15 JunDeep learning needs [expert] data!

“DeepMind Shows AI Has Trouble Seeing Homer Simpson’s Actions” (link), describes how machine learning (deep learning, neural networks, artificial intelligence) needs annotated data to train on so they went with Amazon’s Mechanical Turk service to create it.  That works because we all have visual systems that to a great job.  But some deep learning applications need expertise beyond what humans can do easily.  Yes, I’m thinking of labeling neuroanatomical regions in MRI brain scans.  Neuroanatomy is hard!

Take a look at any recent article on deep learning and you’ll find the standard lament that there is not enough high quality annotated data available.  This burning need is leading to the creation of companies such as understand.ai to do just that.  Investors are starting to notice, and I suspect that getting in now is a very good idea.

Neuromorphometrics has been annotating anatomy from before there was deep learning.  We started out calling the process “segmentation” and more recently I like to describe it as “labeling”  which is a bit more specific than “annotation.”  Whatever it’s called, inquiries about investing are welcome!