Designed and trained a model for the automated detection of R-waves in highly noisy electrocardiograms (ECGs).Obtaining images from Cardiovascular magnetic resonance imaging (MRI) necessitates synchronization with the ECG. However, ECGs are often afflicted with noise, stemming from both the magnetic field of the MRI and the pathologies of the patients, complicating R-wave detection. To address this challenge, we introduced artificial noise into our training dataset, which originally comprised clear ECGs and we evaluated the performance of our model using highly noisy ECG data provided by Dijon Hospital.I employed models designed to process the ECG through its time-frequency representation. Then, I leveraged a model that combines information from both the time-frequency representation and the time series data of ECGs.

Designed and trained more efficient neural network architectures for audiovisual speech generative modeling using variational autoencoders. (Inria MULTISPEECH team at the Laboratoire Lorrain de Recherche en Informatique et ses Applications (LORIA).)Speech perception involves both the acoustic modality and the visual modality, specifically the lip movements of the speaker. We focused on variational autoencoders (VAEs), comprising an encoder to encode the input signal using both speech modalities into a latent representation, and a decoder to reconstruct the input signal from the latent representation.I delved into exploring models with attention mechanisms to capture temporal dependencies within the signal.

Worked on automatic sorting machines for assembly solutions in the automotive industry.Observed challenges in applying computer vision on a production line.