Haven't read the article yet, but as a question about this sort of thing in general: How reliable should we hold DL on up-sampling images of systems that we do not fully understand? It's predicting based on past data, and in the case of seeing something new, rare or unusual...?
“deep learning ... improve[s] the resolution of biological images, but elicits skepticism about its ability to enhance snapshots of sample types that it has never seen before.”https://www.the-scientist.com/news-opinion/ai-networks-generate-super-resolution-from-basic-microscopy-65219 …
-
-
-
You are right!
Adv calc n space topology predicts that unexpected behaviors will emerge. Some great, some very bad. Hence the "Garden of Eden" #AI conundrum!#SafetyFirst#machinelearning#bigdata#IoT#robots#Robotics#aiethics#NLP#DeepLearning#AI4good cc@BanKillerRobots - 2 more replies
New conversation -
-
-
[zooming in] "It's made of dog-slugs!"
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
I was the one claiming in this article that DL will likely fail on predicting abnormalities. Hardly see that this will change over time.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
A deep neural net cannot see something it has never seen before? Sounds like something I've been saying over and over for a long time. It's the fatal flaw of deep learning. Why anybody in their right minds would be calling for a hybrid DL/symbolic AI solution is a mystery.
-
Representationalism, the basis of DL, requires that a DNN must have a prior representation of an object or pattern in memory before it can detect it. The brain has no such limitation, thank God. Marcus is leading the field astray with his hybrid nonsense. Don't listen to him.
End of conversation
New conversation -
-
-
So is it rather dead learning?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.