Gary Marcus

@MaxALittle @DrHughHarvey @DrLukeOR developing thread on deep learning in medicine, feel free to pipe in...

I haven't read your book (yet) but looking forward to some realism! In my experience deep learning is great for computer vision tasks where output is binary with high accuracy - BUT many medical tasks are not binary at all, and accuracy can be hindered by many confounders

The issue with any medical vision applications is that the number of possible medical diagnoses in a given situation is vast, yet not all are represented well in datasets, and rare entities can be critical. For example, if I get a biopsy from a chronic non-healing wound in a...

...young man, it may be pus or it may have granulomatous inflammation indicative of certain infections or it may be an epithelioid sarcoma that’s extremely rare yet a very critical diagnosis to make and not to miss. How can we make deep learning robust for rare critical events?

People who think safe autonomous vehicles are ever going to be feasible just need to go on a trip to Naples or Sicily and see for themselves how driving is done in the rest of the world.

I find these criticisms of deep learning vis-à-vis humans a bit odd. 1-Human doctors also pattern match and are unlikely to come up with diagnoses of conditions they have never encountered in medical training.