The No Free Lunch Theorem of Deep Learning methods

The No Free Lunch Theorem of Deep Learning methods

A review written in the CODED team in collaboration with the Pascal Institute and Oxford Brookes University explains why neural network learning methods are still not widely used in our biology labs. Despite the abondance of publications of new deep learning methods to analyze bio-images, it is surprising that they do not really cross the threshold of biology laboratories. By analyzing 150 methods dedicated to the analysis of images of nuclei, we identify 5 limits that block their diffusion. Only 4 segmentations methods check all the criteria: Cellpose, DeepCell, QCANet, and NuSet.
Deep learning – promises for 3D nuclear imaging: a guide for biologists. Mougeot et al. J Cell Sci (2022) 135 (7): jcs258986.

Last modified: 05/12/2022