Welcome to Arthur Evain's Portfolio

Implemented various Intrusion Detection System (IDS) models on CAN buses based on AI architectures (LSTM, autoencoder, BERT, CNN, adversarial networks). The IDS was subsequently deployed on an Alstef Automated Guided Vehicle (AGV).

Four-person team project focusing on unsupervised classification of an imbalanced dataset of abstract paintings. We extracted geometric features alongside CNN and autoencoder embeddings, then identified clusters using the K-means algorithm. The results were subsequently refined through empirical analysis.

Kaggle competition where I ranked 9th out of 77 participants. The objective was to train a model to classify white blood cell images across 13 classes from a highly imbalanced dataset. My approach combined classical image segmentation methods (non-ML) with fine-tuning a pre-trained neural network using handcrafted extracted features.

Implementation of a patch-based image inpainting algorithm. Below is an application example showing the seamless removal of a horse from the original photograph.

First-year final project conducted in collaboration with a quantum computing researcher. We implemented state-of-the-art cryptography algorithms to verify their theoretical correctness and evaluate their robustness against hardware noise on real quantum machines. We executed experiments on IBM Quantum hardware via API. Below is a quantum circuit generated and tested using the Qiskit library.

First-year team project (5 members). We designed and built an autonomous car powered by a Raspberry Pi and a camera module. After designing the chassis in CAD, we developed autonomous navigation algorithms to detect and reach target beacons on a grid. Our team achieved the best overall performance in our class cohort.

First-year team project (4 members). We developed a C to x86-64 assembly compiler. OCaml libraries were used for lexical analysis and parsing to generate an intermediate JSON abstract syntax tree. A Python backend then interprets this JSON file to emit functional assembly code matching the original C program behavior. Supported features include pointers, local/global variables, arithmetic operations, arrays, functions, printf/scanf, conditional statements (if/else), while loops, and dynamic memory allocation (malloc).

Preparatory class research project (MPI*) presented for French Grandes Écoles oral entrance exams. The study focused on applying the NEAT (NeuroEvolution of Augmenting Topologies) genetic algorithm to autonomous driving. Neural networks received distance measurements from 5 directional wall sensors and outputted steering and acceleration controls. Built in Python, the evolutionary simulation converged toward agents capable of successfully completing complex tracks (shown below).

Built a Python trading application capable of streaming market data and computing various technical indicators. While I initially planned to implement and backtest automated trading strategies based on these indicators, I set the project aside after questioning the long-term viability of standard technical analysis. The codebase is no longer actively maintained but remains available below.