Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Robust, Accurate Stochastic Optimization for Variational Inference
Authors: Akash Kumar Dhaka, Alejandro Catalina, Michael R. Andersen, Måns Magnusson, Jonathan Huggins, Aki Vehtari
NeurIPS 2020 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We show empirically that the proposed framework works well on a diverse set of models... We now turn to validating our robust stochastic optimization algorithm for variational inference (summarized in Algorithm 1) through experiments on both simulated and real-world data. |
| Researcher Affiliation | Academia | Akash Kumar Dhaka Aalto University EMAIL Alejandro Catalina Aalto University EMAIL Michael Riis Andersen Technical University of Denmark EMAIL Måns Magnusson Uppsala University EMAIL Jonathan H. Huggins Boston University EMAIL Aki Vehtari Aalto University EMAIL |
| Pseudocode | Yes | Algorithm 1: Robust Stochastic Optimization for Variational Inference |
| Open Source Code | No | The paper mentions using third-party tools like viabel [23], TensorFlow Probability [9], Stan [4], and arviz [29], but does not state that the authors' own implementation code is open-source or provide a link to it. |
| Open Datasets | Yes | logistic regression [61] on three UCI datasets (Boston, Wine, and Concrete [10]); a high-dimensional hierarchical Gaussian model (Radon [34]), the 8-school hierarchical model [49], and a Bayesian neural network model... on the MNIST dataset [31] |
| Dataset Splits | No | The paper uses various datasets but does not explicitly provide details about specific train, validation, or test dataset splits (e.g., percentages, sample counts, or explicit standard split references) needed for reproduction. |
| Hardware Specification | No | The paper does not explicitly describe the specific hardware (e.g., GPU/CPU models, memory, or cloud instances) used to run the experiments. |
| Software Dependencies | No | The paper mentions using 'viabel', 'TensorFlow Probability', 'Stan', and 'arviz' but does not provide specific version numbers for these software components. |
| Experiment Setup | Yes | In our experiments we used η = 0.01, W = 100, a = 0.5, τ = 1.2, and e = 20. ... we used J = 1 in all of our experiments; the exception is that Fig. 2 used J = 4 ... We also put ELBO at an advantage by doing some tuning of the threshold ϵ, while keeping ϵ = 0.02 when using our MCSE criterion. |