Computer Vision News - June 2023

26 MICCAI Challenge capable of learning from multiple medical domains . The researchers provide data from 17 different datasets, comprising a total of 28 tasks. The algorithm should possess the ability to adapt to a new domain in a few- shot learning scenario where the specific target domain remains undisclosed . Thedatasetsarepresentedinaunifiedformat and accessible through a PyTorch interface . They consist of resized images, facilitating straightforward input to algorithms. While specific domain knowledge is not necessary, it might prove beneficial. The Learn2Learn Challenge focuses on cross-domain few-shot learning and meta- learning in the field of medicine . These are relatively unexplored areas, particularly in realistic scenarios. Existing meta-learning researchoftenrevolvesaroundtoyproblems using similar tasks from the same dataset, which fails to reflect the complexity of real- world clinical tasks and diverse domains. To address this, the organizers have created a benchmark dataset and initiated a challenge forother researchers to test their algorithms. The challenge aims to develop an algorithm THE MICCAI LEARN2LEARN CHALLENGE Stefano Woerner and Susu Sun are PhD students in the Machine Learning for Medical Image Analysis Group in Christian Baumgartner’s lab at the University of Tübingen. They are co-organizers of the MICCAI Learn2Learn Challenge, running now, with results announced at the conference in October. Stefano Woerner Susu Sun