Efficient Gradient-Based Inference through Transformations between Bayes Nets and Neural Nets

Authors: Diederik Kingma, Max Welling

ICML 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Theoretical results are supported by experiments. (Abstract), Experimental results are shown in section 6. (Section 1.1), We applied a Hybrid Monte Carlo (HMC) sampler to a Dynamic Bayesian Network (DBN)... (Section 6.1), trained on a small (1000 datapoints) and large (50000 datapoints) version of the MNIST dataset. (Section 6.2)
Researcher Affiliation Academia Diederik P. Kingma D.P.KINGMA@UVA.NL Max Welling M.WELLING@UVA.NL Machine Learning Group, University of Amsterdam
Pseudocode No No pseudocode or algorithm blocks found.
Open Source Code No No explicit statement or link for open-source code for the described methodology was found.
Open Datasets Yes The model was trained on a small (1000 datapoints) and large (50000 datapoints) version of the MNIST dataset. (Section 6.2)
Dataset Splits No The paper mentions training on MNIST but does not provide specific details on train/validation/test splits, percentages, or sample counts for these splits.
Hardware Specification No No specific hardware details (e.g., GPU/CPU models, memory, or cloud instance types) used for running the experiments are provided.
Software Dependencies No No specific software dependencies with version numbers (e.g., library names, frameworks, or solvers with their versions) are provided.
Experiment Setup Yes For HMC we used 10 leapfrog steps per sample, and the stepsize was automatically adjusted while sampling to obtain a HMC acceptance rate of around 0.9. At each sampling run, the first 1000 HMC samples were thrown away (burn-in); the subsequent 4000 HMC samples were kept. (Section 6.1), For MCEM, we used HMC with 10 leapfrog steps followed by a weight update using Adagrad (Duchi et al., 2010). For MMCL, we used L {10, 100, 500}. (Section 6.2)