Industrial PhD collaboration between the UiT Machine Learning Group and eSmart Systems led to paper at the European Conference on Computer Vision 2020
The European Conference on Computer Vision (ECCV) is one of the premier conferences on machine learning where only the works of the highest quality is presented. This year only 26 % of the 5150 submissions made the cut, one of which was produced by the UiT Machine Learning Group.
- We are very happy with our work being presented at such a prestigious venue as the ECCV. This illustrates the group’s leading role in the country with regards to machine learning and deep learning research, says head of the new center for research-based innovation ”Visual Intelligence”, Robert Jenssen.
The paper, entitled ”SEN: A Novel Feature Normalization Dissimilarity Measure for Prototypical Few-Shot Learning Networks”, introduces a new dissimilarity measure designed for the task of few-shot classification, which aims at identifying unseen classes form just a handful of labeled examples.
- Learning from a limited amount of labeled data is one of the key challenges facing deep learning-based research, says Michael Kampffmeyer. He is one of the authors of the paper and has published numerous papers at top computer vision conferences over the last couple of years.
- Deep learning-based algorithms have demonstrated impressive results on tasks where a lot of labeled data is available. However, they are not as effective when the amount of labeled data is small. The lack of labeled data is a typical setting in many domains, for instance healthcare, and is therefore an important problem to address. Few-shot learning is one of the burgeoning areas of research that address the difficulties of learning from few examples, and in this paper we demonstrate how the proposed dissimilarity measure pushes the state-of-the-art in this field, explains Kampffmeyer.
The paper can be found here and a video presentation of the article is included below.
This work on few-shot learning sprung out from industrial PhD collabora- tion between the the UiT Machine Learning Group and the company eSmart Systems based in Halden, Norway. The goal has been to conduct basic research that also results in benefits when applied to real world problems.
- Moving the frontier in both basic machine learning research and applied research is one of the groups main focus areas, and fruitful collaborations with companies such as eSmart Systems is one way of achieving this goal, says Jenssen.
For more information about the UiT Machine Learning Group’s work on machine learning research, please visit the webpage of the group: https:// machine-learning.uit.no/