Moving on to general symbolic regression, possible domain checks should be done on the problems. Checking for positive / negative definiteness, zero crossings etc. either should limit the use of functions or handle them in a safe way ( e.g. instead of 1/0 -> 1/eps() ).
This could be also used to infer functions itself, e.g. x^2 -> positive semidefinite, x^3 -> indefinite. So we might get hints ( similar to Feynman AI ) here.
Moving on to general symbolic regression, possible domain checks should be done on the problems. Checking for positive / negative definiteness, zero crossings etc. either should limit the use of functions or handle them in a safe way ( e.g. instead of
1/0->1/eps()).This could be also used to infer functions itself, e.g. x^2 -> positive semidefinite, x^3 -> indefinite. So we might get hints ( similar to Feynman AI ) here.