AI Research
I am a Research Scientist specializing in AI-driven deep learning, machine learning, and computer vision. My expertise spans Transformers, Meta-Learning, Large Language Models (LLMs), and Natural Language Processing (NLP). I have conducted applied research in deep learning and computer vision at MindVisionLab (a spin-off of University College London) in partnership with Toyota Research Institute. I also held a three-year postdoctoral position as a Senior Researcher at the University of Glasgow, where I lectured and mentored Ph.D. students.
With 23 publications in top-tier venues (ICCV, ECCV, TOG, T-PAMI) and over 2,100 citations, my research contributions include feature selection, attention models, medical AI, and video-based temporal analysis. I have developed widely used open-source software, including the FSLib MATLAB toolbox for feature selection and a private LLMSuite in PyTorch for Large Language Models. My work has been recognized with multiple awards, including the CVPR Outstanding Reviewer Award, NVIDIA GPU Grant, and MathWorks Research Summit invitation.
At Cosmo Intelligent Medical Devices (Cosmo IMD), Milan, I focus on AI-driven medical image and video analysis, ensuring model compliance and evaluation.
Recent News
LLMSuite Python toolbox
Talk on Vision Transformers
Here are the slides from the vision transformers talk: Vision Transformers Slides
Recent AI Talks
- LLMs Applications (RAG, PAL, CoT, ReAct) [Download slides]
- Transformers, Computational Costs (FSDP, ZeRO), Datasets [Download slides]
- LLM Fine-Tuning, PEFT, LoRA, RLHF, ReST [Download slides]
- Advancements in Video Understanding: TSM Temporal Shift Module
- Non-local Neural Networks
- Interpretability & Attribution Methods ​
SELECTED PUBLICATIONS
Feature Selection, Computer Vision, and Deep Learning:
- Hard-Attention Gates with Gradient Routing for Endoscopic Image Computing
G. Roffo, C. Biffi, P. Salvagnini, A. Cherubini
Published in Medical Image Computing and Computer Assisted Intervention. (MICCAI), 2024
Relevant to: Automated Medical Imaging Analysis
- Infinite Feature Selection: A Graph-based Feature Filtering Approach
G. Roffo, S. Melzi, U. Castellani, A. Vinciarelli, and M. Cristani
Published in IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2020
Relevant to: Content Recommendation,Autonomous Systems, AI-Driven Marketing Analytics, Intelligent Customer Service ChatBots
- Infinite Latent Feature Selection: A Probabilistic Latent Graph-Based Ranking Approach
G. Roffo, S. Melzi, U. Castellani, A. Vinciarelli
Presented at the IEEE International Conference on Computer Vision (ICCV), 2017
Relevant to: Content Recommendation,Autonomous Systems, AI-Driven Marketing Analytics, Intelligent Customer Service ChatBots
- Infinite Feature Selection
G. Roffo, S. Melzi, and M. Cristani
Presented at the IEEE International Conference on Computer Vision (ICCV), 2015
Relevant to: Content Recommendation,Autonomous Systems, AI-Driven Marketing Analytics, Intelligent Customer Service ChatBots
- Automating the Administration and Analysis of Psychiatric Tests: The Case of Attachment in School Age Children
G. Roffo, D.-B. Vo, A. Sorrentino, M. Rooksby, M. Tayarani, S. Di Folco, H. Minnis, S. Brewster, and A. Vinciarelli
Presented at the Conference on Human Factors in Computing Systems (ACM CHI), 2019
Relevant to: Intelligent Customer Service ChatBots, Educational Tools,Virtual Try-On for Retail
Shape Analysis and Matching:
- Discrete Time Evolution Process Descriptor for Shape Analysis and Matching
S. Melzi, M. Ovsjanikov, G. Roffo, M. Cristani, U. Castellani
Published in ACM Transactions on Graphics (ACM TOG), 2018
Relevant to: Virtual Try-On for Retail,Marketing Analytics
Natural Language Processing - Chats/Instant Messaging:
G. Roffo, M. Cristani, L. Bazzani, H. Q. Minh, V. Murino, IEEE International Conference on Computer Vision Workshops (ICCV) 2013.
Relevant to: Enhanced Security Systems.
M. Cristani, G. Roffo, C. Segalin, L. Bazzani, A. Vinciarelli, V. Murino, ACM International Conference on Multimedia (ACMM) 2012. Top 1% Computer Science - Scimago Journal & Country Rank.
Relevant to: Intelligent Customer Service ChatBots.
OTHER PROJECTS
- Infinite Feature Selection: Developed in Matlab, this tool introduces an innovative graph-based feature filtering approach, which has been a cornerstone of Roffo’s research.
- Feature Selection Gates: Python/PyTorch FSG module to enhance Transformers and Vision Transformers. MICCAI 2024.
- FSLib Library for Feature Selection: A comprehensive Matlab toolbox designed to facilitate research and application of feature selection techniques.
- Large Language Models (LLMSuite): A toolbox designed for evaluating large language models, contributing tools and methodologies to assess the effectiveness and efficiency of these models in various applications.
In addition to his primary research, Giorgio has been involved in several educational projects, particularly focusing on teaching Python at the university level. These projects include practical implementations and templates that serve as educational tools for students studying various aspects of machine learning and neural networks:
- Bayes Classifier with Gaussian Mixture Models: A project that demonstrates how to use Gaussian Mixture Models to generate handwritten images, leveraging Bayesian inference techniques.
- Auto-Encoders: A recommender system project that predicts ratings from 1 to 5 using the MovieLens 1M Dataset, implemented using PyTorch and autoencoders.
- Restricted Boltzmann Machines: An implementation of a recommender system based on Restricted Boltzmann Machines, again using the MovieLens 1M Dataset.
- Self-Organizing Maps: A template for applying Self-Organizing Maps (SOM) for banking fraud detection.
- Recurrent Neural Network: A project employing recurrent neural networks with LSTM units for stock price prediction.
- Convolutional Neural Network: A convolutional network designed for object recognition, illustrating the basic principles of CNNs.
- Artificial Neural Network: A simple implementation of an ANN in Python, using Keras to demonstrate fundamental neural network concepts in business analytics.
DATASET & CODE
Open-Source Datasets
I maintain and contribute to several open-source computer vision datasets, enabling research and development in the field.
GitHub Repositories
My GitHub profile hosts a range of code repositories, from innovative Large Language Models to feature selection algorithms.
Reproducible Research
I strive for transparency and reproducibility in my work, providing detailed documentation and code to support my published research.
CONTACT
Giorgio Roffo can be reached through his profiles on major academic and professional networking sites. For insights into his publications, ongoing projects, or to connect professionally, please visit the following links:
- Papers with Code: Research by Giorgio Roffo
AWARDS & RECOGNITION
- Giorgio Roffo has received several notable awards and recognitions that underscore his contributions to the fields of computer vision and machine learning:
- CVPR 2019 Outstanding Reviewers Award: This prestigious award highlights his significant contributions to the peer review process at the top conference in computer vision.
- MathWorks Research Summit Award (2019): Acknowledgement for his innovative research in deep learning, which was recognized at a key industry summit.
- NVIDIA GPU Grant (2017): Roffo was awarded this grant which provided crucial computational resources, aiding in advanced AI research.
EDUCATION
1
Ph.D., Doctor Europaeus in Computer Science
University of Verona, Italy (2014-2017)
Thesis: "Ranking to Learn and Learning to Rank"
Awarded with the prestigious Doctor Europaeus certification, recognizing his research's adherence to the high international standards and its broad academic significance.
Thesis: "Ranking to Learn and Learning to Rank"
Awarded with the prestigious Doctor Europaeus certification, recognizing his research's adherence to the high international standards and its broad academic significance.
Postdoctoral Researcher
University of Glasgow, UK (2017-2020)
2
Master Level I in Computer Game Development (MGDev)
University of Verona, Italy (2013)
Specialized program concentrating on game development technologies and methodologies, enhancing his skills in complex software system design.
Specialized program concentrating on game development technologies and methodologies, enhancing his skills in complex software system design.
3
MS, Master's Degree in Computer Science and Engineering
University of Verona, Italy (Completed 2011)
Focused on advanced computational techniques and software engineering principles.
Focused on advanced computational techniques and software engineering principles.
4
BS, Bachelor Degree in Computer Science and Engineering
University of Verona, Italy (Completed 2009)
His undergraduate studies laid the groundwork for his deep technical expertise in algorithms, programming, and system design.
His undergraduate studies laid the groundwork for his deep technical expertise in algorithms, programming, and system design.
Research and Work Experience
- AI Research Scientist, COSMO Intelligent Medical Devices (IMD), Milan, Italy
(2022 - Present)
At COSMO IMD, Roffo focuses on leveraging deep learning and computer vision to innovate in medical applications. His work includes developing AI systems and novel deep learning architectures that improve the effectiveness and accessibility of medical diagnostics and treatments.
- Research Scientist, Mind Vision Labs, London, UK
(2020 - 2022)
Collaborated with UCL London and Toyota Research Institute (TRI) Brussels on autonomous driving research. Specialized in vision transformers and advanced computer vision algorithms to enhance the reliability and safety of autonomous vehicle technologies.
- Research Associate, University of Glasgow, School of Computing Science, UK
(2017 - 2020)
Contributed to the EPSRC Project: Multimodal Deep Learning. Focused on attention models and transformer networks, developing solutions that are not only technically robust but also interpretable and scalable across different applications.
- Ph.D. Student, University of Verona, Italy
(2014 - 2017)
Conducted seminal research on "Ranking to Learn and Learning to Rank" in pattern recognition, which has been recognized with a Doctor Europaeus certificate, indicating the high international standards of his work.
- Research Assistant, Italian Institute of Technology, Pattern Analysis and Computer Vision department
(2012)
Worked on advanced computer vision algorithms and pattern analysis, contributing to foundational research that supports automated visual understanding technologies.
- Intern, University of Glasgow, School of Psychology
(2012, Summer)
Explored the intersection of technology and psychology, particularly focusing on how machine learning can be applied to understand and predict human behavior based on visual and interactive cues.
