Special session on Learning with Few Annotations
Frenay Benoit
Special session on Learning with Few Annotations
In many application areas, deep learning (DL) models constitute effective solutions, but their development still requires vast amounts of annotated data. In many contexts, due either to the annotation process or to the expertise required to generate high-quality labels, annotated data are scarce, noisy, and sometimes even absent. The design of DL methods that are less dependent or robust to variable amounts and quality of annotations is therefore one of the hottest topics in DL. The goal of this special session on learning with few annotations is to gather (i) researchers who aim to alleviate the dependence of DL models on annotations, as well as (ii) researchers working in application domains where annotations are scarce. Special attention should be given to challenging datasets, such as domains with high imbalance between classes, a very large number of classes, temporal data, etc.
Topics of interest include (but are not limited to):
- Unsupervised, self-supervised, semi-supervised, and supervised representation learning
- Weakly supervised learning (noisy, incomplete, or imprecise labels) and learning with noisy labels
- Contrastive representation learning
- Metric learning
- Transfer learning
- Meta-learning
- Prototypical, dictionary and subspace learning
- Few-shot and zero-shot learning
- Foundation models and vision-language models for label-efficient learning
- Pseudo-labelling methods (self-training / co-training / consistency regularization, ...)
- Multiple instance learning, positive unlabeled learning
- Synthetic data generation, data augmentation
- Datasets and benchmarks
- Active, human-in-the-loop learning, efficient annotation
- Efficient fine-tuning, domain adaptation, domain generalization
- Applications
Important Dates:
Submissions Open : Friday, September 11, 2026
Paper Submission Deadline : Wednesday, November 18, 2026
Author Notification : Friday, January 22, 2027
Conference Dates : 21–23 April 2027, Bruges, Belgium
Submission:
Submissions to special sessions follow the same review procedure, format and submission rules as regular ESANN papers (single-blind, 6 pages).
Authors should indicate the corresponding special session when submitting through https://www.esann.org/.
Organizers:
Lorenzo Baraldi : University of Modena and Reggio Emilia, Italy
Benoît Frenay : University of Namur, Belgium
Diana Mateus : Centrale Nantes, France
Ariel Basso : University of Namur, Belgium
Pierre Poitier : University of Namur, Belgium