Several autoencoder analysis functions could be sped up by storing the input/decoding radial graph rather than recomputing it over and over again.

[InfluenceFunctional]