FastNRTF
Efficient 3D reconstruction and relighting of complex scenes with global illumination effects using Neural Radiance Transfer Fields.
See moreABOUT ME
COMPUTER & DATA SCIENTIST
Hi, my name is Elia and I'm a MSc Data Science student. I work on and love AI in all its shapes (Data Science, Machine Learning and Deep Learning), especially applied to my main areas of interest and passions: Computer Vision, Computer Graphics, Virtual Reality, Game Development, Music and Art.
Since I was a small kid I was fascinated by computers and videogames and such interest grew stronger as I got older. In 2021 I finally became a computer scientist by getting an Engineering of Computing Systems Bachelor’s degree. After that, specializing on AI was a no brainer choice for me: I always wanted to not only work in the field I love, but to research for its improvement and evolution as well. For this purpose, Artificial Intelligence is perfect as it can be applied to anything while being enormously interesting as a topic on its own.
Deep Learning & ML90%
Computer Vision & Graphics 85%
Data Analytics 85%
Software Development 80%
Game Development 80%
Some Tools
Languages:
AI and Data Science:
Game Development:
Web Development:
-
MY JOURNEY UNTIL NOW
-
Science High School Diploma
Republic of San Marino’s upper secondary school 2013-2018GPA: 100 /100 with honours
-
Engineering of Computing Systems Bachelor’s degree
Politecnico di Milano2018 - 2021GPA: 108/110 13th university in Engineering and Technology (2020 QS W.U. Rankings) Merit Based Scholarship 2020, Best Freshmen Prize 2019San Marino Merit Contribute 2019, 2020, 2021
-
MSc Data Science
EPFL, Swiss Federal Institute of Technology2021 - PresentCurrent GPA: 5.2/6 10th university in Computer Science and Information Systems (2022 QS World University Rankings)
-
Where next?
Projects
All the images with the four coloured squares on the bottom right were created using DALL·E 2 AI by OpenAI to metaphorically represent the content of the projects.
RocketMan
A Virtual-Reality "escape-room-like" videogame for Oculus Quest, developed with Unity Engine using only the basic Oculus Framework. Game awarded as best game and trailer of the course ( year 2021/2022).
See more
2D Priors for 3D human reconstruction
Potential of 2D Priors for Improving Robustness of single-image 3D Human Reconstruction, with CLIP supervision and Multimodal learning.
See more
Data Visualization Website
Market analysis of mobile apps dataset with interactive plots, with D3.js and Google Charts.
See more See website
ZO-AdaMM vs FO-AdaMM
Implementation and comparison of zero order vs first order method on the AdaMM (aka AMSGrad) optimizer: analysis of convergence rates and minima shape.
See more
Noise2Noise Lite
Noise2Noise is an AI denoiser trained with noisy images only. We implemented a ligther version which trains faster on smaller pictures without losing performance and an even simpler one where every low-level component was implemented from scratch, including a reimplementation of autograd.
See more
Insurance company & Piccardi Music
Two data analysis projects to practice Pandas, Data Wrangling, Data Vizualization, Regression, Observational Studies, Statistics and Supervised Learning ( w/ scikit-learn and statsmodels).
See more
Political party classifier from quotes
A Quotebank data analysis to build a political party classifier based on quotes, explained with a data story. Includes sentiment analysis, grammatical structure/complexity analysis and topic labeling.
See more See data story
Q-NIM
Reinforcement Learning: Q-Learning and Deep Q-Learning to train artificial agents that can play the famous game of Nim.
See more
HCGM and VuCondat for Semidefinite Programming
Semidefinite Programming with Homotopy Conditional Gradient Method (HCGM) and Vu-Condat methods for solving two problems: Fashion-MNIST classification using k-means clustering and geometric embedding for the Sparsest Cut Problem.
See more
RMSProp and AMSGrad for MNIST image classification
Implementation and comparison of SGD, SGD with momentum, RMSProp and AMSGrad optimizers on the Image classification task using MNIST dataset.
See more
CSI using Blind Image Deconvolution and Frank Wolfe
Blind Image Deconvolution and Frank-Wolfe's algorithm to deblur a license plate for Crime Scene Investigation (CSI).
See more
WGAN learns the distributon of a MoG
A Wasserstein Generative Adversarial Network that learns the distribution of a Mixture of Gaussian, using weight clipping or spectral normalization.
See more
Image Reconstructor
An Image Reconstructor that applies fast proximal gradient method (FISTA) to the wavelet transform of an image using L1 and Total Variation (TV) regularizations.
See more
FO & PROX
Implementation and brief comparison of different First Order and different Proximal gradient methods, comparison of their convergence rates.
See more
Higgs Boson Classifier using LHC CERN data
An AICrowd Challenge: Logistic Regression classifier that predicts whether an event's decay signature was the one of a Higgs Boson.
See more
Road Segmentation
An AICrowd Challenge: CNN classifier that predicts whether the pixels of an image represent a road or not.
See more
Master of Renaissance
An online multiplayer board game coded in java, playable both on a javaFX GUI or on CLI, with many server features to make it robust to connection problems.
See more
PoliMusic
A website that lets the user upload his songs on a server. It was made in two versions: Pure HTML (thin Client) and Thymeleaf Rich Internet Application (thick Client).
See more
Aerial Photography Simulation for photogrammetric flight design
Python software with GUI, capable of simulating an aerial photo given a pair DEM-Orthophoto (in GeoTIFF format).
See more
edU
Time (ms) and space (RAM usage) optimized command line text editor with multiple undo/redo feature.
See more
STEALTH
A simple command line arcade game, created in order to train with the basics of C programming after the first semester at my Bachelor's degree
See moreContact Me
