Few-shot Learning for Intelligent Multimedia Systems

Guest Editors

Jiachen Yang, Tianjin University, China (yangjiachen@tju.edu.cn)

Houbing Song, Embry-Riddle Aeronautical University, USA (songh4@erau.edu)

Qinggang Meng, Loughborough University, UK (Q.Meng@lboro.ac.uk)

Aims and scope

Multimedia data is all around us, e.g. videos, images, texts, etc. In recent years, we have achieved many excellent performances on the processing and analysis of multimedia data through deep learning. However, the current deep learning method relies on quite large-scale datasets, which is time-consuming to annotate and seems still far away from our desired intelligence. Considering the way humans how to deal with multimedia data, such as classification of images or videos, we can easily complete the recognition task from only a handful of data rather than millions of data. As for the intelligent multimedia systems, the knowledge-driven is clearly more appropriate than data-driven. So, it is still challenging for the existed techniques to complete intelligent pattern recognition from only a few labelled multimedia data, called few-shot learning, which aims to develop a model with good generalization based on a few samples. But the few-shot learning for intelligent multimedia systems should be a fundamental step towards the desired artificial intelligence, as it pursues the combination of natural intelligence with algorithm flexibility and extensibility.

The goal of this special issue is to assemble recent advances in the few-shot learning based multimedia analysis and processing. The multimedia data of interest covers a wide spectrum, ranging from images, texts, audios, to kinds of videos. We expect the contribution focusing on the innovative few-shot techniques, including methodological and algorithmic methods to solve the theoretical and practical multimedia problems. We also encourage the contributions on various deployable few-shot applications.

Topics of interest include but are not limited to:

• Survey of few-shot learning towards intelligent multimedia systems
• Theoretical studies of few-shot learning, e.g., information theory, entropy
• Few-shot learning towards image/video classification, object detection, and segmentation
• Few-shot learning towards multimedia data generation, fusion, and augmentation
• Few-shot learning towards applications, e.g., multimedia quality assessment
• Few-shot learning towards special hardware architectures and deployments
• New dataset and benchmark for few-shot intelligent multimedia systems

Important dates

Manuscript submission deadline: November 15, 2021

Decision notification: December 30, 2021

Author revisions due (if applicable): January 30, 2022

Final decision notification: February 28, 2022

Submission Guidelines

Papers submitted to this special issue must be original and must not be under consideration for publication in any other journal or conference. The manuscripts will be peer-reviewed strictly following the reviewing procedures. All submitted papers will be evaluated on the basis of relevance, significance of contribution, technical quality, scholarship, and quality of presentation, by at least three independent reviewers. The papers must be written in English and must not exceed 30 pages (single column, double space, 12 pt font, including figures, tables, and references). Authors should prepare their manuscript according to the journal’s Submission Guidelines at https://www.springer.com/journal/530