Accelerating deep neural network learning using data stream methodology

This dataset is a reproducibility package accompanying the article Approximate Importance-Based Sampling for Neural Network Training and contains the standalone code, saved experimental results, metric trajectories, and generated figures needed to reproduce the publication-relevant experiments. It supports rerunning and verifying three groups of MNIST-based neural network training experiments: the effect of mini-batch size, the effect of drift-detector parameters, and comparison with the previous method. The package includes scripts to regenerate all experiments and save trajectory files, scripts to recreate figures from those saved results, and copies of the publication figures already generated during the original study.