Erwan DAVID

Research internship in Aalto, Finland

For 3 months, from June to August 2024, I did a research internship in AI with genetic algorithms at the Computer Science Department at Aalto University in Helsinki.

I used Python to simulate the natural evolution of a hostile ecosystem. Individuals are controlled by artificial neural networks and evolve through genetic mutations. They must learn to develop individual and group strategies to survive in this environment.
This research project is the continuation of a study project I started during my first 2 years at engineering school.

Internship at Sopra Steria Next

At the end of my 1st year at Arts et Métiers, I did a 4 weeks engineering internship at Sopra Steria Next, an IT consulting firm.
As part of a project to visualize road accidents for the Ministry of the Interior, I programmed in Python:

• an algorithm for computing surface density (KDE)
• a clustering algorithm (DBSCAN) with the SKlearn library

Natural evolution of an ecosystem - Python - IA - Vanilla

As part of a group project in artificial intelligence using Python I simulated the natural evolution of individuals in a hostile environment.
Each individual, driven by a dynamic neural network, must understand for himself that eating plants is the only way to survive. Neural networks evolve genetically by transfer and mutation of weights and neurons.

This project is entirely Vanilla except for the 5 libraries: Numpy, Random, Matplotlib, OpenCV and Pyqtree.

Random generation of an environment - Python - Vanilla

This project carried out under Python deals with the procedural generation of complex environments.
I've coded a program that generates 3D environments in an entirely random but coherent way. Each region of the map is defined by a biome (or type of environment) depending on its temperature and average precipitation rate. The local altitude is then correctly adapted.
Many mathematical concepts are used: Voronoi diagram, Lennard-Jonnes potential, Perlin noise, histogram equalization, Bezier curves or even Gauss masks.

This project is entirely Vanilla except for the 3 libraries: Numpy, Random, Matplotlib.

Learning a labyrinth by reinforcement - Python - IA - Keras

This project carried out under Python is about artificial intelligence and more particularly reinforcement learning.
I have coded here a machine learning algorithm that allows an agent to learn to solve a maze by himself.
I realized up to 3 different types of reinforcement algorithms:

• I created a mono-agent learning algorithm with a Q-table
• I created a mono-agent deep learning algorithm (containing a neural network)
• I created a multi-agent learning algorithm with a Q-table per agent

This project was carried out using 4 libraries: Keras, Numpy, Random, Matplotlib.

Procedural generation of a maze - Python - Vanilla

This project carried out under Python touches on the procedural generation of a labyrinth.
I have coded a program that generates an entirely different maze with each new call. An obvious condition is that the entrance and exit must be connected by a path in order for the maze to be solvable.
This project was essentially divided into 2 distinct parts:

• Creation of a maze with given dimensions
• Numerical resolution of the maze by Lee's method

This project is entirely Vanilla except for the 3 libraries: Numpy, Random, Matplotlib.

Image recognition - Python - IA - Keras - Arts et Métiers

As part of a group project in artificial intelligence under Python:

• I created neural networks with Keras to classify images from a clothing image database
• I reduced their prediction uncertainties by modifying the parameters and hyperparameters and then by aggregating different networks (ensemblist method)

This project was carried out using 3 libraries: Keras, Numpy, Matplotlib.

TIPE Error detection - Python - Vanilla - Competitive entrance exams to Grandes Ecoles

As part of a group project to optimize tipping bucket rain gauges :

• I made 3 Python programs to detect errors in the measurement of rain gauges
• I designed in SolidWorks and 3D printed an anti-debris filter

This project is entirely Vanilla except for the 2 libraries: Numpy, Matplotlib.

Help Out - volunteer

I participate in outreach activities to help the homeless, providing them with hot drinks, blankets, a listening ear, and kind words. The outreach takes place during the cold season, mainly in autumn and winter.

ARPEJ 78 - Volunteer trainer

I've worked with high-school and middle-school students with academic difficulties. I helped them review their lessons and do their homework in Maths, Physics, English and History-Geography. My aim was to give them self-confidence and to develop their autonomy

La main à la pâte - Volunteer trainer

Thanks to my background, I was able to help school and college teachers to teach and inspire their students to love science and technology. My aim was to develop their curiosity and critical thinking skills by leading classes or practical sessions.