Some recent results on computing with 'neural nets'
Hava T. Siegelmann, Eduardo D. Sontag · 2005
Neural networks, understood here as systems consisting of linearly interconnected dynamical elements (integrators in continuous-time, or delay lines in discrete time) and scalar nonlinearities, provide an appealing model of analog computation. The authors describe some recent results on the approximation properties and the computational capabilities of such dynamical systems.>