An Adaptive Learning Approach To Parameter Estimation For Hybrid Petri Nets In Systems Biology

In this work, we investigate adaptive learning techniques in hybrid Petri nets (HPNs) that can model biological systems. In particular, based on a state space formulation we develop a decisionaided adaptive gradient descent (DAAGD) algorithm capable of cost-effectively estimating the parameters used in an HPN model. Contrary to standard gradient descent techniques, the DAAGD algorithm does not require prior knowledge, i.e., information about the discrete transitions' firing instants. Simulations of a gene regulatory network assess the performance of the proposed DAAGD algorithm against standard gradient descent algorithms with full, imperfect and no prior knowledge.