Data-Efficient Augmentation for Training Neural Networks

Authors: Tian Yu Liu, Baharan Mirzasoleiman

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments demonstrate that our method achieves 6.3x speedup on CIFAR10 and 2.2x speedup on SVHN, and outperforms the baselines by up to 10% across various subset sizes.
Researcher Affiliation Academia Tian Yu Liu Department of Computer Science University of California, Los Angeles tianyu@cs.ucla.edu Baharan Mirzasoleiman Department of Computer Science University of California, Los Angeles baharan@cs.ucla.edu
Pseudocode Yes Algorithm 1 CORESETS FOR EFFICIENT DATA AUGMENTATION
Open Source Code Yes 1Our code can be found at https://github.com/tianyu139/data-efficient-augmentation
Open Datasets Yes Our experiments demonstrate that our method achieves 6.3x speedup on CIFAR10 and 2.2x speedup on SVHN... Similarly, on Tiny Image Net and Image Net... We demonstrate the effectiveness of our approach applied to CIFAR10 (Res Net20, Wide Res Net28-10), CIFAR10-IB (Res Net32), SVHN (Res Net32), noisy-CIFAR10 (Res Net20), Caltech256 (Res Net18, Res Net50), Tiny Image Net (Res Net50), and Image Net (Res Net50)...
Dataset Splits No The paper does not explicitly specify the training/test/validation dataset splits, or how a validation set was used for hyperparameter tuning or early stopping criteria.
Hardware Specification Yes For example, the state-of-the-art augmentation of [36], which appends every example with its highest-loss transformations, increases the training time of Res Net20 on CIFAR10 by 13x on an Nvidia A40 GPU (c.f. Sec. 6).
Software Dependencies No The paper mentions general tools and models (e.g., Res Net, Wide Res Net) but does not provide specific version numbers for software dependencies like Python, PyTorch, or TensorFlow.
Experiment Setup Yes For all methods, we select a new augmentation subset every R epochs. ...training on and augmenting 10% coresets yields 65.4% accuracy... ...Tiny Image Net and Image Net on Res Net50 (90 epochs, R = 15). ...R = 20.