StackedTempering

Julia code accompanying the paper "Accelerated sampling with stacked restricted boltzmann machines" by Jorge Fernandez-de-Cossio-Diaz, Clément Roussel, Simona Cocco, and Rémi Monasson.

Installation

You will need Julia installed on your machine.

This pacakge is not registered in the Julia General registry. We provide an independent registry that can be used to install this package and its dependencies. To install, run the following commands at the Julia Pkg REPL:

pkg> registry add https://github.com/2024stacktemperingrbm/2024-ICLR-StackedTempering-JuliaRegistry.git
pkg> add StackedTempering RestrictedBoltzmannMachines

See https://github.com/2024stacktemperingrbm/2024-ICLR-StackedTempering-JuliaRegistry for instructions.

Usage

The main function is stacked_tempering, which samples the stack of RBMs exchanging configurations between neighboring layers, while respecting the Metropolis rule.

Example:

using RestrictedBoltzmannMachines: BinaryRBM
using StackedTempering: sample_bottom_to_top
using StackedTempering: sample_top_to_bottom
using StackedTempering: stacked_tempering

# Create stack of RBMs
rbms = (
    BinaryRBM(randn(7), randn(5), randn(7,5)),
    BinaryRBM(randn(5), randn(3), randn(5,3)),
    BinaryRBM(randn(3), randn(2), randn(3,2))
)

# Sample the stack of RBMs, bottom to top
hs = sample_bottom_to_top(rbms, falses(7, 10))

# Sample the stack of RBMs, top to bottom
vs = sample_top_to_bottom(rbms, falses(2, 10))

# sample the stack of RBMs, respecting Metropolis transitions between the layers
stacked_tempering(rbms, vs, hs)

Citation

If you use this code in a publication, please cite:

Jorge Fernandez-de-Cossio-Diaz, Clément Roussel, Simona Cocco, and Remi Monasson. "Accelerated sampling with stacked restricted boltzmann machines." The Twelfth International Conference on Learning Representations (2024).

Or you can use the included CITATION.bib.