I am currently a master student at ENS Paris-Saclay (M2 MVA) and ENSAE Paris (Engineering cycle). My academic journey and projects center around statistics and machine learning, with a particular focus on deep learning in recent years.
-
Moving Mahalanobis (Report in French): A method to perform Anomaly Detection on unimodel time series by leveraging a pre-trained model. Online setup enabling continuous inference without any model training.
-
Clinical Trial Duration: An econometric project investigating factors prolonging clinical trials. Leveraged NLP techniques and multiple variables to extract features from clinical trial descriptions.
-
Modern Bert for Clinical Trial: An application and analysis of ModernBert for clinical trial named entity recognition
-
Online Stochastic Matching: Report and comparison of different methods for online stochastic matching
-
LSTM For Portfolio Allocation: Use of Recurrent-NN Architecture to handle constrained portfolio allocation.
-
LongRoPE, Ding & Al. 2024: A positional encoder to manage high context length (1M+)
-
Pseudo Inverse Guided Diffusion Models for inverse problems, Song & Al. 2023: A Problem Agnostic Method to solve inverse problems using diffusion models
-
LiTE Pytorch, M. Devanne, M. Ismail-Fawaz & Al. 2023: A pytorch implementation of the model LiTE for time series classification.
-
Detecting changes in slope with an L0 penalty, Maidstone & Al. 2019: Changepoint detection with l0 penalty by using dynamic programming
-
Prototypical Networks for Few-shot Learning, Snell & Al. 2017: A neural network designed to handle few shot learning by "learning to separate"
-
Mixture Model for Random Graphs, Daudin & Al. 2006: A statistical model to perform clustering on Unoriented graphs.
-
Video Models for deepfake detection: Work conducted @ Entrust to use video models and training strategy to perform deep fake detection with strong generalisation and fair results
-
GAN Models for financial synthetic data: Work conducted with HSBC to study the potential of generated time series using GAN-models
