Sébastien Kéroack

Personal projects

  • This project is the source code of my multilingual portfolio, designed for any PHP/Apache-compatible server.
  • Key features include dynamic language loading, adaptive theming (light/dark mode), meta tag optimization for social media, form security with reCAPTCHA v2, and a structured architecture for efficient handling of static and dynamic content.
High-Level Overview Diagram

Neural Architecture Search

2021 - 2023

  • This project generates custom neural architectures, tailored to specific goals, using a controller trained through reinforcement learning.
  • The agents are trained on clusters, enabling large-scale data collection. This data is used to refine the controller, which learns to produce increasingly robust architectures to achieve the desired objectives.
High-Level Overview Diagram

Forecasting Foreign Exchange Market Movements through Economic News Analysis

2020 - 2021

  • This project enables sentiment analysis of economic news using a language model (LLM).
  • An agent retrieves and processes economic news, which is then stored in a database. A web interface allows for manual annotation of this data to enrich the training dataset. A pre-trained LLM is then fine-tuned.
High-Level Overview Diagram
  • This project is a software library dedicated to deep learning, developed in C++ and CUDA. It enables the design, training, and deployment of neural networks.
  • It supports both supervised and unsupervised training, as well as advanced techniques for regularization, normalization, and weight optimization, inspired by scientific publications.
High-Level Overview Diagram