Speaker
Iacopo Mastromatteo
(SISSA)
Description
I will present a general framework which can be used to
reconstruct probability distributions for strings of binary
variables. While the problem of inference can be analytically
controlled for small systems, a description of some of the
regularization prescriptions needed to treat large systems will
be provided, together with a discussion concerning their
symmetries. Finally, I will present several possible applications
of these methods, namely i) exact, fast inference for 1-D
periodic systems, ii) exact, fast inference for tree-like graphs,
iii) approximate, fast inference for generic graphs.
