Learning from Scarce Data 2022 : CBMI Special Session: Learning from scarce data challenges in the media domain

CALL FOR PAPERS:
Special session on "Learning from scarce data challenges in the media domain"
(in conjunction with CBMI 2022, September 14-16, Graz, Austria)

Website: https://cbmi2022.org/call-for-special-session-papers/
Contact: Hannes Fassold, JOANNEUM RESEARCH, [email protected]
Paper deadline: April 10, 2022

Deep learning-based algorithms for multimedia content analysis need a large amount of annotated data for effective training, e.g., for image classification on the ImageNet dataset, each class comprises several thousand annotated samples. Having a dataset of insufficient size for training usually leads to a model which is prone to overfitting and performs poorly in practice. But in many real-world applications in multimedia content analysis, it is not possible or not viable to gather and annotate such a large training data. This may be due to the prohibitive cost of human annotation, ownership/copyright issues of the data, or simply not having enough media content of a certain kind available.

To address this issue, a lot of research has been performed in recent years on learning from scarce data/learning from limited data. There are a variety of ways to work around the problem of data scarcity like using transfer learning, domain transfer or few-shot learning.

The special session on “Learning from scarce data” aims to provide a forum for novel approaches on learning from scarce data for multimedia content analysis, with a focus on the media domain.

The topics of interest include, but are not limited to:

-Transfer learning
-Synthetic data generation
-Domain transfer/adaptation
-Semi-supervised and self-supervised learning, e.g. to take advantage of large amounts of unlabeled media archive content
-Few-shot learning (classification, object detection etc.), which is useful e.g. for adding new object classes to an automatic tagging engine for media archive content.
-Benchmarking and evaluation frameworks for content from the media domain
-Open resources, e.g., software tools for learning from scarce data in the media domain



Session Organisers:
-Dr. Giuseppe Amato, CNR-ISTI, Pisa
-Prof. Bogdan Ionescu, AI Multimedia Lab, Politehnica University of Bucharest, Romania
-Hannes Fassold, JOANNEUM RESEARCH, Graz