“NEURAL NETWORKS are called black boxes because the training algorithm finds the optimal node values to solve a problem, but looking at the solution it is impossible to tell why that solution works without decomposing every element of the network. In other words, the node values are extremely sensitive to the context (nodes they connect to), so you have to map out the entire thing to understand it.”
https://www.reddit.com/r/todayilearned/comments/3d3vct
linkedin post 2022-06-29 05:12:48
Nigel Fleming
September 21, 2022