About
Hi, I’m Federico 👋, a research scientist at the CENTAI Institute in Torino. I hold a Ph.D. in Data Science from Sapienza University of Rome.
My research focuses on artificial intelligence for sociotechnical systems. I currently work on:
- developing new empirical and theoretical reinforcement learning algorithms,
- applying network-science methods to analyze internal representations of large language models for improved safety and honesty, and
- auditing LLM behaviors related to sociodemographic variation, personas, and model-based web-search capabilities.
Feel free to reach out if you’re interested in collaborating ✉️:
federico dot cinus [at] centai dot eu
Latest News
November 2025
- 📄 New preprint (Under review): Auditing LLM Editorial Bias in News Media Exposure .
October 2025
- 📊 🔍 Released an interactive tool to explore inferred sociodemographic patterns across subreddits: reddit-fabric.
- 📄 New preprint (Under review): Online Minimization of Polarization and Disagreement via Low-Rank Matrix Bandits .
August 2025
- 🎤 Honored to present our paper at the International Joint Conference on Artificial Intelligence (IJCAI 2025) in Montreal 🇨🇦.
July 2025
- 🎉 Paper accepted at EPJ Data Science: Measuring how computer science research translates into innovation and development.
May 2025
- 🎓 Earned my Ph.D. in Data Science (cum laude) from Sapienza University of Rome.
- 🎉 Paper accepted at KDD 2025: Engagement-Driven Content Generation with LLMs.
April 2025
- 🎉 Paper accepted at IJCAI 2025: Minimizing Polarization and Disagreement in the Friedkin-Johnsen Model with Unknown Innate Opinions. See you in Montreal!
- 🎓 Presented at WWW 2025 in Sydney.
March 2025
- 🎉 New paper accepted at TIST.
- 🎓 Completed reviews for KDD 2025.
February 2025
- 📄 New Preprint: Minimizing Polarization and Disagreement in the Friedkin-Johnsen Model with Unknown Innate Opinions. Read it here.
- 📄 New Preprint: On the Inference of Sociodemographics on Reddit. Check it out here.
January 2025
- 🎉 Accepted at The Web Conf 2025! Our paper Exposing Cross-Platform Coordinated Inauthentic Activity in the Run-Up to the 2024 U.S. Election has been accepted! Read the preprint here.
December 2024
- 🎓 I'm happy to be reviewing papers for The Web Conference 2025.
- 📄 New Preprint: We've just published a preprint exploring how large language models can generate content optimized for user engagement using "social" reinforcement learning and opinion dynamics. Check it out here.