Stochastic Gradient Richardson-Romberg Markov Chain Monte Carlo

Authors: Alain Durmus, Umut Simsekli, Eric Moulines, Roland Badeau, Gaël RICHARD

NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We support our findings using both synthetic and real data experiments.
Researcher Affiliation Academia 1: LTCI, CNRS, T el ecom Paris Tech, Universit e Paris-Saclay, 75013, Paris, France 2: Centre de Math ematiques Appliqu ees, UMR 7641, Ecole Polytechnique, France
Pseudocode Yes We provide a pseudo-code of SGRRLD in the supplementary document.
Open Source Code No The paper states, 'We have implemented SGLD and SGRRLD in C by using the GNU Scientific Library for efficient matrix computations,' and 'We provide a pseudo-code of SGRRLD in the supplementary document,' but does not provide a concrete link or explicit statement about the availability of the C implementation code.
Open Datasets Yes compare SGRRLD against SGLD on three large movie ratings datasets, namely the Movie Lens 1Million (ML-1M), Movie Lens 10Million (ML-10M), and Movie Lens 20Million (ML-20M) (grouplens.org).
Dataset Splits No The paper states, 'We randomly select 10% of the data as the test set and use the remaining data for generating the samples,' but does not explicitly specify a separate validation split or its proportion.
Hardware Specification Yes All our experiments are conducted on a standard laptop computer with 2.5GHz Quad-core Intel Core i7 CPU, and in all settings, the two chains of SGRRLD are run in parallel.
Software Dependencies No The paper states, 'We have implemented SGLD and SGRRLD in C by using the GNU Scientific Library for efficient matrix computations,' but does not specify version numbers for C or the GNU Scientific Library.
Experiment Setup Yes In our first experiment, we set d = 1, σ2 θ = 10, σ2 x = 1, N = 1000, and the size of each minibatch B = N/10. We fix the step size to γ = 10 3. In order to ensure that both algorithms are run for a fixed computation time, we run SGLD for K = 21000 iterations where we discard the first 1000 samples as burn-in, and we run SGRRLD for K = 10500 iterations accordingly, where we discard the samples generated in the first 500 iterations as burn-in.