Avatar

Lorenzo Perini

Research Scientist @ Meta

Biography

Lorenzo Perini is a research scientist in the Central Applied Science (CAS) team at Meta focusing on probabilistic machine learning. He earned his BSc in Mathematics from the University of Florence in 2017, and his MSc in Mathematical Engineering from Politecnico di Torino in 2019, where he specialized in data statistics and network optimization. His Master’s thesis, conducted in collaboration with Tierra S.p.A., explored predictive maintenance using Hidden Markov Models and Autoencoders.

In 2019, Lorenzo began his PhD in the DTAI Lab at KU Leuven under Prof. Dr. Jesse Davis, focusing on uncertainty quantification and anomaly detection. His work has been recognized with several fellowships, including a PhD fellowship from the Research Foundation – Flanders (FWO) and the Scientific Prize Gustave Boël-Sofina Fellowship for talented researchers for a long stay abroad. During his PhD, he was a visiting researcher at the University of Helsinki and completed an internship at Bosch Center for Artificial Intelligence (BCAI).

Lorenzo has published papers in prestigious conferences such as NeurIPS, ICML, KDD, AAAI, IJCAI, and ECAI. His main research interests include Uncertainty Quantification and Anomaly Detection, often with the introduction of the human-in-the-loop, e.g. via Active Learning and Learning to Reject. In March 2024, he defended his doctoral dissertation titled operational, uncertainty-aware, and reliable anomaly detection.

Interests

  • Uncertainty Quantification
  • Anomaly/Out-Of-Distribution Detection
  • Human-In-The-Loop
  • Weakly Supervised Learning
  • Learning to Reject
  • Bayesian Learning
  • Calibration

Education

  • Ph.D. in Machine Learning

    Oct 2019 - March 2024

    KU Leuven

  • MSc of Mathematical Engineering

    Oct 2017 - Jul 2019

    Politecnico di Torino

  • BSc of Mathematics

    Oct 2014 - Jul 2017

    Università degli Studi di Firenze

Skills

Probability and Statistics

Machine Learning

Python

Experience

 
 
 
 
 

Research Scientist, Central Applied Science

Meta

Dec 2024 – Present London, United Kingdom
 
 
 
 
 

Postdoctoral Researcher

KU Leuven

Apr 2024 – Nov 2024 Leuven, Belgium
 
 
 
 
 

Research Scientist Intern

Bosch Center for AI (BCAI)

Sep 2023 – Jan 2024 Renningen, Germany
Task: Research on designing a data quality metric to evaluate synthetic anomalies for training/tuning anomaly detectors. Outcome: Paper under submission.
 
 
 
 
 

Visiting Ph.D. Researcher

University of Helsinki

Feb 2022 – Jul 2022 Helsinki, Finland
Task: Research on estimating the contamination factor’s posterior distribution using Bayesian Learning and Variational Inference. Outcome: Paper at ICML 2023.
 
 
 
 
 

Ph.D. Researcher

KU Leuven

Oct 2019 – Mar 2024 Leuven, Belgium
Responsibilities include:

  • Publishing papers in peer-reviewed venues;
  • Giving talks;
  • Supervising thesis;
  • Reviewing papers;
  • Teaching assistant.
 
 
 
 
 

Data Scientist Intern

Tierra S.p.A.

Feb 2019 – Jul 2019 Turin, Italy
Tasks: data engineering, time-series analysis, deep learning, predictive maintenance, anomaly detection. Outcome: MSc thesis.

Grants & Awards

PhD Fellowship fundamental research (FWO)

PhD grant for the research project ‘‘Measuring and Exploiting the Uncertainty in Anomaly Detection’'.

Scientific prize Gustave Boël-Sofina Fellowship

F.R.S.-FNRS & FWO grant for talented PhD students for a long research stay abroad.

Overall ECML-PKDD Engagement Award 2020

Multiple nominations by the conference session chairs as particularly engaging speaker.

Recent & Upcoming Talks

Operational, Uncertainty-Aware, and Reliable Anomaly Detection

Anomaly detection methods aim to identify examples that do not follow the expected behavior. For various reasons, anomaly detection is …

Unsupervised Anomaly Detection with Rejection @ NeurIPS23

Anomaly detection aims at detecting unexpected behaviours in the data. Because anomaly detection is usually an unsupervised task, …

Learning from Positive and Unlabeled Multi-Instance Bags in Anomaly Detection @ KDD23

In the multi-instance learning (MIL) setting instances are grouped together into bags. Labels are provided only for the bags and not on …

Estimating the Contamination Factor’s Distribution in Unsupervised Anomaly Detection @ ICML23

Anomaly detection methods identify examples that do not follow the expected behaviour, typically in an unsupervised fashion, by …

How to Allocate your Label Budget? Choosing between Active Learning and Learning to Reject in Anomaly Detection @ AAAI23

Anomaly detection attempts at finding examples that deviate from the expected behaviour. Usually, anomaly detection is tackled from an …