Frequentist Uncertainty in Recurrent Neural Networks via Blockwise Influence Functions

Authors: Ahmed Alaa, Mihaela Van Der Schaar

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

Reproducibility Variable Result LLM Response
Research Type Experimental Using data from a critical care setting, we demonstrate the utility of uncertainty quantification in sequential decision-making.
Researcher Affiliation Academia 1University of California, Los Angeles 2Cambridge University.
Pseudocode Yes Algorithm 1 BJ-based uncertainty estimation in RNNs
Open Source Code No The paper does not provide an explicit statement or link for open-sourcing the code.
Open Datasets Yes We conducted experiments on data from the Medical Information Mart for Intensive Care (MIMIC-III) (Johnson et al., 2016) database
Dataset Splits No We generated 1000 training sequences using the model in (9) with T = 10 time steps and trained a vanilla RNN model to predict the label yt on the basis of the input sequence x1:t.
Hardware Specification No The paper does not provide specific details about the hardware used for running experiments.
Software Dependencies No The paper discusses various RNN architectures (simple RNN, LSTMs, GRUs) and models but does not provide specific software dependencies with version numbers.
Experiment Setup Yes To ensure a fair comparison, we fix the hyperparameters of the underlying RNNs in all baselines to the following: Number of layers: 1, Number of hidden units: 20, Learning rate: 0.01, Batch size: 150, Number of epochs: 10, Number of steps: 1000.