Learning with limited supervision
Friday 03/06/22, 2:00pm @ UCLouvain
Ismail Ben Ayed (ETS Montreal)
Abstract:
Despite their unprecedented performances when trained on large-scale labeled data, deep networks are seriously challenged when dealing with novel (unseen) tasks and/or limited labeled instances. This generalization challenge occurs in a breadth of real scenarios and applications. 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 very limited supervision. Specifically, I will discuss state-of-the-art models, which leverage unlabeled data with priors, and connect them under a unifying regularization 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 convoluted meta-learning strategies. Classical and simple losses, such as information-theoretic or graph regularization, well-established in clustering and semi-supervised learning, achieve outstanding performances.
References:
O. Veilleux, M. Boudiaf, P. Piantanida and I. Ben Ayed, Realistic Evaluation of Transductive Few-Shot Learning, Neural Information Processing Systems (NeurIPS), 2021
M. Boudiaf, I. M. Ziko, J. Rony, J. Dolz, P. Piantanida, I. Ben Ayed, Transductive information maximization for few-shot learning, Neural Information Processing Systems (NeurIPS), 2020
I. M. Ziko, J. Dolz, E. Granger and I. Ben Ayed, Laplacian regularized few-shot learning, International Conference on Machine Learning (ICML), 2020
H. Kervadec, J. Dolz, M. Tang, E. Granger, Y. Boykov, I. Ben Ayed, Constrained-CNN losses for weakly supervised segmentation, Medical Image Analysis (MedIA), 2019
M. Bateson, H. Lombaert, I. Ben Ayed, Souce-free domain adaptation for image segmentation, Medical Image Computing and Computer-Assisted Intervention (MICCAI), 2022
H. Kervadec, H. Bahig, L. Letourneau-Guillon, J. Dolz, I. Ben Ayed, Pixel-wise supervision for segmentation: A few global shape descriptors might be surprisingly good!, Medical Imaging with Deep Learning (MIDL), 2021
M. Tang, F. Perazzi, A. Djelouah, I. Ben Ayed, C. Schroers, Y. Boykov, On regularized losses for weakly supervised CNN segmentation, European Conference on Computer Vision (ECCV), 2018