Detecting cancers with neural networks requires a balance of rewards and penalties
Read now →Researchers improved AI tumor detection by penalizing it for learning where the scans originated
Darcy Murphy
Computer Science
University of Manchester
I’m getting towards the end of a part time MSc in “Digital Biology” (Computer Science with some Healthcare Informatics thrown in). I previously trained as an early years teaching assistant and worked at a primary school breakfast club, which I think made me much better at communicating ideas as simply as possible. My MSc project is about a technique to stop neural networks from learning bias that you know is in your training data, which I’m applying to a neural network that predicts heart disease from EGC recordings. I’m especially interested in ethical issues in machine learning and how those intersect with technical issues with biased data. I have three cats and a pet fish called Keith that I adopted from a friend who couldn’t keep him!
Researchers improved AI tumor detection by penalizing it for learning where the scans originated
Computer programs used in 46 states incorrectly label Black defendants as “high-risk” at twice the rate as white defendants