MEDISCHARGE: Automated Clinical Discharge Summaries
GitHub Repository ACL Paper EPFL-MAKE at โDischarge Me!โ: An LLM System for Automatically Generating Discharge Summaries of Clinical Electronic Health Records Project Summary This project is focused on the automation of generating discharge summaries from clinical Electronic Health Records (EHRs) using Large Language Models (LLMs). The system, MEDISCHARGE, is designed to generate two main sections of a discharge summary: Brief Hospital Course (BHC) and Discharge Instructions (DI). Built on the Meditron-7B model, it leverages a context window extension and a dynamic information selection framework to handle large EHRs and efficiently select the most important content when needed. ...
Disaster Change Captioning
Report Website Project Summary This project revolves around the task of disaster change detection using remote sensing imagery, particularly from satellite sources. The objective is to create a comprehensive dataset and develop models to detect and analyze changes between pre- and post-disaster images. The dataset includes over 60 disaster events, along with pixel-level annotations of disaster-specific changes. ...
Screen Tastes: The User-Critic Divide in Cinema
Website During the course Applied Data Analysis (CS-401) at EPFL, we had to work on a data analysis project on one of the proposed datasets. With my team (Baptiste Lecoeur, Enzo Palmisano, Jamil Maj & Mariella Daghfal) we chose to work on the CMU Movie Summary Corpus dataset. ...
Occupancy FLEP
Github Backend Github Frontend occupancy.flep.ch Introduction As a masterโs student in Data Science at EPFL, Lausanne, I often found myself in a common predicament faced by many of my peers: the search for a quiet, unoccupied room for studying or group discussions. This task, seemingly simple, was surprisingly challenging due to the lack of accessible and comprehensive information about room availability on campus. ...
Analyzing EPFL Restaurant Menus
Analyzing EPFL Restaurant Menus from Data Scraping to Network Visualization using Python and D3.js, while performing Exploratory Data Analysis (EDA) and Automation with GitHub Actions.
GitHub Actions Scraping
I recently had to run a scraping script on a schedule in order to retrieve news articles from daily RSS Feeds (see more). I decided to use GitHub Actions to do so. This post is a quick tutorial on how to do it. ...
EPFL GraphSociative
GitHub Live Demo Have you ever wondered about the intricate connections within EPFLโs associations? How do associations interconnect? How many accreditations does individuals have? ...
RSS Trends
GitHub Kaggle Dataset Medium Live Demo Inspired by and curious about Google News articles grouping by event I challenged myself into replicating its state of the art. ...
Fine-tuning T5 for Q/A on EPFL Courses Materials
Report Introduction The project, undertaken with Delabarre Luca and Nemo Fabrice, aimed to train a large language model (LLM) for assisting students in understanding EPFL course materials. This initiative sought to provide quick, detailed answers without human interaction, thereby streamlining the educational process. ...
Deep Q-Learning for Epidemic Mitigation
GitHub Repository Report Introduction In the Spring Semester of 2023 we embarked, with Romain Birling, on a mini-project as part of the Artificial Neural Networks/Reinforcement Learning course at EPFL. Our focus was on Deep Q-learning for Epidemic Mitigation, aiming to study the impact of different policies on epidemic control. ...