Robustness to Spurious Correlations via Human Annotations

Authors: Megha Srivastava, Tatsunori Hashimoto, Percy Liang

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

Reproducibility Variable Result LLM Response
Research Type Experimental Empirically, we show improvements of 5 10% on a digit recognition task confounded by rotation, and 1.5 5% on the task of analyzing NYPD Police Stops confounded by location.
Researcher Affiliation Academia 1Computer Science Department, Stanford University. Correspondence to: Megha Srivastava <megha@cs.stanford.edu>.
Pseudocode No The paper does not contain any pseudocode or clearly labeled algorithm blocks.
Open Source Code Yes Reproducibility We provide all source code, data, and experiments as part of a worksheet on the Coda Lab platform: https://bit.ly/uvdro-codalab.
Open Datasets Yes We evaluate the efficacy of UV-DRO on synthetic domain shifts on the MNIST digit classification task. [...] We consider the task of trying to detect false positives or police stops that do not result in arrests by training classifiers on data from police stops spanning 20032014 in New York City (NYCLU, 2019).
Dataset Splits Yes We tuned hyperparameters such as the learning rate, regularization, and DRO parameters using a held-out validation set, which we describe in the appendix.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies No The paper mentions the use of "Fast Text Sent2Vec library" but does not specify its version number. It also refers to optimization methods like "batch gradient descent with Ada Grad" but without corresponding software package versions.
Experiment Setup No The paper states: "We tuned hyperparameters such as the learning rate, regularization, and DRO parameters using a held-out validation set, which we describe in the appendix." While it indicates that hyperparameters were tuned and described elsewhere, it does not provide their specific values or detailed configuration settings in the main text.