Welcolme to my personal page!

I am a researcher working at the intersection of machine learning and particle physics. My work focuses on developing principled statistical and generative modeling methods for high-energy physics, including normalizing flows, transformers, and foundation models for collider data. I am particularly interested in building reliable, scalable ML systems for scientific discovery, bridging modern AI techniques with the demands of fundamental physics.

Present research

Below is a summary of my current research avenues:

Foundation models in High Energy Physics

High-energy physics provides a natural testbed for developing large-scale foundation models: it offers vast amounts of structured, high-dimensional data governed by well-defined physical principles. My research focuses on building transformer-based architectures trained directly on collider data, moving beyond simulation-driven pipelines toward models that learn general-purpose representations from real LHC events. These models can be fine-tuned for tasks such as jet classification, fast simulation, and anomaly detection. A central component of my work is ensuring that such foundation models are interpretable, scalable, and statistically reliable.

Large-scale ML validation

Reliable AI for science requires statistically rigorous validation, particularly in high-dimensional settings. I develop principled methods to assess generative and density-estimation models used in collider physics. This work has led to the current development of GENOVA, a validation framework designed to quantify generative fidelity and robustness in complex scientific datasets. By hihg dimensional hypothesis testing, my goal is to establish scalable standards for trustworthy machine learning in high-energy physics and other data-intensive sciences. Moreover, I also foucs on assesing the limits of ML classification by robust methods based on the Neyman-Pearson lemma.

Data-driven new physics searches

The increasing complexity and volume of LHC data call for new strategies beyond traditional hypothesis-driven searches. My research develops data-driven and weakly supervised methods for discovering new physics, including anomaly detection, likelihood learning with normalizing flows, and ML-adapted reinterpretation of collider results. I also work on data-driven approaches to accelerate parameter exploration in search for new physics. These approaches aim to transform collider phenomenology into a more adaptive, ML-assisted framework where new physics can be systematically explored in high-dimensional parameter spaces