Computer Vision News - August 2020

Raghavendra Selvan 25 Best of MIDL 2020 port it to segmentation or regression, which may entail some methodological development or rigorous validation. Although still at the research stage, Raghav can foresee the work maturing to the point where it could be used in a real clinical setting. “This method features only linear components and linear decisions, so it translates into better interpretability ,” he explains. “If you want to use a model in a clinical setting, they’re not interested in numbers, they want to know what something means. With a linear model regime, there’s plenty of scope for improving interpretability in a clinical setup . That’s something we would like to investigate a bit further, but I’d be happy to see someone else in the community take it up too. It’s such an interesting topic.” The first time Raghav encountered tensor networks was at NeurIPS in December last year. He was so fascinated he ended up spending his Christmas break getting to grips with them. “We’ve heard so much about convolutional neural networks and feedforward neural networks,” he says. “I was like, what is this thing? Then I started reading and started thinking this was something really cool. I didn’t know if it would pan out to medical image data, but I think that’s where our contribution is to adapt it in a way that it works.” Raghav’s paper will be featured in a special edition of MELBA, the new journal being launched in conjunction with MIDL. "… with less than 10 per cent on a single GPU you can do what convolutional neural networks do with four GPUs!"