Heure: 13h30
Résumé: Despite their unprecedented performances when trained on large-scale labeled data, deep learning and artificial intelligence models are seriously challenged when dealing with novel (unseen) classes and limited labeled instances. In contrast, humans can learn new tasks easily from a handful of examples, by leveraging prior experience and context. Few-shot learning attempts to bridge this gap, and has recently triggered substantial research efforts. This talk discusses recent developments in the general, wide-interest subject of learning with limited supervision. Specifically, I will discuss state-of-the-art models, which leverage unlabeled data with structural priors, and connect them under a unifying information-theoretic perspective. Furthermore, I will highlight recent results, which point to important limitations of the standard few-shot benchmarks, and question the progress made by an abundant recent few-shot literature, mostly based on complex meta-learning strategies. Classical and simple loss functions, such as the Shannon entropy or Laplacian regularization, well-established in the clustering literature, achieve outstanding performances.
Biographie: Ismail Ben Ayed is a Full Professor at the ETS Montreal, where he holds a research Chair on Artificial Intelligence in Medical Imaging. He is also affiliated with the University of Montreal Hospital Research Centre (CRCHUM). His interests are in computer vision, optimization, machine learning and medical image analysis algorithms. Ismail authored over 100 fully peer-reviewed papers, mostly published in the top venues of those areas, along with 2 books and 7 US patents. In the recent years, he gave over 30 invited talks, including 5 tutorials at flagship conferences (MICCAI’14, ISBI’16, MICCAI’19, MICCAI’20 and MICCAI’21). His research has been covered in several visible media outlets, such as Radio Canada (CBC), Quebec Science Magazine and Canal du Savoir. His research team received several recent distinctions, such as the MIDL’21 best paper award and several top-ranking positions in internationally visible contests. Ismail served as Program Committee for MICCAI’15, MICCAI’17 and MICCAI’19, and as Program Chair for MIDL’20. Also, he serves regularly as reviewer for the main scientific journals of his field, and was selected several times among the top reviewers of prestigious conferences (such as CVPR’21, NeurIPS’20 and CVPR’15).
Note: La présentation sera donnée en français.