Marco Romanelli

Research Associate

NYU Tandon School of Engineering

Bio

I am a Research Associate at NYU Tandon School of Engineering: Electrical and Computer Engineering. My interests lie at the intersection between information theory and machine learning on the one hand, and privacy and security on the other. I received my PhD from École Polytechnique and Inria in 2020.

Interests

Information Theory applied to Privacy and Security
Machine Learning applied to Privacy and Security
Safety in AI
Information Leakage Measurement

Education

PhD in Computer Science, 2020
École Polytechnique, Inria, Lix
MSc in Computer and Automation Engineering, 2017
Università di Siena, Dipartimento di Ignegneria dell'Informazione e Scienze Matematiche
BSc in Ingengeria Informatica e dell'Informazione, 2014
Università di Siena, Dipartimento di Ignegneria dell'Informazione e Scienze Matematiche

News

Check out our new preprint Low-rank fine-tuning for LLMs: A fairness perspective ! In this paper we analyzed the effects of Low Rank fine-tuning (LoRA) on the fairness and bias of Large Language Models (LLMs).

May 31, 2024

Our paper Beyond the Norms: Detecting Prediction Errors in Regression Models was accepted to appear as a spotlight at the Forty-first International Conference on Machine Learning (ICML 2024)! In this paper we propose a framework to evaluate uncertainty in regression models!

May 9, 2024

Our paper Disparate Impact on Group Accuracy of Linearization for Private Inference was accepted to appear at the Forty-first International Conference on Machine Learning (ICML 2024)! In this paper we analyzed the effects on Fairness due to ReLU budgeting for Private Inference speedup.

May 6, 2024

Check out our new preprint Disparate Impact on Group Accuracy of Linearization for Private Inference! In this paper we analyzed the effects on Fairness due to ReLU budgeting for Private Inference speedup.

Feb 2, 2024

Our new paper On the (In)feasibility of ML Backdoor Detection has been accepted to appear at AISTATS 2024. In this paper we propose a theoretical framework that encompasses the problems of Backdoor and OOD detection, showing for the first time their intrinsic limitation which results in a ground breaking infeasibility theorem.

Jan 21, 2024

Our new paper Optimal Zero-Shot Detector for Multi-Armed Attacks has been accepted to appear at AISTATS 2024. In this paper we propose a minimum-regret optimal framework to aggregate adverarial attack detectors to minimize the worst-case risk.

Jan 20, 2024

Our new paper Retrieval-Guided Reinforcement Learning for Boolean Circuit Minimization has been accepted to appear at ICLR 2024. In this paper we proposed an OOD detection based strategy to inform the reward function for policies in RL, with specific applications to secure automated hardware design.

Jan 18, 2024

Our new paper A Data-Driven Measure of Relative Uncertainty for Misclassification Detection has been accepted to appear at ICLR 2024. In this paper we proposed a data-driven method, powered by a statistical diversity and dissimilarity metric, to detect incorrect classification at test time by assessing the uncertainty of a given model.

Jan 16, 2024

Our new paper MEAD: A Multi-Armed Approach for Evaluation of Adversarial Examples Detectors, which sets up a new evaluation framework for adversarial example detection mechanisms, has been accepted to appear at ECMLPKDD 2022.

Sep 18, 2022

I am excited to start a new position as Research Associate at NYU Tandon School of Engineering: Electrical and Computer Engineering.

Sep 16, 2022

Software

The main software repository can be found at GitHub. My previus contributions can be found at GitLab, GitLab@MIPAN, GitHub@LEAVES.