Computer Vision News - July 2018

Juan Caicedo is a postdoc researcher at the Broad Institute of MIT and Harvard. He works with a multidisciplinary team of biologists and computer scientists, led by Anne Carpenter , and the lab is mostly focused on analysing biomedical images. More specifically, microscopic images for understanding the effects of treatments, either genetical or chemical perturbations. He speaks to us before his poster presentation at CVPR. The work Juan is presenting is about analysing microscopic images in order to learn representations for those images without any labels. The technique is called weakly supervised learning . He is excited to see so many different researchers working in weakly supervised learning at CVPR 2018. Juan tells us that he wants to learn representations for these images in order to compare the effects of different treatments. Imagine a patient with cancer and you take some cells of this patient and capture images of those cells under the microscope. Then you want to know what happens if you apply certain drugs to those cells. Just by looking at the images, is the population of cells going to have a positive response or a negative response? For many problems in computer vision, including microscopic images, there are not nice and clean labelled datasets. Since they don’t have labels for these specific patients, they want to make a system that can learn without labels or additional information. 40 Juan Caicedo Thursday “We take the images and those images usually have multiple cells, so we use neural networks to extract each individual cell from the image.”