Learning to Compose Domain-Specific Transformations for Data Augmentation

Authors: Alexander J. Ratner, Henry Ehrenberg, Zeshan Hussain, Jared Dunnmon, Christopher Ré

NeurIPS 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In our experiments, we show the efficacy of our approach on both image and text datasets, achieving improvements of 4.0 accuracy points on CIFAR-10, 1.4 F1 points on the ACE relation extraction task, and 3.4 accuracy points when using domain-specific transformation operations on a medical imaging dataset as compared to standard heuristic augmentation approaches.
Researcher Affiliation Academia Alexander J. Ratner , Henry R. Ehrenberg , Zeshan Hussain, Jared Dunnmon, Christopher Ré Stanford University {ajratner,henryre,zeshanmh,jdunnmon,chrismre}@cs.stanford.edu
Pseudocode No The paper does not include a section or figure explicitly labeled 'Pseudocode' or 'Algorithm', nor does it present any structured algorithm blocks.
Open Source Code Yes Our hope is that the proposed method will be of practical value to practitioners and of interest to researchers, so we have open-sourced the code at https://github.com/HazyResearch/tanda.
Open Datasets Yes We ran experiments on the MNIST [18] and CIFAR-10 [17] datasets, using only a subset of the class labels to train the end classification models and treating the rest as unlabeled data. We applied our approach to the Employment relation extraction subtask from the NIST Automatic Content Extraction (ACE) corpus [6]... Digital Database for Screening Mammography (DDSM) dataset [15, 4, 26]
Dataset Splits Yes We selected hyperparameters of the generator via performance on a validation set.
Hardware Specification No The paper does not provide specific details about the hardware used for experiments, suchs as CPU or GPU models, memory, or specific cloud infrastructure.
Software Dependencies No The paper does not provide specific version numbers for any software dependencies or libraries used in the experiments.
Experiment Setup Yes We select a fixed sequence length for each task via an initial calibration experiment (Fig. 5b). We use L = 5 for ACE, L = 7 for DDSM + DS, and L = 10 for all other tasks. We selected hyperparameters of the generator via performance on a validation set. We then used the trained generator to transform the entire training set at each epoch of end classification model training. For MNIST and DDSM we use a four-layer all-convolutional CNN, for CIFAR10 we use a 56-layer Res Net [14], and for ACE we use a bi-directional LSTM. Additionally, we incorporate a basic transformation regularization term as in [24] (see Supplemental Materials), and train for the last ten epochs without applying any transformations as in [11].