Tuan Anh Le

c3: lightweight incrementalized mcmc for probabilistic programs using continuations and callsite caching

02 January 2017

notes on (Ritchie, Stuhlm├╝ller, & Goodman, 2016).

summary

this paper proposes a way to make lightweight metropolis hastings (lmh) more efficient. the high level idea is to save as much computation as possible in addition to the reuse of random choices in the original lmh algorithm. the main ideas are

there are further optimizations such as automatic cache adaptation, thunking, copy-on-write (only write changes to cache if proposal is accepted), and using hash-tables for storing random choices.

experiments show promising results.

note to self


references

  1. Ritchie, D., Stuhlm├╝ller, A., & Goodman, N. D. (2016). C3: Lightweight Incrementalized MCMC for Probabilistic Programs using Continuations and Callsite Caching. Proceedings of the 19th International Conference on Artificial Intelligence and Statistics (AISTATS). Retrieved from http://arxiv.org/abs/1509.02151
    @article{ritchie2016c3,
      author = {Ritchie, Daniel and Stuhlm\"{u}ller, Andreas and Goodman, Noah D.},
      title = {C3: Lightweight Incrementalized MCMC for Probabilistic Programs using Continuations and Callsite Caching},
      booktitle = {Proceedings of the 19th International Conference on Artificial Intelligence and Statistics (AISTATS)},
      year = {2016},
      keywords = {probabilistic programming, inference, mcmc, metropolis-hastings},
      url = {http://arxiv.org/abs/1509.02151}
    }
    

[back]