For #1, yes, what you say makes sense and fits my understanding of the relevant terminology. It’s conceivable that a neural network would learn a good approximate lookup table instead, but if the architecture were sufficiently capable (EG, neural turing machines) learning search would be more compact & more accurate.
For #1, yes, what you say makes sense and fits my understanding of the relevant terminology. It’s conceivable that a neural network would learn a good approximate lookup table instead, but if the architecture were sufficiently capable (EG, neural turing machines) learning search would be more compact & more accurate.