Dynamical System Implementations of Sparse Bayesian Learning

Despite its state of the art performance in many applications, the sparse Bayesian learning (SBL) procedure can be expensive to implement, limiting its use in practice. In this paper, we use the locally competitive algorithm (LCA) framework to develop two continuous time dynamical systems whose trajectories converge to a minimum of the SBL objective. The resulting systems are neurally feasible and can be implemented using primitives from analog electronics, potentially opening the SBL procedure to new applications.