The actual normal-stage source size is controlled by one fixed envelope. This includes the chosen terminal coordinate and canonical boundary coefficient, with the polynomial first shear retained.
The source size used by the canonical packet solve is bounded by one explicit polynomial-exponential envelope, including the base-sized boundary coefficient and both reciprocal time intervals.
Actual selected terminal coordinates and the canonical boundary parameter have fixed polynomial caps. They are inputs to the uniform normal-stage source envelope.
Terminal constant, given by 8*(activationConstant CM CH+1)*(1+3*(1+embeddingCost)^2).
Equations
- EulerParentPacketParameterCaps.terminalConstant CM CH = 8 * (EulerTransverseActivationSelection.activationConstant CM CH + 1) * (1 + 3 * (1 + EulerPacketParentLabelBounds.embeddingCost) ^ 2)
Instances For
Boundary constant, given by boundaryLocalizationC1*(CM+2)+1.
Equations
Instances For
Terminal cap, given by 1+terminalConstant gradientConstant hessianConstant.
Equations
Instances For
Source constant, given by EulerPacketParameterEnvelope.boundConstant C (boundaryConstant gradientConstant) terminalCap.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Envelope, given by parameterEnvelope J (sourceConstant C) 320 20 1000 X n.
Equations
- EulerNormalPacketParameters.envelope J C X n = EulerPacketUniformFrequencyScales.parameterEnvelope J (EulerNormalPacketParameters.sourceConstant C) 320 20 1000 X n
Instances For
Frequency spec, constructed using frequencyCostSpec.
Equations
- One or more equations did not get rendered due to their size.