3 ms·
Neural networks do exactly what you are describing as "symbolic reasoning". It seems to be a common thing recently to dismiss modern ML techniques as curve fitt
by viuphiet 7y ago
Neural networks do exactly what you are describing as "symbolic reasoning". It seems to be a common thing recently to dismiss modern ML techniques as curve fitting, but these fundamental models are extremely powerful.
Neural networks are capable of approximating any system to arbitrary precision.
- uoaei 7y agoThis is theoretically true, but it's like saying "computers can compute any function, given enough time and resources". There is a need to construct logically-deduced models which impose an inductive bias so that your regression methods are efficient. That's where reasoning comes in, and where automated reasoning methods should be useful.