I'm Davide Evangelista, a Junior Assistant Professor in the Department of Computer Science and Engineering at the University of Bologna.
I work on inverse problems in imaging: reconstructing images from measurements that are noisy, incomplete, or expensive to acquire. Much of my research concerns medical imaging, especially computed tomography, where I combine numerical optimization and regularization with deep learning.
Research
A recurring question in my work is how much we can trust a learned reconstruction. An image can look convincing and still be inconsistent with the measurements. I study the stability of neural networks for inverse problems and develop methods that use learned information within a mathematical model of the acquisition process. This includes work on sparse-view CT, adaptive regularization, and reducing contrast agent doses.
More recently, I’ve been working with diffusion models and flow matching as image priors. I’m interested in how to guide these models with measured data, recover missing structures, and make reconstruction less computationally expensive. I also work on diffusion models for weather downscaling, where the variability between generated samples matters as much as each individual prediction.
Alongside imaging, I collaborate on continual learning and language models. Recent projects cover normalization for data whose distribution changes over time, the continuity of language models, mathematical reasoning benchmarks, and hallucination detection through hidden representations.
The publications page has the papers and preprints behind these projects.
Teaching and code
I teach computational imaging, mathematical methods for machine learning, and deep generative models. My course pages collect lecture notes and practical material, with Python implementations alongside the mathematics.
Code for my research and teaching is available on GitHub. You can also find me on Google Scholar, or email me at davide.evangelista5@unibo.it.