Post-training Iterative Hierarchical Data Augmentation for Deep Networks

Authors: Adil Khan, Khadija Fraz

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

Reproducibility Variable Result LLM Response
Research Type Experimental 4 Experiments, This section presents the results of the empirical validation of IHDA on three image classification benchmarks: CIFAR-10, CIFAR-100 [1], and Image Net [32].
Researcher Affiliation Academia Adil Khan Khadija Fraz Institute of Data Science and Artificial Intelligence Innopolis University Universitetskaya St, 1, Innopolis, Russia, 420500 a.khan@innopolis.ru, k.fraz@innopolis.university
Pseudocode Yes Algorithm 1: The algorithm to compute potential of a point p hl (X), Algorithm 2: The IHDA Algorithm
Open Source Code No The paper does not include an unambiguous statement about releasing source code for the described methodology, nor does it provide a direct link to a code repository.
Open Datasets Yes CIFAR-10, CIFAR-100 [1], and Image Net [32]., Sussex-Huawei Locomotion-Transportation (SHL) challeneg dataset [33]
Dataset Splits Yes For CIFAR datasets, the validation set had 5000 images, which were taken from the training set. For Image Net, we used its reduced subset, which was created by randomly choosing 150 classes and 50,000 samples. From this reduced subset, we held out 5000 images for the validation set to tune the hyperparameters.
Hardware Specification No The paper does not specify any hardware details such as GPU models, CPU types, or cloud computing resources used for running the experiments.
Software Dependencies No The paper does not provide specific software dependencies or library versions, only mentioning general techniques like backpropagation and models like VAEs.
Experiment Setup Yes In all experiments, the spread of RBF γ in Algorithm 1 was set to 0.05. For other hyper-parameters (including p, and w), we held out a part of the training dataset as the validation set to find their optimum values. The hyper-parameter p and w, for each experiment, are selected from the interval [0, 1], with a step size of 0.05, based on the performance on the validation set.