laferraille Goto Github PK

cesal_scraper

Personal project that automates the process of checking room availability at Cesal Residential by scraping the Cesal website.

eeg_signals_prediction

End-to-end Machine Learning project to predict EEG signals for automated sleep staging. The methodology used achieved a promising F1 score of 67.5% while putting us at the top 15% of the leaderboard of the Kaggle competition.

efficient-fine-tuning-techniques-for-deep-learning-models

This repo aims to explore efficient fine-tuning techniques for deep learning models, focusing on reducing the computational and environmental costs associated with training large-scale models. We emphasize methods like Low-Rank Adaptation (LoRA), VeRA, and Quantization, which aim to enhance the fine-tuning process for foundational models.

face-recognition

Personal project on face recognition using OpenCV and face-recognition packages. This aims to understand better on faces are detected on an image/video.

gaia_mistral

graph-based-segmentation-for-car-detection-in-mario-kart

Explore graph-based algorithms for car detection in Mario Kart images and videos with tree-based segmentation and Max-Flow/Min-Cut

laferraille

link-prediction-in-an-actor-co-occurrence-network

Predict the existence or non-existence of links between the nodes of an actor graph using ML and a Graph Auto-Encoder

linkedin_scraper

This project provides a straightforward notebook designed for scraping LinkedIn to collect basic user information. The code is versatile, allowing for easy modifications to cater to specific needs or use-cases.

liver-cancer-detection

Multivariate Data Analysis of a cohort of patients with liver cancer subtypes. Prediction task leveraging dimensionality reduction techniques and multiway tensor analysis.

overcome-over-smoothing-in-graph-neural-networks

Evaluate oversmoothing in graph neural networks (GNNs) focusing specifically on two methods: AdaEdge and MADReg.

vision-transformers

This repository is dedicated to exploring the Vision Transformer (ViT) architecture through a series of comprehensive experiments. It provides insightful demonstrations using straightforward implementations on popular datasets like MNIST and CIFAR, and a on a more challenging scenario which is ovarian-subtype classification.