As a French PhD student, I am passionate to whatever comes close to Artificial Intelligence & Earth Observation. Whether it is theoretical content with exploring state-of-the-art models or more concrete applicative programming with Jupyter Notebooks, I always find myself curious about what the world is up to !
Additionally, I am currently exploring the depth of SAR imagery, understanding its underlying concepts such as Polarimetric SAR or Interfometry.
You may also explore the personal projects I find myself doing on my free time hosted on this website. I have focused on exploring different facets of Deep Learning with applications in Multimodal Learning, Computer Vision or Automatic Feature extraction for Remote Sensing applications. All these projects’ repositories are hosted on Github.
PhD in Applied Physics, 2020-2023
SONDRA Laboratory, CentraleSupélec, Gif-sur-Yvette, France
MSc in Artificial Intelligence & Multimodal Interaction, with Distinction, 2019-2020
Heriot-Watt University, Edinburgh, Scotland
Engineering degree in Computer Science, 2017-2020
EISTI, Cergy, France
BSc degree in Computer Science, 2015-2018
Cergy-Pontoise University, Cergy, France
Working on problematics of change detection in SAR Time Series of forests with the help of Deep Learning methods.
SONDRA Laboratory is a laboratory mixing 4 entities: French ONERA, Supélec (known today as CentraleSupélec), the National University of Singapore and the DSO of Singapore.
As a deep learning Intern, I have trained, tuned and tested a model capable of doing building segmentation using satellite imagery. In this internship, I have tried multiple models (Mask RCNN, UNet, Deep UNet) and tried to take the best out of them all. The code was written using Keras with Tensorflow Back-End and was manipulated using a web-based RESTFul GUI with Flask and HTML5 technologies.
Also, multiple postprocessing technologies were considered and tried such as Logistic Regression (using scikit-learn) or Conditional Random Field (using pycrf).
State of the art models for Similarity Learning are all based on Deep Learning architecture using Siamese Network [Gregory et al., 2015]. They define a feature extraction pipeline that creates a latent representation of input data. This embedding vector is semantically highly descriptive and can be used for the computation of distances between data records to measure similarity. While similarity learning is a popular topic, the combination of multiple modalities has not yet been attracted most of the attention of the field. In the context of Duplicate Product Identification, both the textual description of the product and their pictures can be used to make the similarity decision. This context of using data descriptors, e.g images and text, of different modalities require to rethink the concept of Siamese Network to perform multimodal similarity learning. In this work, multiple approaches have been explored: unimodal & multimodal Similarity Learning algorithms. The latter, combining embeddings across multiple modalities through gradient sharing method was proven to outperform any other combination of unimodal approaches through the use of N-Way & F-Beta Scoring. Furthermore, our analysis of the impact on the learnt features of the combination of multiple modalities has given insights on how they can collaborate to optimize the training function by detecting the most informative features for each modality. Comparing the weights of a multimodal siamese network to unimodal network helped to better evaluate cross-modality data profiles captured within the embeddings.