Go directly to page contentent
TEAM LADE

Alberto Cazzaniga

Researcher at Laboratory of Data Engineering

I am a Researcher at the Research and Innovation Technology Institute (RIT) at Area Science Park in Trieste, where I coordinate the activities of the Laboratory of Data Engineering (LADE) focusing on applications of Artificial Intelligence to material and life sciences.

I study the emergence of meaningful features in deep-learning models trained by self-supervision, and in particular transformers. I believe that understanding computational strategies of these models will help enhancing their performance in applications, and make them more robust and trustworthy. I enjoy looking at neural networks through the lenses of geometry, inspired and influenced by my background in Algebraic Geometry.

I teach Natural Language Processing at the University of Trieste and Generative AI at the Master in high Performance Computing at SISSA and ICTP.

 

Research Interests 
  • Large Transformer Models
  • Geometry of Representations
Experience & Education
  • Master in High Performance Computing (ICTP & SISSA) 2020
  • PhD in Mathematics, University of Oxford 2016
  • Masterclass in Moduli Spaces, Utrecht University, 2011
Latest pubblications
02/06/2026
Visual Instruction Tuning Aligns Modalities through Abstraction
Abstract Visual instruction tuning effectively adapts a pre-trained Large Language Model (LLM) to process image…
Go to the news Visual Instruction Tuning Aligns Modalities through Abstraction
11/06/2026
Tumour evolution as ground truth for cancer whole-genome sequencing
Abstract Cancer genomes are shaped by evolutionary processes that couple mutagenesis, clonal selection, chromosomal instability,…
Go to the news Tumour evolution as ground truth for cancer whole-genome sequencing
01/07/2026
When Seeing Overrides Knowing: Disentangling Knowledge Conflicts in Vision-Language Models
Abstract Vision-language models (VLMs) increasingly combine visual and textual information to perform complex tasks. However,…
Go to the news When Seeing Overrides Knowing: Disentangling Knowledge Conflicts in Vision-Language Models
18/09/2025
The Narrow Gate: Localized Image-Text Communication in Native Multimodal Models
Abstract Recent advances in multimodal training have significantly improved the integration of image understanding and…
Go to the news The Narrow Gate: Localized Image-Text Communication in Native Multimodal Models
01/03/2026
Preserving Historical Truth: Detecting Historical Revisionism in Large Language Models
Abstract Large language models (LLMs) are increasingly consulted for historical information by citizens, journalists, and…
Go to the news Preserving Historical Truth: Detecting Historical Revisionism in Large Language Models
28/11/2025
Are LLMs Good Safety Agents or a Propaganda Engine?
Abstract Large Language Models (LLMs) are trained to refuse to respond to harmful content. However,…
Go to the news Are LLMs Good Safety Agents or a Propaganda Engine?