Gennaro Gala
PhD candidate in Probabilistic Machine Learning @ TU/e
I'm Gennaro, a deep learner working on tractable/self-organizing generative modelling. I have broad interests in (equivariant) deep learning and particularly like generative models as VAEs, GANs, flows, diffusion, tractable probabilistic models; and architectures as GNNs, transformers, and vector quantized models.
📧 gennarogala1396@gmail.com gengala13 github.com/gengala
🎓 google scholar 📰 Curriculum Vitae
🎓 Education
PhD candidate at UAI group
Apr 2021 - Apr 2025, Eindhoven University of Technology (NL)
Master's Degree in Computer Science & Machine Intelligence
Oct 2018 - Dec 2020, University of Bari Aldo Moro (IT)
110/110 with honors (higher possible grade)
Bachelor's Degree in Computer Science
Oct 2015 - Oct 2018, University of Bari Aldo Moro (IT)
110/110 with honors (higher possible grade)
💼 Experience
Intern @ Italian National Research Council
Bari, Italy – (Mar 2018 - Sep 2018)
Intern researcher working on ML for marine biology 🐬
📚 Selected Publications
Scaling Continuous Latent Variable Models
as Probabilistic Integral Circuits
Gennaro Gala, Cassio de Campos, Antonio Vergari, Erik Quaeghebeur
TL; DR: We provide a comprehensive pipeline to build and scale hierarchical continuous latent variable models through the lenses of DAG-shaped probabilistic (integral) circuits
📄 arxiv preprint (under review)
Probabilistic Integral Circuits
Gennaro Gala, Cassio de Campos, Robert Peharz, Antonio Vergari, Erik Quaeghebeur
TL; DR: We add integral units to the language of tractable circuits to represent continuous latent variables and learn hierarchical tree-shaped neural continuous mixtures
📄 AISTATS paper gengala/pic
E(n)-equivariant Graph Neural Cellular Automata
Gennaro Gala, Daniele Grattarola, Erik Quaeghebeur
TL; DR: By replacing standard graph convolutions with E(n)-equivariant ones, we design and propose a class of isotropic automata that we call E(n)-GNCAs. These models are lightweight, but can nevertheless handle large graphs, capture complex dynamics and exhibit emergent self-organising behaviours.
📄 TMLR paper gengala/egnca
Continuous mixtures of tractable probabilistic models
Alvaro H.C. Correia*, Gennaro Gala*, Erik Quaeghebeur, Cassio de Campos, Robert Peharz
TL; DR: We investigate a hybrid approach: (small-dimensional) continuous intractable mixtures of tractable probabilistic models. While these models are analytically intractable, they are amenable to numerical integration schemes based on a finite set of integration points.
📄 AAAI paper AlCorreia/cm-tpm
Bayesian Structure Scores for Probabilistic Circuits
Yang Yang*, Gennaro Gala*, Robert Peharz
TL; DR: We develop Bayesian structure scores for deterministic PCs, i.e., the structure likelihood with parameters marginalized out, which are well known as rigorous objectives for structure learning in probabilistic graphical models.
📄 AISTATS paper
🛠️ Skills
- [Probabilistic Models] [Deep Learning] [Generative AI] [Neural Cellular Automata]
- [pytorch] [lightning] [python] [C/c++] [MATLAB] [TensorFlow] [sklearn]