Mirrored Langevin Dynamics

Authors: Ya-Ping Hsieh, Ali Kavis, Paul Rolland, Volkan Cevher

NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we report promising experimental results for LDA on real datasets.
Researcher Affiliation Academia Laboratory for Information and Inference Systems (LIONS), EPFL, Lausanne, Switzerland {ya-ping.hsieh, ali.kavis, paul.rolland, volkan.cevher}@epfl.ch
Pseudocode Yes Algorithm 1 Stochastic Mirrored Langevin Dynamics (SMLD)
Open Source Code No The paper does not provide concrete access to source code (specific repository link, explicit code release statement, or code in supplementary materials) for the methodology described.
Open Datasets Yes We implement the SMLD for LDA on the Wikipedia corpus with 100,000 documents, and we compare the performance against the SGRLD [26].
Dataset Splits No The paper does not provide specific dataset split information for training, validation, and test sets. It mentions a "test data" but no explicit validation set or overall split percentages.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper mentions "python, np.exp" but does not specify version numbers for Python or any libraries, which is required for reproducibility.
Experiment Setup Yes We implement the deterministic MLD for sampling from an 11-dimensional Dirichlet posterior (3.6) with n1 = 10,000, n2 = n3 = 10, and n4 = n5 = = n11 = 0, which aims to capture the sparse nature of real observations in topic modeling. We set αℓ= 0.1 for all ℓ. We perform a grid search over the constant step-size for both algorithms, and we keep the best three for MLD and SGRLD. For each iteration, we build an empirical distribution by running 2,000,000 independent trials... In order to keep the comparison fair, we adopt exactly the same setting as in [26], including the model parameters, the batch-size, the Gibbs sampler steps, etc.