Lorenzo Bini

Bio

Greetings everyone, I'm Lorenzo, PhD candidate in Artificial Intelligence at the University of Geneva. I work with Stéphane Marchand-Maillet as a member of the VIPER group, closely collaborating with the DMML group. My research focuses on developing self-supervised and active learning strategies to streamline costly labeling in high-dimensional data. I specialize in Graph Neural Networks, from adversarial learning to representation learning and anomaly detection in self- or weakly-supervised settings. I'm also building generative models (e.g., DDPMs and Flow Matching) for 3D genomics, flow cytometry, scRNA, spatial transcriptomics, and multi-omics data. Lately, I've been using LLMs to streamline routine hospital processes.

In collaboration with the Geneva University Hospital (HUG), I'm working on detecting Minimal Residual Disease of Acute Lymphoblastic and Myeloid Leukemia from flow cytometry data, where SSL and AL methods reduce the burden of physician-led annotation. I'm also a teaching assistant for “Introduction to Computational Finance,” “Natural Language Processing,” and “Information Retrieval” at the CUI. Recently, I've joined Jian Ma's group at the Ray and Stephanie Lane Computational Biology Department at Carnegie Mellon University as a PhD researcher, working on GenAI for epigenomics and gene regulation.

Check out our latest works:

• The first comprehensive benchmark for multi-class single-cell classification on flow cytometry data, using GNNs and other state-of-the-art single-cell deep learning techniques.
• A plug-in module to inject biological priors into state-of-the-art GNNs for hierarchical single-cell classification: FCHC-GNN.
• Decoding attention for domain-dependent interpretable GNNs: the first study revealing the emergence of Massive Activations (MAs) within GNN attention mechanisms, plus an Explicit Bias Term (EBT) to switch them off.
• LapDDPM: a conditional graph diffusion probabilistic model for robust and high-fidelity scRNA-seq generation that combines Laplacian positional encodings, spectral adversarial perturbations, and conditional score-based diffusion.
• A new self-supervised graph learning framework that bypasses negative sampling via spectral bootstrapping and adversarially enhanced Laplacian-based signals.
• Our MAs paper was accepted as an oral presentation at ECAI 2025.

Publications

Most recent publications on Google Scholar .

• Selected
• All

* indicates first-author equal-contribution.

Vitae

Full CV in PDF.

• Carnegie Mellon University 2025 — current

  PhD Research Internship, Computational Biology Department
  GenAI for epigenomics and gene regulation in Jian Ma's group
• University of Geneva PhD — current

  PhD Candidate in Artificial Intelligence
  VIPER group with Stéphane Marchand-Maillet working on graph generative models and self-supervised/adversarial learning
• Geneva University Hospital (HUG) Current collaboration

  Minimal Residual Disease detection in Leukemia from flow cytometry data
  Active/self-supervised learning for rare AML/ALL event detection
• University of Geneva, CUI Teaching — current

  Teaching Assistant
  Computational Finance; Natural Language Processing; Information Retrieval

Contact

Email: lorenzo.bini [AT] unige.ch

Geneva-based: Department of Computer Science, BAT A, Route de Drize 7 1227 Carouge, Switzerland. Office Room #221.

Pittsburgh/CMU-based: Ray and Stephanie Lane Computational Biology Department, Gates Hillman, 5000 Forbes Avenue Pittsburgh, PA 15213, USA. Office Room #7601.