|Title:||Machine learning determination of dynamical interactions: From the Ising model to biological data (Part 1)|
|Date (JST):||Wed, Jun 19, 2019, 11:00 - 12:00|
|Place:||Seminar Room A|
In the first part of this talk (Guido Cossu) we discuss training a set of Restricted Boltzmann Machines (RBM) on two-dimensional Ising spin configurations at various temperatures and validate the training procedure by monitoring several estimators such as the log-likelihood. We present a closed form expression for extracting the values of n-point interactions between the visible nodes, as well as the success of the RBM in predicting the Ising couplings.
In the second part (Ava Khamesh), after describing the derivation of the analytical formula for the couplings in more detail, we discuss training an RBM on binary disease data across the UKBiobank population to extract couplings between traits/diseases. We then present some of the expected predictions, both in terms of measurement of observables and the couplings. We further present some of the challenges in training and extracting potential biological meanings from such interactions.