Computer Vision News - November 2022

7 Xingjian Zhen analyze the neurons and compare two different networks. Here, we’re offering a new method that hasn’t been used in the community before to efficiently compare two networks and offer some ability to condition on a network, which previous methods did not provide. It’s a simple and elegant way to solve a popular problem! ” As we wrap up the interview, we uncover what may have been fueling the team’s success when we ask them to tell us something about UW-Madison that we do not know. “ It has the best ice creamof all universities! ” Zihang laughs. “ That’s what I wanted to say, ” Xingjian adds. “ I think Zihang is still missing The Daily Scoop. ” “ He’s right, ” Zihang confirms. “ I’m in New York, but I still miss the ice cream at Wisconsin-Madison! ” problems in any field, the best solutions are generally the simple ones, ” he points out. “ The beautiful part is borrowing something from textbook statistics on machine learning and seeing how applicable it is in computer vision literature. We’re all glad that the reviewers and committee members saw what we saw in this work. ” Zihang believes that researchers in the deep learning community have always been interested in how to qualitatively or quantitatively understandwhat anetwork has learned . “ Five or six years back, people did some beautiful visualizations of the neurons within deep neural networks to see what they had learned, ” he recalls. “ More recently, researchers tried to use canonical correlation analysis (CCA) to S T PER CCV