Boosting Variational Inference With Locally Adaptive Step-Sizes
Authors: Gideon Dresdner, Saurav Shekhar, Fabian Pedregosa, Francesco Locatello, Gunnar Rätsch
IJCAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We present empirical evidence demonstrating that our method is both faster and more robust than existing methods. The adaptive methods also yield more parsimonious variational approximations than previous techniques. (Section 1 - Introduction) |
| Researcher Affiliation | Collaboration | Gideon Dresdner1 , Saurav Shekhar1 , Fabian Pedregosa2 , Francesco Locatello1 , Gunnar R atsch1 1 Dept. for Computer Science, ETH Zurich, Universit atsstrasse 6, 8092 Zurich, Switzerland 2 Google Research dgideon@ethz.ch |
| Pseudocode | Yes | Algorithm 1 Template for Boosting VI algorithms and Algorithm 2 Find step-size with approximate backtracking |
| Open Source Code | Yes | Source code to reproduce our experiments is available here: https://github.com/ratschlab/adaptive-stepsize-boosting-bbvi |
| Open Datasets | Yes | For this task we used the CHEMREACT1 dataset which contains 26,733 chemicals, each with 100 features. (...) For this task we used the EICU COLLABORATIVE RESEARCH database [Goldberger et al., 2000]. (...) 1http://komarix.org/ac/ds/ |
| Dataset Splits | No | The paper mentions using training and test data and evaluating across 10 replicates, but does not provide specific split percentages or sample counts for train/validation/test sets. |
| 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 various software components and techniques (e.g., 'black-box VI solvers', 'gradient descent', 'Python') but does not provide specific version numbers for any of them. |
| Experiment Setup | No | The paper mentions the number of iterations (50 for CHEMREACT, 40 for EICU) and replicates (10), and the base family (Laplace distributions), but does not provide specific details on hyperparameters like learning rates, batch sizes, or optimizer settings. |