Hamiltonian Monte Carlo Without Detailed Balance

Jascha Sohl-Dickstein, Mayur Mudigonda, Michael DeWeese
Proceedings of the 31st International Conference on Machine Learning, PMLR 32(1):719-726, 2014.

Abstract

We present a method for performing Hamiltonian Monte Carlo that largely eliminates sample rejection. In situations that would normally lead to rejection, instead a longer trajectory is computed until a new state is reached that can be accepted. This is achieved using Markov chain transitions that satisfy the fixed point equation, but do not satisfy detailed balance. The resulting algorithm significantly suppresses the random walk behavior and wasted function evaluations that are typically the consequence of update rejection. We demonstrate a greater than factor of two improvement in mixing time on three test problems. We release the source code as Python and MATLAB packages.

Cite this Paper


BibTeX
@InProceedings{pmlr-v32-sohl-dickstein14, title = {Hamiltonian Monte Carlo Without Detailed Balance}, author = {Sohl-Dickstein, Jascha and Mudigonda, Mayur and DeWeese, Michael}, booktitle = {Proceedings of the 31st International Conference on Machine Learning}, pages = {719--726}, year = {2014}, editor = {Xing, Eric P. and Jebara, Tony}, volume = {32}, number = {1}, series = {Proceedings of Machine Learning Research}, address = {Bejing, China}, month = {22--24 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v32/sohl-dickstein14.pdf}, url = {https://proceedings.mlr.press/v32/sohl-dickstein14.html}, abstract = {We present a method for performing Hamiltonian Monte Carlo that largely eliminates sample rejection. In situations that would normally lead to rejection, instead a longer trajectory is computed until a new state is reached that can be accepted. This is achieved using Markov chain transitions that satisfy the fixed point equation, but do not satisfy detailed balance. The resulting algorithm significantly suppresses the random walk behavior and wasted function evaluations that are typically the consequence of update rejection. We demonstrate a greater than factor of two improvement in mixing time on three test problems. We release the source code as Python and MATLAB packages.} }
Endnote
%0 Conference Paper %T Hamiltonian Monte Carlo Without Detailed Balance %A Jascha Sohl-Dickstein %A Mayur Mudigonda %A Michael DeWeese %B Proceedings of the 31st International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2014 %E Eric P. Xing %E Tony Jebara %F pmlr-v32-sohl-dickstein14 %I PMLR %P 719--726 %U https://proceedings.mlr.press/v32/sohl-dickstein14.html %V 32 %N 1 %X We present a method for performing Hamiltonian Monte Carlo that largely eliminates sample rejection. In situations that would normally lead to rejection, instead a longer trajectory is computed until a new state is reached that can be accepted. This is achieved using Markov chain transitions that satisfy the fixed point equation, but do not satisfy detailed balance. The resulting algorithm significantly suppresses the random walk behavior and wasted function evaluations that are typically the consequence of update rejection. We demonstrate a greater than factor of two improvement in mixing time on three test problems. We release the source code as Python and MATLAB packages.
RIS
TY - CPAPER TI - Hamiltonian Monte Carlo Without Detailed Balance AU - Jascha Sohl-Dickstein AU - Mayur Mudigonda AU - Michael DeWeese BT - Proceedings of the 31st International Conference on Machine Learning DA - 2014/01/27 ED - Eric P. Xing ED - Tony Jebara ID - pmlr-v32-sohl-dickstein14 PB - PMLR DP - Proceedings of Machine Learning Research VL - 32 IS - 1 SP - 719 EP - 726 L1 - http://proceedings.mlr.press/v32/sohl-dickstein14.pdf UR - https://proceedings.mlr.press/v32/sohl-dickstein14.html AB - We present a method for performing Hamiltonian Monte Carlo that largely eliminates sample rejection. In situations that would normally lead to rejection, instead a longer trajectory is computed until a new state is reached that can be accepted. This is achieved using Markov chain transitions that satisfy the fixed point equation, but do not satisfy detailed balance. The resulting algorithm significantly suppresses the random walk behavior and wasted function evaluations that are typically the consequence of update rejection. We demonstrate a greater than factor of two improvement in mixing time on three test problems. We release the source code as Python and MATLAB packages. ER -
APA
Sohl-Dickstein, J., Mudigonda, M. & DeWeese, M.. (2014). Hamiltonian Monte Carlo Without Detailed Balance. Proceedings of the 31st International Conference on Machine Learning, in Proceedings of Machine Learning Research 32(1):719-726 Available from https://proceedings.mlr.press/v32/sohl-dickstein14.html.

Related Material