Layered feedforward neural network is relevant to empirical physical formula construction: A theoretical analysis and some simulation results

摘要: Abstract We theoretically establish that, contrary to superficial observation, constructing an empirical physical formula (or law interchangeably) explain the phenomenon is inherently full with several serious obstacles. show that appropriate layered feedforward neural network (LFNN) relevant overcome significantly these To this purpose, we first form a five element set of obstacles pertaining construction. Second, suitably chosen LFNN can each obstacles, because arbitrarily accurately estimates unknown whether experimental variables are deterministic or probabilistic. offer general approach, treat uses non-parametric method sieves estimation. The allows one increase properly number hidden neurons growing sample size. Finally, support our theory, present some simulation results large Here use artificial rather than real data simply in order not prefer any specific equation.