Passing Expectation Propagation Messages with Kernel Methods
This work addresses a specific computational problem in probabilistic inference for researchers and practitioners, representing an incremental improvement by automating message passing in EP.
The authors tackled the computational bottleneck of estimating multivariate integrals in expectation propagation by learning a kernel-based message operator that directly maps incoming messages to outgoing ones, bypassing expensive integral computations during inference.
We propose to learn a kernel-based message operator which takes as input all expectation propagation (EP) incoming messages to a factor node and produces an outgoing message. In ordinary EP, computing an outgoing message involves estimating a multivariate integral which may not have an analytic expression. Learning such an operator allows one to bypass the expensive computation of the integral during inference by directly mapping all incoming messages into an outgoing message. The operator can be learned from training data (examples of input and output messages) which allows automated inference to be made on any kind of factor that can be sampled.