Write a Blog >>
Tue 21 Jan 2020 14:00 - 14:30 at St Claude - C

Universal probabilistic programming systems (PPSs) provide a powerful framework for specifying rich and complex probabilistic models. However, it comes at the cost of substantially complicating the process of drawing inferences from the model. In particular, inference can become challenging when the support of the model varies between executions. Though general-purpose inference engines have been designed to operate in such settings, they are typically inefficient, often relying on proposing from the prior to make transitions. To address this, we introduce a new inference framework: Divide, Conquer, and Combine (DCC). DCC divides the program into separate straight line sub-programs, each of which has a fixed support allowing more powerful inference algorithms to be run locally, before recombining their outputs in a principled fashion. We show how DCC can be implemented as an automated and general-purpose PPS inference engine, and empirically confirm that it can provide substantial performance improvements over previous approaches.

Tue 21 Jan

lafi-2020
14:00 - 15:05: LAFI (né PPS) - C at St Claude
lafi-202014:00 - 14:30
Talk
Yuan ZhouUniversity of Oxford, Hongseok YangKAIST, Yee Whye TehUniversity of Oxford, Tom RainforthDepartment of Statistics, University of Oxford
lafi-202014:32 - 14:47
Talk
Alexander K. LewMassachusetts Institute of Technology, USA, Benjamin ShermanMassachusetts Institute of Technology, USA, Marco Cusumano-TownerMIT-CSAIL, Austin GarrettMIT, Ben ZinbergMIT, Vikash MansinghkaMIT, Michael CarbinMassachusetts Institute of Technology
lafi-202014:49 - 15:05
Talk