Deep Adaptive Design: Amortizing Sequential Bayesian Experimental Design

              Adam Foster * 1 Desi R. Ivanova * 1 Ilyas Malik 2 Tom Rainforth 1

             Abstract                 using a likelihood p(y|θ, ξ) and a prior p(θ), where ξ is
   We introduce Deep Adaptive Design (DAD),           our controllable design and θ is the set of parameters we
   a method for amortizing the cost of adaptive         wish to learn about. We then optimize ξ to maximize the ex-
   Bayesian experimen ...