Advances in Black-Box VI: Normalizing Flows, Importance Weighting, and Optimization
Authors: Abhinav Agrawal, Daniel R. Sheldon, Justin Domke
NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate components relating to optimization, flows, and Monte-Carlo methods on a benchmark of 30 models from the Stan model library. |
| Researcher Affiliation | Academia | 1College of Information and Computer Sciences, University of Massachusetts Amherst 2Department of Computer Science, Mount Holyoke College {aagrawal, sheldon, domke}@cs.umass.edu |
| Pseudocode | No | The paper describes algorithms and procedures in text but does not include any formally labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper mentions using existing tools like Autograd and Stan but does not provide a link or explicit statement about releasing the source code for their own methodology or implementation. |
| Open Datasets | Yes | We evaluate each method using a benchmark of 30 models from the Stan Model library [35, 36]. |
| Dataset Splits | No | The paper discusses evaluation using 10,000 fresh samples and mentions ADVI's step-size selection based on ELBO after 200 iterations, but it does not specify a distinct validation set with percentages or counts for hyperparameter tuning or model selection in a general sense separate from the final evaluation. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU/CPU models or memory amounts used for running experiments. |
| Software Dependencies | No | The paper mentions software like "Autograd, a Python automatic differentiation library [22]" and "Stan, a state-of-the-art probabilistic programming framework [4]" but does not specify their version numbers or other required software dependencies with versions. |
| Experiment Setup | Yes | During optimization, all methods have the same computational budget, measured as 100 "oracle evaluations" of the log p per iteration, and are optimized for 30,000 iterations. |