(e.g. trained from Moritz or Mihai, or experimentally measured from Hamutal or Mina)
Local motifs
Global structure
Functional decomposition
See also (Hoppe, Grande, 2025, Don’t be afraid of Cell Complexes)
Decompose into modes
This is a global description of the graph
Modes
Low dim representation of functions on the graph
Linear, random connectivity within each subpopulation
Transfer function in terms of functional Fourier modes
What we might explore
Daniel Moreno Soto
Consistent criterion even when no model is correct
Strength of criterion depends on dataset size
More samples
One model is correct
No model is correct
Selection criterion (lower ⇒ better model)
BIC
Bayes factor
MDL
AIC
elpd
EMD
(ours)
May be useful for testing model hypotheses
(René, Longtin, 2025; Selecting fitted models under epistemic uncertainty)