Continual Learning in Computer Vision: Theory and Applications

Dear Colleagues,

In less than a decade, we have witnessed the rise of deep neural networks as the main paradigm for supervised and unsupervised learning. However, despite the impressive results achieved in a wide variety of fields, the current training approaches still suffer from a range of problems related to their incapability to adapt to scenarios different from the ones they are trained in—a capability which, instead, is a key factor in the way people learn, showing that the road to human-like artificial intelligence is still very long. While we can train very good models that carry out individual tasks, it is still unclear how to train models that can learn sequentially to perform multiple tasks sequentially.

Continual learning addresses the design and training of models with a built-in capability to adapt to multiple tasks without suffering from catastrophic forgetting and to work well in test conditions different from those seen during training. The impact of successful methods for continual learning would both affect the foundations of machine learning and extend the potential of existing solutions, across multiple application fields. Indeed, one can envisage how a model that can identify pulmonary diseases from X-ray images could benefit from re-using features learned to perform segmentation, and how useful a model that can perform both tasks would be to a physician.

In this Special Issue, we invite authors to send theoretical and application contributions related to continual learning, including but not limited to the following topics:

• Critical surveys on continual learning
• Training approaches for continual learning
• Model design in continual learning
• Memory-based techniques
• Dataset distillation and model distillation
• Continual learning on real-world datasets
• New datasets for continual learning
• Continual learning for expert domains (e.g., medicine)
• Continual active learning

Dr. Simone Palazzo
Dr. Carmelo Pino
Guest Editors

Manuscript Submission Information

Manuscripts should be submitted online at www.mdpi.com by registering and logging in to this website. Once you are registered, click here to go to the submission form. Manuscripts can be submitted until the deadline. All submissions that pass pre-check are peer-reviewed. Accepted papers will be published continuously in the journal (as soon as accepted) and will be listed together on the special issue website. Research articles, review articles as well as short communications are invited. For planned papers, a title and short abstract (about 100 words) can be sent to the Editorial Office for announcement on this website.

Submitted manuscripts should not have been published previously, nor be under consideration for publication elsewhere (except conference proceedings papers). All manuscripts are thoroughly refereed through a single-blind peer-review process. A guide for authors and other relevant information for submission of manuscripts is available on the Instructions for Authors page. Journal of Imaging is an international peer-reviewed open access monthly journal published by MDPI.

Please visit the Instructions for Authors page before submitting a manuscript. The Article Processing Charge (APC) for publication in this open access journal is 1800 CHF (Swiss Francs). Submitted papers should be well formatted and use good English. Authors may use MDPI's English editing service prior to publication or during author revisions.

• Continual learning
• Catastrophic forgetting
• Continual learning datasets
• Domain-specific continual learning
• Class-incremental learning
• Task-incremental learning