About Me
Work in Progress
Education
M.Sc. AI and Robotics
Sapienza Università di Roma
2023 - Current
Final GPA: 30/30
Best Student of the course according to Sapienza “Exam Bonus” Award.
Recipient of Sapienza “Meritorious Students” Merits.
Currently enrolled as a second year student following the Honours’ Programme.
Attended the 3rd “European Summer School on Artificial Intelligence” (ESSAI).
B.Sc. Computer and System Engineering
Sapienza Università di Roma
2020 - 2023
Final Grade: 110 / 110 cum laude
Final GPA: 29.29/30
Recipient of Sapienza “Meritorious Students” Merits.
Thesis: Object Re-Identification for Multi-Object Tracking Applied to Precision Agriculture.
High School Diploma, Liceo Scientifico Tradizionale
Liceo Statale Democrito
2015 - 2020
Final Grade: 100 / 100 cum laude
Throughout high school, I participated in various student competitions in Math, Physics, Biology, Chemistry, and Computer Science at the school, regional, and national levels, both individually and as a team leader.
During my fourth year of high school, I was selected for and attended the “Scuola Estiva Nazionale per Studenti sulla Fisica Moderna 2019” (SENS-FM2019), which took place at the University of Udine.
Experience
Research Experience
ALCOR Lab, DIAG, Sapienza Università di Roma
June 2024 - February 2025
Research collaboration with the ALCOR Lab at the Department of Computer, Control, and Management Engineering (DIAG) of Sapienza University of Rome, conducted during my master’s degree. The work focused on improving ground-to-aerial image matching techniques using semantic segmentation to enhance alignment and comparison of images from different viewpoints.
This collaboration led to the publication of the paper “Enhancing Ground-to-Aerial Image Matching for Visual Misinformation Detection Using Semantic Segmentation” of which I’m a co-first author.
Publications
[1] Emanuele Mule*, Matteo Pannacci*, Ali Ghasemi Goudarzi*, Francesco Pro, Lorenzo Papa, Luca Maiano and Irene Amerini. Enhancing ground-to-aerial image matching for visual misinformation detection using semantic segmentation. In Proceedings of the Winter Conference on Applications of Computer Vision (WACV) Workshops, pages 795-803, February 2025. (article, repository, website)
*: Equal contribution
Highlighted Projects
Implementation in PDDL and IndiGolog of a planning task from the italian card game Machiavelli. We take into consideration a single turn of a player during the course of the game. The goal is to check whether it’s possible for the player to win from the current configuration (board state and player’s hand) and, in the positive case, to find a sequence of legal actions that do so.
Implementation in Python of 2 class structures representing RDFS-based Knowledge Graphs. The first version uses a Neo4J Session to perform updates, syntax check and deduction directly on the graph through queries in Cypher. The second version performs these operations in Python and is able to interact with a Neo4J Session for importing/exporting Knowledge Graphs.
Implementation of a simple 2D trajectory planner for ROS in C++. The code provides a node that listens to the required informations (current position, map structure and goal position), then it uses these informations to build the trajectory planning problem and solves it applying the A* algorithm. The resulting path is then visualized using OpenCV.
Implementation of ES (Evolutive Strategy) and GA (Genetic Algorithm) using multiple independent populations in the context of Multi-Agent Reinforcement Learning to train neural networks (Neuro-Evolution). The agents learn to play classic games from the PettingZoo library including Connect Four and TicTacToe by having the populations compete against each other.
Implementation of augmentation strategies and trasformer-based models for NLI (Natural Language Inference) on the FEVER dataset. The augmentation strategies are based on WSD (Word Sense Disambiguation) and SRL (Semantic Role Labeling) informations provided with the dataset.
Implementation of two methods to transfer the style of a 2D image onto a 3D object using differentiable rendering. The first method stylizes multiple 2D views of the 3D object and uses them as optimization targets for texture and mesh updates. The second method directly computes content and style losses on the rendered images and backpropagates the gradients to the 3D parameters.
Implementation of an ensemble of GINs (Graph Isomorphism Network) using Pytorch Geometric for a Graph Classification task on a Protein-Protein Association (PPA) dataset in presence of noise. The models of the ensembles are trained on different bootstraps of the original dataset using a noise-robust loss and regularization techniques.