Computer Vision News - May 2023

23 Abby Stylianou of progress dealing with uncertainty? They are. This is clearly something that, broadly, there's a lot of interest in. I think it's something that I've not seen as much in the image retrieval community. I will admit that it just may be that I have missed important work. There's so much coming out of our field, right? It is entirely possible there is important work that I have not seen yet. But no, I think as a field, we do generally realize we need much better notions of uncertainty and confidence. But it's something I really want to build into our models because I think it's really important for investigators. You've told us a lot about your current work. And I see that you have several decades of career still in front of you to do. Do you see yourself on this path forever?  [ laughs ] It's funny when I look at my like, here's the thing in the query image that we focused on, and here's the thing in the resulting image. Much of computer vision when we're trying to do explainability, we make heat maps. They're kind of a horrible way of explaining anything. It doesn't really help, especially for somebody who's not an expert, which is what we're really dealing with, what the model thinks is important. So I think at having better natural language explanations, for example. We've really had this very recent explosion of impressive language models. I would love to be able to tell an investigator the reason we think this is similar. It is the red couch or something like that. And to give them the ability to correct that and say like, no, I don't care about the red couch. Focus on this. You're hearing all of the things that I'mwriting into my next grant! Isn't the scientific community making a lot