Reparameterization Gradient for Non-differentiable Models

Authors: Wonyeol Lee, Hangyeol Yu, Hongseok Yang

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

Reproducibility Variable Result LLM Response
Research Type Experimental We experimentally compare our gradient estimator (OURS) to the score estimator (SCORE), an unbiased gradient estimator that is applicable to non-differentiable models, and the reparameterization estimator (REPARAM), a biased gradient estimator that computes only Rep Gradθ (discussed in Section 3). We evaluate our estimator on three models for small sequential data: temperature [33] models..., textmsg [1] is a model..., influenza [32] is a model... Table 1 compares the average variance of each estimator for N = 1... Figure 1 shows the ELBO objective, for different estimators with different N s, as a function of the iteration number.
Researcher Affiliation Academia Wonyeol Lee Hangyeol Yu Hongseok Yang School of Computing, KAIST Daejeon, South Korea {wonyeol, yhk1344, hongseok.yang}@kaist.ac.kr
Pseudocode No The paper includes mathematical derivations and theorems but does not contain explicitly labeled 'Pseudocode' or 'Algorithm' blocks.
Open Source Code Yes Code is available at https://github.com/wonyeol/reparam-nondiff.
Open Datasets Yes temperature [33] models the random dynamics of a controller..., textmsg [1] is a model for the numbers of per-day SNS messages over the period of 74 days..., influenza [32] is a model for the US influenza mortality data in 1969.
Dataset Splits No The paper uses benchmark models but does not explicitly provide details about train/validation/test dataset splits, specific percentages, or sample counts.
Hardware Specification No The paper does not provide specific details regarding the hardware (e.g., CPU, GPU models, memory) used to run the experiments.
Software Dependencies No The paper mentions that the implementation is 'written in Python and uses autograd [18]' but does not specify version numbers for these software components.
Experiment Setup Yes We optimize the ELBO objective using Adam [11] with two stepsizes: 0.001 and 0.01. We run Adam for 10000 iterations and at each iteration, we compute each estimator using N {1, 8, 16} Monte Carlo samples.