Ensembles of Locally Independent Prediction Models

Authors: Andrew Ross, Weiwei Pan, Leo Celi, Finale Doshi-Velez5527-5536

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

Reproducibility Variable Result LLM Response
Research Type Experimental Across a variety of synthetic and real-world tasks, we find that our method improves generalization and diversity in qualitatively novel ways, especially under data limits and covariate shift.
Researcher Affiliation Academia Andrew Slavin Ross,1 Weiwei Pan,1 Leo Anthony Celi,2 Finale Doshi-Velez1 1Paulson School of Engineering and Applied Sciences, Harvard University, Cambridge, MA 02138, USA 2Massachusetts Institute of Technology, Cambridge, MA, 02139, USA
Pseudocode No The paper includes mathematical formulations and descriptions of procedures, but it does not contain any explicitly labeled 'Pseudocode' or 'Algorithm' blocks.
Open Source Code Yes Code to replicate all experiments is available at https://github.com/dtak/lit.
Open Datasets Yes Next, we test our method on several standard binary classification datasets from the UCI and MOA repositories (Lichman 2013; Bifet et al. 2010). These are mushroom, ionosphere, sonar, spectf, and electricity...
Dataset Splits Yes For all datasets, we randomly select 80% of the dataset for training and 20% for test, then take an additional 20% split of the training set to use for validation.
Hardware Specification No The paper states that models were 'trained in TensorFlow' but does not provide any specific details about the hardware used, such as GPU/CPU models, memory, or cloud computing resources.
Software Dependencies No The paper mentions training models in 'TensorFlow' but does not specify a version number for TensorFlow or any other software dependencies, libraries, or packages used in the experiments.
Experiment Setup Yes For the experiments that follow, we use 256-unit single hidden layer fully connected neural networks with rectifier activations, trained in Tensorflow with Adam. For the real-data experiments, we use dropout and L2 weight decay with a penalty of 0.0001.