About

I am a Postdoctoral Researcher with a joint affiliation with Inria Sophia Antipolis, in the McTAO team, and Université Côte d’Azur, in the Laboratoire J.A. Dieudonné (LJAD). I am supervised by Samuel Vaiter and Jean-Baptiste Caillau.

I received my Ph.D. in Computer Science from the University of Genoa, where I was supervised by Lorenzo Rosasco and Silvia Villa.

My research focuses on black-box and zeroth-order optimization, namely the design, implementation, and theoretical analysis of optimization algorithms in settings where first-order information is not available. I study these methods across a broad range of settings, including deterministic and stochastic environments, smooth and non-smooth objectives, convex and non-convex problems, compositional functions, and bilevel optimization. My interests include finite-difference methods, variance reduction, and zeroth-order Langevin-based methods for non-convex optimization.

A related direction of my research concerns optimization over probability distributions, with applications to large language models (LLMs). I am particularly interested in memory-efficient zeroth-order methods for LLM pre-training and fine-tuning, as well as fast first-order methods for reinforcement-learning-based optimization, where efficiently estimating the REINFORCE gradient while reducing the cost of sampling is a central challenge.

I am also interested in applying machine learning and optimization to scientific and engineering problems, including high-energy physics, robotics, biology, and large-scale machine learning. My work includes reinforcement learning for olfactory navigation in turbulent environments, gradient-free optimization for tuning humanoid robot walking controllers, scalable kernel methods for large-scale machine learning, and optimization-based techniques for adversarial machine learning, with applications including malware detection and large language models.

Main Research Interests

• Black-box and zeroth-order optimization
• Optimization over probability distributions
• Non-convex and non-smooth optimization
• Stochastic and bilevel optimization
• Optimization for large language models
• Adversarial machine learning
• Reinforcement learning and navigation
• Machine learning for scientific applications

Contact

You can contact me by email at marco.rando0396@gmail.com (marco.rando0396 [at] gmail.com)