Parabolic Approximation Line Search for DNNs

Authors: Maximus Mutschler, Andreas Zell

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

Reproducibility Variable Result LLM Response
Research Type Experimental We performed a comprehensive evaluation to analyze the performance of PAL on a variety of deep learning optimization tasks. Therefore, we tested PAL on commonly used architectures on CIFAR-10 [31], CIFAR-100 [31] and Image Net [13]. ... All in all, we trained over 4500 networks with Tensorflow 1.15 [1] on Nvidia Geforce GTX 1080 TI graphic cards.
Researcher Affiliation Academia Maximus Mutschler and Andreas Zell University of Tübingen Sand 1, D-72076 Tübingen, Germany {maximus.mutschler, andreas.zell}@uni-tuebingen.de
Pseudocode Yes Algorithm 1 The basic version of our proposed line search algorithm. See Section 4 for details. ... The full version of PAL including all additions is given in Appendix B Algorithm 2.
Open Source Code No The paper mentions using "TensorFlow 1.15" but does not state that the authors' own implementation code for PAL is open-source or provide a link to it.
Open Datasets Yes We focus on CIFAR-10, as it is extensively analyzed in optimization research for deep learning. However, on random samples of MNIST, CIFAR-100 and Image Net we observed the same results. ... We performed a comprehensive evaluation to analyze the performance of PAL on a variety of deep learning optimization tasks. Therefore, we tested PAL on commonly used architectures on CIFAR-10 [31], CIFAR-100 [31] and Image Net [13].
Dataset Splits No The paper mentions using validation accuracy but does not specify the explicit training, validation, and test dataset splits (e.g., percentages or exact counts) needed for reproduction.
Hardware Specification Yes All in all, we trained over 4500 networks with Tensorflow 1.15 [1] on Nvidia Geforce GTX 1080 TI graphic cards.
Software Dependencies Yes All in all, we trained over 4500 networks with Tensorflow 1.15 [1]...
Experiment Setup Yes A detailed explanation of the experiments including hyperparameters and data augmentations used are given in Appendix D.8.