Data Augmentation as Feature Manipulation

Authors: Ruoqi Shen, Sebastien Bubeck, Suriya Gunasekar

ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our main contribution is a detailed analysis of data augmentation on the learning dynamic for a two layer convolutional neural network in the recently proposed multi-view data model by Allen-Zhu & Li (2020b). We complement this analysis with further experimental evidence that data augmentation can be viewed as feature manipulation.
Researcher Affiliation Collaboration 1University of Washington. Part of this work was done as a intern at Microsoft Research. 2Microsoft Research.
Pseudocode No The paper describes algorithms and derivations using mathematical formulas and prose, but it does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not contain any explicit statements about releasing source code for the methodology described, nor does it provide a link to a code repository.
Open Datasets Yes We complement our analysis with experiments on CIFAR-10 and synthetic datasets, where we study data augmentation in more generality.
Dataset Splits No The paper mentions 'training examples' and 'test dataset' (e.g., 'full CIFAR-10 dataset which has 50000 training examples for 10 classes' and 'We use the standard CIFAR-10 test dataset') but does not specify exact training/validation/test split percentages or counts for reproducibility.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU models, CPU types, memory) used to run the experiments.
Software Dependencies No The paper mentions training networks (ResNet20) and using SGD, but it does not specify any software dependencies with version numbers (e.g., PyTorch, TensorFlow versions, or specific library versions).
Experiment Setup Yes In all configurations, we train a Res Net20 network using SGD for 120 epochs with momentum 0.9, weight decay 0.005, and learning rate starting at 0.1 and annealed to (0.01, 0.001) at epochs (40, 80).