mayhaps (do you have a linky, sounds hilarious). But you can’t force feed them thousands of images, and then put a current one and have it highlight areas of concern or saying probability or something. In any case, cancer on images is the more grifty one, cause they just have millions of datapoints, instead of like 50 haphazard biopsies and pet scans
I didn’t really go through it to see if they compare it with ML performance, but it’s still a fun anecdote.
Plus, you don’t have to feed an ML model or clean its pen. Or train new ones when the current generation dies. I don’t really know if there is –or should be– a practical application in scale of these findings, but it’s pretty neat!
mayhaps (do you have a linky, sounds hilarious). But you can’t force feed them thousands of images, and then put a current one and have it highlight areas of concern or saying probability or something. In any case, cancer on images is the more grifty one, cause they just have millions of datapoints, instead of like 50 haphazard biopsies and pet scans
Here’s the link the article I read was referencing: https://pmc.ncbi.nlm.nih.gov/articles/PMC4651348/
I didn’t really go through it to see if they compare it with ML performance, but it’s still a fun anecdote.
Plus, you don’t have to feed an ML model or clean its pen. Or train new ones when the current generation dies. I don’t really know if there is –or should be– a practical application in scale of these findings, but it’s pretty neat!