Deep Adaptive Design Amortizing Sequential Bayesian Experimental Design
What this work does
Optimizes a design policy over entire experimental histories using sequential contrastive lower bounds on information gain. History pooling supports fast adaptive design after upfront training.
Connection to discovery
A direct precedent for learning experiment selection rather than merely solving a fixed question, with train-time amortization and test-time evidence gathering separated.
Evidence
Location finding, temporal discounting and a death-process model compare information bounds and deployment cost; the paper also contrasts nonmyopic and myopic designs.
Scope and limitations
The prior, likelihood and latent parameter family are supplied. Informative parameter estimation within that family is narrower than discovering an unanticipated mechanism or representation.
THE RESEARCH CONVERSATION
Discussion & reactions
Share a question, an insight, or a replication experience. Sign in with GitHub to join the conversation.