Neural network designing method and digital-to-analog fitting method

摘要

A neural network designing method forms a RNN (Recurrent Neural Network) circuit to include a plurality of oscillating RNN circuits configured to output natural oscillations, and an adding circuit configured to obtain a sum of outputs of the plurality of oscillating RNN circuits, and inputs discrete data to the plurality of oscillating RNN circuits in order to compute a fitting curve with respect to the discrete data output from the adding circuit.

主权项

1. A neural network designing method comprising:
a forming procedure causing a computer to form a RNN (Recurrent Neural Network) circuit to include a plurality of oscillating RNN circuits configured to output natural oscillations, and an adding circuit configured to obtain a sum of outputs of the plurality of oscillating RNN circuits; and a computing procedure causing the computer to input discrete data to the plurality of oscillating RNN circuits, in order to compute a fitting curve with respect to the discrete data output from the adding circuit, wherein the computing procedure includes
computing and storing in a storage unit a0,a1, . . . , a2n-1,b1, . . . , b2n-1represented by the following formulas from the discrete data y0=0, y1, . . . , y2n-1, y2n=0,a0⁢=12⁢n⁢∑i=12⁢n-1⁢yi=12⁢n⁢∑i=02⁢n⁢yiaq=ψ⁡(q2⁢n)4⁢n⁢⁢π2⁢⁢∑i=12⁢n-1⁢yi⁡(-cos⁢⁢(i-1)⁢q⁢⁢πn+2⁢⁢cos⁢⁢i⁢⁢q⁢⁢πn-cos⁢⁢(i+1)⁢q⁢⁢πn)q=1,2⁢⁢…⁢⁢2⁢n-1bq=ψ⁡(q2⁢n)4⁢n⁢⁢π2⁢∑i=12⁢n-1⁢yi⁡(-sin⁢⁢(i-1)⁢q⁢⁢πn+2⁢sin⁢⁢i⁢⁢q⁢⁢πn-sin⁢⁢(i+1)⁢q⁢⁢πn)q=1,2⁢⁢…⁢⁢2⁢n-1ψ⁡(x)=ⅆ2ⅆx2⁢log⁢⁢Γ⁡(x) where ψ(x) denotes a trigamma function that is defined by a second derivative of a logarithm of a gamma function, and a0, a1, . . . , a2n-1, b1, . . . , b2n-1 are fixed values independent of the time t, and
computing the fitting curve defined by the following formula, using the fixed values a0,a1, . . . , a2n-l, b1, . . . , b2n-1,y⁡(t)=a0+∑q=12⁢n-1⁢aq⁢cos⁢⁢q⁢⁢πn⁢t+∑q=12⁢n-1⁢bq⁢sin⁢⁢q⁢⁢πn⁢t where cos (qπt/n) and sin(qπt/n) are oscillating functions output from the plurality of oscillating RNN circuits.