Etalumis: Bringing Probabilistic Programming to Scientific Simulators at Scale

Authors: Atilim Gunes Baydin (University of Oxford), Lei Shao (Intel Corporation), Wahid Bhimji (Lawrence Berkeley National Laboratory), Lukas Heinrich (European Organization for Nuclear Research (CERN)), Lawrence F. Meadows (Intel Corporation), Jialin Liu (Lawrence Berkeley National Laboratory), Andreas Munk (University of British Columbia), Saeid Naderiparizi (University of British Columbia), Bradley Gram-Hansen (University of Oxford), Gilles Louppe (University of Liege), Mingfei Ma (Intel Corporation), Xiaohui Zhao (Intel Corporation), Philip Torr (University of Oxford), Victor Lee (Intel Corporation), Kyle Cranmer (New York University), Mr Prabhat (Lawrence Berkeley National Laboratory), Frank Wood (University of British Columbia)

Abstract: Probabilistic programming languages (PPLs) are receiving widespread attention for performing Bayesian inference in complex generative models. However, applications to science remain limited because of the impracticability of rewriting complex scientific simulators in a PPL, the computational cost of inference, and the lack of scalable implementations. To address these, we present a novel PPL framework that couples directly to existing scientific simulators through a cross-platform probabilistic execution protocol and provides Markov chain Monte Carlo (MCMC) and deep-learning-based inference compilation (IC) engines for tractable inference. To guide IC inference, we perform distributed training of a dynamic 3DCNN-LSTM architecture with a PyTorch-MPI-based framework on 1,024 32-core CPU nodes of the Cori supercomputer with a global minibatch size of 128k: achieving a performance of 450 Tflop/s through enhancements to PyTorch. We demonstrate a Large Hadron Collider (LHC) use-case with the C++ Sherpa simulator and achieve the largest-scale posterior inference in a Turing-complete PPL.


Presentation: file


Back to Technical Papers Archive Listing