Correcting Predictions for Approximate Bayesian Inference

Authors: Tomasz Kuśmierczyk, Joseph Sakaya, Arto Klami4511-4518

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the approach empirically in several problems, confirming its potential. ... We conduct a series of experiments, using variational approximation as q(θ). We first compare the method against the alternative of calibrating the posterior inference to account for the loss for a matrix factorization model, and then demonstrate improved decisions for a sparse regression model and a multilevel model for cases with approximations.
Researcher Affiliation Academia Tomasz Ku smierczyk, Joseph Sakaya, Arto Klami Helsinki Institute for Information Technology HIIT Department of Computer Science, University of Helsinki {tomasz.kusmierczyk, joseph.sakaya, arto.klami}@helsinki.fi
Pseudocode No The paper does not contain any pseudocode or clearly labeled algorithm blocks.
Open Source Code Yes The code to reproduce our experiments is available online.1 1https://github.com/tkusmierczyk/correcting approximate bayesian predictions
Open Datasets Yes on a subset of last.fm data (Bertin-Mahieux et al. 2011). ... We use the radon data and the multi-level model... (Gelman and Hill 2006). ... We apply the model on the corn data (Chen and Martin 2009).
Dataset Splits Yes we split the data randomly into equally sized training and test set. ... randomly split into equally sized training and test subsets.
Hardware Specification No The paper mentions 'computational resources' in the acknowledgements but does not specify any particular hardware components like CPU or GPU models, or memory details.
Software Dependencies No The paper mentions 'probabilistic programming tools, such as Stan (Carpenter et al. 2017) and Edward (Tran et al. 2016)' and that a model is 'implemented using the publicly available Stan code', but it does not provide specific version numbers for these or other software dependencies used in the experiments.
Experiment Setup Yes In our experiments, we use a simple feed-forward network with 3 hidden layers (with 20, 20 and 10 nodes) with Re LU activation and Adam optimizer with learning rate = 0.01... ...we use L = 5 in our empirical experiments. ... we use S = 1000 in other experiments. ... we use B = 20 quantiles.