FL-net, an autoencoder-based neural network, maps imaginary-time Green's functions to spectral functions with roughly 20% lower loss than MEM and previous networks, but larger hidden layers make it more noise-sensitive.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cond-mat.str-el 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Analytic Continuation by Feature Learning
FL-net, an autoencoder-based neural network, maps imaginary-time Green's functions to spectral functions with roughly 20% lower loss than MEM and previous networks, but larger hidden layers make it more noise-sensitive.