Gaston Mazzei 26, Physicist, Software Developer
gastonmazzei95@gmail.com
I'm Gaston Mazzei. Please press 'More Info' or explore specific topics on the right by clicking on them. Thank you for your time!
About Myself
Born in southern South America and very Passionate about Nature, some of its details along with Humankind's mathematical models were formally learnt during a 6-year-long Bsc+Msc Physics degree. During that time several personal and academic projects were accomplished, and a little bit under 3 years of work experience was accumulated.
With a clear list of favourite topics, taking great pride in being philosophically aligned with the European Union's Ethics, and with a "Plan Quantique" scholarship to to study High-Performance-Computing and it's link with Quantum Computing via the Master in Parallel & Distributed Computing at the recently created Behemoth the Paris-Saclay University, my current place of residence is France.
Favourite topics
From the field of physics my favourite topics are statistical physics, (quantum) complex systems, and fluid dynamics. In all cases the inclusion of noise and the relationship with topology. In Computer Science, I find great joy in the techniques of High Performance Computing, Optimization, and Machine Learning.
Other interests
My personal interests include practising sports such as trekking or amateur Olympic-style weightlifting (and maybe beach-volley?); I'm quite an histrionic individual so I enjoy joking and playing music, and with friends spending time watching movies and/or cooking. I'm deeply passionate about animals, and I enjoy both classical music and specific composers from the German, French and Italian baroque.
C++ Developer (intern)
jan 2022 - May 2022
During the first four months of 2022 I am an intern at the VENICE (Virtual & Augmented Reality) Laboratory under the supervision of Dr. Christian Sandor.
Network Automation & Optimization Engineer
jan 2021 - aug 2021
As a Network Automation & Optimization Engineer at Iquall Networks my role was to both model networks using NoSQL databases and to design implement and test algorithms using Python and C++ in order to provide automated solutions to international telecommunication companies inside propietary software. The company is a local (made in Argentina) offerer of cost-effective flexible transparent responsive and low-latency software inside which telecommunication companies can easily scale their expanding infrastructure without increasing complexity.
Physics Machine Learning Developer (intern)
feb 2020 - dec 2020
During the almost 1-year-long research internship at ICAS (the International Center for Advanced Studies) several machine-learning topics were learnt from the renowed High Energy Physics researcher Dr. Ezequiel Alvarez. As part of the Institute's transfer of technology mandate as well as my own personal Master's thesis the website AI-Friendly.com was made: it consists of a free online customizable neural network.
Jr. Business Intelligence
nov 2018 - nov 2019
During the Part-Time Jr. B.I. job at an alpine-styled and medium-sized (about 100 employees and 4000 guests per month) four-star hotel in the Argentina's southern Patagonia region statistics was collected and executive reports were produced on a monthly basis. Vainilla sections of the reports included occupation and cash flow charts; more exotic ones were such as machine-learning-assisted proposals of target customers by identifying constant ratios between services consumed by guests inside the historical data. Finally in the context of the Hotel certifying its eco-friendliness while reducing its carbon footprint is that the latter was also computed considering the daily temperature and the monthly gas and electric consumption (extract of the report).
Microfluidic Technician (intern)
jun 2017 - jun 2018
Undergraduate research internship at CNEA-CAC (National Atomic Energy Comission) and
Network Diffusion and Synchronization Simulator
Efficiently study diffusion and synchronization
NEDISS v1 is a parallel (openMP) and distributed (MPI) simulator written in C++ to integrate ODE's of systems that interact via a graph. It is aimed towards sparse nonlinear ODE's that model systems that exhibit synchronization and in this case there is a 'visual-mode output' for which new equation plugins are super-easy to write (e.g. video for Kuramoto Eq. plugin). Users can add their own equations and/or graph by deriving from a templated class. Future plans include producing documentation and adding random-walks-based analysis of fractional diffusion.
Tracking Orchestra
convert movement to music
The Tracking Orchestra uses python to convert object's movement to music using OpenCV's non-AI-based multi-tracking capabilities. An example is available here. The instrument assigned to each can be changed. It is not 'canned' as it is not Dockerized but only requires being ran on Linux w/ Python3 and manually installing two library requirements.
AI-Friendly
artificial intelligence made friendly
AI-Friendly is my Fullstack Development flagship: it's a dynamic website built with Nginx and the open-source automation server called Jenkins that makes publicly-available for free a dense neural network that is trained over user-provided datasets in Microsoft Excel format. I developed it during my 'Software Development & Data Science Internship' at ICAS as per its state-sponsored transfer-learning mandate. In particular it has been used in highschool courses and private-company training such as 'AI for executives'.
Bitcoin Autoencoder
bitcoin automatic trading with a neural network
Open Source and dynamic website built with TensorflowJS and Flask that implements an automatic trading strategy based on a multi-thresholded strategy implemented in an neural network autoencoder's latent space. It's not considering transaction costs so it's a 'toy model'.
TicTacToe Neural Net
tictactoe against a neural net
Open Source and dynamic website built with TensorflowJS that uses a neural net to play tic tac toe. Intended for use in high-level ML courses. There is only ONE strategy that beats the neural net... can you find it?! For a verbose 'More info' students can explore the console output (in Chrome 'Ctrl+Shift+I'): if you wanna help improve this verbose explanation you are welcome via a pull request!
Covid Arg Fake News,
bayesian interpretation of covid data
Open Source and dynamic website built with Python's Flask that draws random examples of the absurd implications of the Argentinian's corrupted coronavirus database using bayesian statistics. Spanish only.
My Personal Page
the page you are looking at
The current site is a static open-source website hosted by GitHub and built with Ruby's Jekyll using Bulma. Please note that it is open source so you can use it for free if you inform me about it and don't mind your name appearing here.
Master M1 in Quantum Parallel and Distributed Computing
aug 2021 - apr 2022
Currently under the Plan Quantique's Quantum Scholarship to study Quantum and Distributed Computer Science at the Paris-Saclay University: a recently created Behemoth that ranks 14th in the 2021 Shangai Ranking (1st in Math; 9th in Physics).
MSc Physics
mar 2015 - feb 2021
Licenciatura en Ciencias Fisicas' is a 6-year degree; internationally equivalent to a MSc Physics. The University of Buenos Aires is a 200-year-old university that ranks 66 in the QS world ranking 2021 (transcripts: original & translated; grade: 8.4/10; the thesis defense is available on YouTube; mirror).
BSc Experimental Physics
mar 2015 - oct 2020
'Physics Research Assistant' is a degree awarded by the University of Buenos Aires in 2020. Internationally equivalent to a Bachelor in Experimental Physics; it is an intermediate degree for the MSc Physics degee and it requires taking extra experimental courses (transcript; grade: 8.2/10).
AI-Friendly.com: Artificial Intelligence Made Friendly
october 2021
Open source web application that allows high-level access to a customizable neural network for free. Published in the 34th Argentinian Symposium of Artificial Intelligence (ASAI - ISSN 1666-1079 p53-54) at the JAIIO 50 (50th Argentine Informatics and Operational Research Journal - link to proceedings and article).
Image Inpainting Applied to Art: Completing Escher’s Print Gallery
july 2021
Image Inpainting applied to M.C. Escher's Print Gallery at the LatinX in AI (LXAI) Research Workshop - part of ICML 2021's Affinity Workshops (link to JLXAI journal - Id 19 - also available on Arxiv).
Delta Hedging with Transaction Costs: Dynamic Multi-Scale Strategy using Neural Nets
may 2021
A portfolio strategy is parametrized by hedging at different periods and computing the weights with a convolutional 1D neural network (published in MACI VIII; the eight journal of the Argentinian Society of Applied Mathematics; ISSN: 2314-3282 pp. 459-462 - also available on Arxiv).
Option Pricing Model with Transaction Costs
may 2017
Numerical simulations of the Black Scholes PDE with non-differentiable transaction costs and the analytical solution for exponential costs (published in MACI VI; the sixth journal of the Argentinian Society of Applied Mathematics; ISSN: 2314-3282 pp. 569-573 - also available on Arxiv).
TOEIC English Language Certificate
Dec 2021
English language evaluation at a paper-based TOEIC examination (results pending).
LXAI Sponsorship for NVIDIA CUDA C++ Course
Nov 2021
Introductory synchronous course on CUDA C++ programming taught by NVIDIA Solutions Architect Paul Graham and sponsored by the LXAI (Latins in A.I.) organization [certificate]
Plan Quantique Scholarship
Aug 2021 - Apr 2022
Won the Plan Quantique scholarship for a period of one year to study quantum and high performance computing in France [proof].
IELTS English Language Certificate
Dec 2020
IELTS UKVI (score: 8.0/9.0 - verification code).
Periodical evaluations at Workera.ai
2020-2021
Periodical measurements of AI & CS knowledge as a member of the Stanford-related community called workera (certificate)
Best Undergraduate Paper Award
2017
Won the Best Undergraduate Paper award at ASAMACI's 6th Congress for the work 'Option Pricing Model with Transaction Costs' (ISSN: 2314-3282 pp. 569-573; congress link; Arxiv; -proof of the certificate is pending-)
Characterization of AI Detector over Kernel Trace
a natural metric for reverse-engineerability
Program written in Bash and Python to (1) collect user's input of what they are doing at the PC and (2) train a neural network to classify the different tasks by watching at the (Linux) kernel trace
Distributed Ring Election Simulation
distributed algorithm simulation
Python simulation showing the communication complexity as a function of the number of nodes for the LCR leader election distributed algorithm (link)
Synchronous Distributed Algorithms Simulation
leader election - breadth-first tree - routing problem
Numerical simulation recovering the predicted communication complexity for several synchronous distributed algorithms in graphs sampled from the subset of Erdos Renyi graphs that are connected (link)
Fluid Mechanics Data Analysis
analyzing resonance
Program analyzing the resonance effect of a 2020 viral toad video (repository - video)
Neural Net Learns Algorithms
neural net sorts numerical arrays
Jupyter notebook written in Python showing how a neural network can successfully learn to sort numerical arrays with +90% accuracy (link)
Simulating Seasonality's Effect on Covid Crash
AI + modelling
The start of the Covid Crisis coincided with the months where USA unemployment flexibility historically peaks. This script uses Facebook Research Prophet's seasonal functionalities to model how could the Covid Crash had affected the unemployment if it had ocurred at another point of the year (repository)
Physics simulation: Tic Tac Toe
dense neural net
Tic Tac Toe: the optimum markovian strategy was learnt by a two-layer fully-connected neural network as an example of ai-friendly.com 's potential applications to use it in both my own Master's thesis and colleagues' highschool-level AI and Statistics courses.
Physics simulation: Quantum Tunneling
dense neural net
Quantum Tunneling: the transmission coefficient is fitted by a shallow fully-connected neural network that learns from binary observations identified with the collapse of the quantum wave-function. It was an example of ai-friendly.com 's potential applications to use it in my Master's thesis.
Physics simulation: Exoplanets
dense neural net
Exoplanets: Kepler's third law is learnt by a shallow fully-connected neural network and thus the orbital period can be estimated from the radius and the solar mass. It was an example of ai-friendly.com 's potential applications to use it in both my own Master's thesis and colleagues' highschool-level AI and Statistics courses.
Physics simulation: Group Theory
dense neural net
Group Theory: detection of Lie algebras with a total number of different irreductible representations in a product descomposition that exceed a threshold. It was used as an example of ai-friendly.com 's potential applications in my Master's thesis.
Physics simulation: The Ising Model
dense neural net
The Ising Model with a magnetic field: classification of parameters that exceed a given threshold for the Energy and the Magnetizaion. It was used as an example of ai-friendly.com 's potential applications in my Master's thesis.
Physics simulation: Nonlinear Equations
dense neural net
Quadratic two-dimensional homogeneous ODEs: detection of unstable equilibria as an example of ai-friendly.com 's potential applications to display in my Master's thesis.
Physics simulation: Higgs Boson
dense neural net
Detection of the Higgs Boson in LHC data: the Higgs Boson is detected at a concentration 1% in LHC collisions in format LHCO by a shallow fully-connected neural network. It was used as an example of ai-friendly.com 's potential applications in my Master's thesis.
Physics simulation: Electrical Circuits
dense neural net
Circuits' max power: detection of pasive-components' values that lead to currents above a certain threshold for diode-perturbed common circuits. It was an example of ai-friendly.com 's potential applications to use it in both my own Master's thesis and colleagues' highschool-level AI and Statistics courses.
AI vs Humans
random forests classifier
A Musical Experiment: 200 humans were surveyed to measure the average human accuracy at recognizing classical music authors in between 8 options from short sound samples. Different AI-based-classifiers were measured against that benchmark and the conclusion was that a Random Forests Ensemble model performed with an average 3x accuracy than of humans.
Fundamentals of Accelerated Computing with CUDA C/C++
november 2021
Introductory synchronous course on CUDA C++ programming taught by NVIDIA Solutions Architect Paul Graham and sponsored by the LXAI (Latins in A.I.) organization [certificate].
TALENT
july 2021
TALENT (Training in Advanced Low Energy Nuclear Theory) is a 2-week summer school held virtually and hosted by the European Centre for Theoretical Studies in Nuclear Physics and Related Areas; it covered both pure machine learning and pure low energy nuclear physics concepts as well as covering several intersections of both areas (course repo and proof [1][2] of assistance).
Geophysical Fluid Dynamics in FORTRAN
july 2021
An introductory course to FORTRAN programming with emphasis in parallelization and fluid mechanics libraries. The course was provided by the French-Argentine Institute of Climate Studies (IFAECI) and the Department of Oceanic and Atmospheric Sciences of the University of Buenos Aires (DCAO-UBA) (codes).
Machine Learning
2020
'Machine Learning' was a single-course at the Saint Martin National University (UNSAM) dictated by astrophysics researcher Dr. Rodrigo Diaz (proof: academic transcript; grade: 8/10).
Mathematical Modelling & Analysis
july 2016 - july 2018
Mathematical Modelling and Data Analysis courses taken at the Department of Mathematics and Institute of Computing of the University of Buenos Aires
1) Exploratory Data Analysis (grade: 10/10)
2) Quantitative Analysis in Finances (grade: 10/10)
3) Mathematical Modelling of Complex Social Systems.
ASAI - JAIIO 50
october 2021
Presented two works at ASAI: the Argentinian Symposium of Artificial Intelligence. The first one was a work-in-progress (sometimes in this web called 'the Escher Project') that was published in an ICML 2021 workshop (1) and the second one was a paper about AI-Friendly that was itself published in this journal (2). The videos are in Spanish.
ML Collective Research Jam 2
june 2021
Presented AI-Friendly in a research jam organized by ML Collective (an independent nonprofit organization that supports open collaboration in machine learning research)
LXAI Social @ ICLR
may 2021
Presented the Escher Inpaint Project as part of 'ML Techniques for Art Reconstruction' in the pitch for research projects applying for mentorship at ICLR Social Virtual Room n4416 inside LXAI's event LatinX in AI Social at ICLR 2021 (presentation link; and further proof)
MACI VIII
may 2021
Presented the recent publication 'Delta Hedging with Transaction Costs: Dynamic Multi-Scale Strategy using Neural Nets' at the MACI VIII virtual sessions (proof - link to presentation)
ML Collective Research Jam 1
april 2021
Presented the Escher Inpaint Project in a research jam organized by ML Collective (an independent nonprofit organization that supports open collaboration in machine learning research)
22nd Giambiagi Winter School: Artificial Intelligence and Deep Learning in Physics
nov 2020
Presented a virtual poster about ai-friendly.com at the 22nd Giambiagi Winter School (link to conference presentation)
MACI VI
may 2017
Presented the paper 'Option Pricing Model with Transaction Costs' at MACI VI's Quantitative Finance Session (poster)
DF-UBA
may 2017
Presented the results of the calibration and characterization of a standing-wave optical resonator Nd:YAG laser at the Department of Physics inside the Buenos Aires University (poster)
Google Form Randomizer
decrease bounce & increase info
JavaScript Architecture for GoogleScripts that allows to downsize Google-Form surveys by randomizing the shown questions and thus allows for simpler forms with more total questions. Further info is available at this Medium series [1][2].
Darkest & Lightest
mini-Tool for front-end design
Darkest & Lightest is a mini Python script that reports the HEX value of a picture's darkest and lightest values. I have used it as a tool while choosing colors during FrontEnd Design.
LHCO to CSV
Convert LHCO format to CSV
LHCO to CSV is a python code for converting cumbersome .LHCO files into comfortable .csv ones. It was written during the LHC Olympics during 2020 in an attempt to provide quick preprocessing for AI-focused colleages.
Telegram Geo-Recommender Bot
respond to users w/ nearest market
Open Source Python Application for a Telegram Chatbot that indicates the nearest market based on the location. It responds to user input by looking info in a pre-supplied database.
Peat
Python wrapper for CPU constraints
Peat is intended as a Python script wrapper that limits both CPU and memory usage at the cost of calling Docker. It used an image with MKL optimizations for Intel processors but that image may disappear from DockerHub in the future. Memory constraints do not work as the patch for docker-compose memory bug has not been applied; so at the end of the day it is currently only a CPU constrainer.
NYU Deep Learning Translator
2020
Translated from English to Spanish a lecture over language models at Yann LeCun's Deep Learning course at NYU (proof).
Rock-guitar Videoclips in Spanish
2019
Created 13 animated musical videoclips of original rock-guitar music amateurly sang in Spanish by myself (link).
Self-watering vertical garden
2018
A self-watering vertical flower pot was designed and a prototype was manufactured (presentation).
J.S.Bach Lutheran Poem Harmonization Mix
2018
YouTube medley of post-reformation Lutheran poems harmonized by J.S.Bach with score and lyrics (link).
Classical Music on Nylon Guitar
2017-2018
YouTube playlist with short instrumental works (nylon string and rich in counterpoint - link).
Electronic Music
2015-2017
Short Electronic Music clips (link).
Instrumental Music
2011-2015
Classic & Rock songs (link).
Battle of the Bands
2013
As a guitarist of Xibalba I played an original song (Robulo) and a cover (Peaches in Regalia - Frank Zappa) at the Battle of the Bands at the EMBA (Buenos Aires Music School) sponsored by the Lara Studio
Band Videoclip
2012
As a guitarist of Xibalba I played the guitar in the videoclip of our original song (Chromatique)
Big Live Concert
2012
As a guitarist of Xibalba I played at the City Bar Concert (+200 people)
Collective Tribute
2010
Participated in a Spanish guitar community's musical video tribute after the passing away of the rock figure Ronnie James Dio (link).
