Learning via Surprisability — the power of absence

Machine learning almost always looks for what’s present. I built a method that also listens for what’s missing. Learning via Surprisability (originally Latent Personal/Population Analysis) is an information-theoretic way to represent people and populations in high-dimensional space, grounded in one idea: the absence of an expected feature carries information. It attributes authorship, detects impersonation, and spots outliers in temporal — and even sparse biological (immunology) — data, outperforming unsupervised methods, matching black-box models, and running far more efficiently. I’ve extended it to visualization and summarization interfaces that make absence perceptible and actionable, and I am building an ERC Advanced proposal around the approach.

Selected work: interpretable timeline analysis via surprisability (Mokryn et al., 2025); domain-based Latent Personal Analysis for impersonation detection (Mokryn & Ben-Shoshan, 2021); and LPA applied to B-cell clone diversity in immunology (Alon et al., 2021).

References

2025

Chaos

Interpretable Transformation and Analysis of Timelines through Learning via Surprisability

Osnat Mokryn, Teddy Lazebnik, and Hagit Ben-Shoshan

Chaos: An Interdisciplinary Journal of Nonlinear Science, 2025

Bib

@article{mokryn2025interpretable,
  title = {Interpretable Transformation and Analysis of Timelines through Learning via Surprisability},
  author = {Mokryn, Osnat and Lazebnik, Teddy and Ben-Shoshan, Hagit},
  journal = {Chaos: An Interdisciplinary Journal of Nonlinear Science},
  volume = {35},
  number = {7},
  pages = {073134},
  year = {2025}
}

2021

UMUAI

Domain-Based Latent Personal Analysis and its Use for Impersonation Detection in Social Media

Osnat Mokryn and Hagit Ben-Shoshan

User Modeling and User-Adapted Interaction, 2021

Bib

@article{mokryn2021domain,
  title = {Domain-Based Latent Personal Analysis and its Use for Impersonation Detection in Social Media},
  author = {Mokryn, Osnat and Ben-Shoshan, Hagit},
  journal = {User Modeling and User-Adapted Interaction},
  volume = {31},
  number = {4},
  pages = {785--828},
  year = {2021}
}

Front. Immunol.

Using Domain-Based Latent Personal Analysis of B Cell Clone Diversity Patterns to Identify Novel Relationships between the B Cell Clone Populations in Different Tissues

Uri Alon, Osnat Mokryn, and Uri Hershberg

Frontiers in Immunology, 2021

Bib

@article{alon2021bcell,
  title = {Using Domain-Based Latent Personal Analysis of B Cell Clone Diversity Patterns to Identify Novel Relationships between the B Cell Clone Populations in Different Tissues},
  author = {Alon, Uri and Mokryn, Osnat and Hershberg, Uri},
  journal = {Frontiers in Immunology},
  volume = {12},
  pages = {642673},
  year = {2021}
}