Implicit Semantic Data Augmentation for Deep Networks
Authors: Yulin Wang, Xuran Pan, Shiji Song, Hong Zhang, Gao Huang, Cheng Wu
NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive empirical evaluations on several competitive image classification benchmarks show that ISDA consistently improves the generalization performance of popular deep networks, especially with little training data and powerful traditional augmentation techniques. |
| Researcher Affiliation | Collaboration | Yulin Wang1 Xuran Pan1 Shiji Song1 Hong Zhang2 Cheng Wu1 Gao Huang1 1Department of Automation, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology (BNRist), 2Baidu Inc., China |
| Pseudocode | Yes | Algorithm 1 The ISDA Algorithm. |
| Open Source Code | Yes | Code for reproducing our results is available at https://github.com/blackfeatherwang/ISDA-for-Deep-Networks. |
| Open Datasets | Yes | We use three image recognition benchmarks in the experiments. (1) The two CIFAR datasets consist of 32x32 colored natural images in 10 classes for CIFAR-10 and 100 classes for CIFAR-100, with 50,000 images for training and 10,000 images for testing, respectively. (2) Image Net is a 1,000-class dataset from ILSVRC2012[29], providing 1.2 million images for training and 50,000 images for validation. |
| Dataset Splits | Yes | In our experiments, we hold out 5000 images from the training set as the validation set to search for the hyper-parameter λ0. |
| Hardware Specification | No | On Image Net, due to GPU memory limitation, we approximate the covariance matrices by their diagonals, i.e., the variance of each dimension of the features. |
| Software Dependencies | No | The paper refers to common deep learning architectures and optimization algorithms (SGD), but does not explicitly list specific software dependencies (e.g., libraries, frameworks) with version numbers required for reproduction. |
| Experiment Setup | Yes | The hyper-parameter λ0 for ISDA is selected from the set {0.1, 0.25, 0.5, 0.75, 1} according to the performance on the validation set. On Image Net, due to GPU memory limitation, we approximate the covariance matrices by their diagonals, i.e., the variance of each dimension of the features. The best hyper-parameter λ0 is selected from {1, 2.5, 5, 7.5, 10}. |