Akinori HIDAKA

Tokyo Denki University

Relaxation of Constraints of Supervised Learning for Deep Bioimaging

Image recognition technology based on deep learning has made remarkable advances. In particular, the practical application of problem setting for which a tremendous amount of images for learning can be prepared relatively easily (e.g., on-road object recognition from vehicle-mounted camera images) has gotten ahead of others. On the other hand, images in the bioimaging field are costly and time-consuming and require skills in terms of individual acquisition; thus, it is often difficult to prepare a large number of high-quality images for learning.
In this study, we will develop technologies that will allow for deep learning from as few bioimaging field images as possible, thereby enabling automatic content understanding, image sharpening, and super-resolution of microscopic images, etc., based on the framework of unsupervised learning, semi supervised learning, and unpaired supervised learning.