elke},
  booktitle = {IEEE MIT Undergraduate Research Technology Conference (URTC)},
  title = {Are Fair Learning To Rank Models Really Fair? An Analysis Using Inferred Gender},
  year = {2022},
  keywords = {Analytical models;Inference algorithms;Fair ranking;learning to rank;demographic inference;fairness;uncertainty},
}