I am a Machine Learning engineer. I recently finished my PhD studies at École Polytechnique Fédérale de Lausanne in the INDY Lab, advised by Prof. Patrick Thiran. I am interested in the generalization ability of deep neural networks with a particular focus on classification tasks with limited and/or noisy-labeled datasets. In my research, I focus on finding techniques that require no access to ground truth labels while being informative about the model generalization and the dataset itself.
News
- July 2023: I successfully defended my PhD thesis titled “Deep Learning Generalization with Limited and Noisy Labels”. You can find it here.
- January 2023: Our paper titled “Leveraging Unlabeled Data to Track Memorization” is accepted at ICLR 2023. Check out this twitter thread to learn more about it.
Publications
Leveraging Unlabeled Data to Track Memorization
Mahsa Forouzesh, Hanie Sedghi, Patrick Thiran
ICLR 2023, ArXiv, OpenReview, Github
Disparity Between Batches as a Signal for Early Stopping
Mahsa Forouzesh, Patrick Thiran
ECML/PKDD 2021, ArXiv, Published Version, Github
Generalization Comparison of Deep Neural Networks via Output Sensitivity
Mahsa Forouzesh, Farnood Salehi, Patrick Thiran
ICPR 2020, Oral Presentation, ArXiv, Published Version, Github
Pre-prints
Differences Between Hard and Noisy-labeled Samples: An Empirical Study
Mahsa Forouzesh, Patrick Thiran