Wednesday, October 6, 2010

Computational Energy Surface

Essentially, a computational energy surface is an n-dimensional surface that represents the voltage output for any combination of input voltages. A well-designed neural network will have valleys, or self-consistent stable points, at the places that correspond to particular “answers” for a given set of input voltages. The shape of this computational energy surface is determined by the connections between the amplifiers, the strengths (resistance) of each of those connections, and any external currents applied to the amplifiers.

No comments: