报告摘要：
The inverse Isingproblem means to reconstruct interactions (external fields and couplings)from observations (magnetizations and correlation functions). It is equivalent to a maximum entropy inference of the probability distribution, given megnetizationsand correlation functions, and has applications to diverse fields such as neuroscience, systems biology / genetics. The kinetic inverse IsingProblem means by analogy to reconstruct external fields and couplings from time series data from a kinetic Isingmodel. I will survey recent developments in the areawith an emphasis on connections between statistical physics and machine learning. I willalso cover recent results by us on extending the Belief Propagation schemes to non-equilibriumspin systems.
This is joint work with John Hertz, Yasser Roudi, MikkoAlava, ZengHongfLiand HamedMahmoudi.