Splitting Steepest Descent for Growing Neural Architectures

Authors: Lemeng Wu, Dilin Wang, Qiang Liu

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We test our method on both toy and realistic tasks, including learning interpretable neural networks, architecture search for image classification and energy-efficient keyword spotting. Table 1 shows the results on the Google speech commands benchmark dataset (Warden, 2018), in which our method achieves significantly higher accuracy than the best model (DS-CNN) found by Zhang et al. (2017), while having 31% less parameters and Flops.
Researcher Affiliation Academia Qiang Liu UT Austin lqiang@cs.utexas.edu Lemeng Wu * UT Austin lmwu@cs.utexas.edu Dilin Wang UT Austin dilin@cs.utexas.edu
Pseudocode Yes Algorithm 1 Splitting Steepest Descent for Optimizing Neural Architectures
Open Source Code No The paper does not provide any statement or link indicating that the source code for the described methodology is publicly available.
Open Datasets Yes We apply our method to split the prototype layer starting from a single neuron on MNIST... We experiment with two popular deep neural architectures, Mobile Net (Howard et al., 2017) and VGG19 (Simonyan & Zisserman, 2015)... on CIFAR-10. Table 1 shows the results on the Google speech commands benchmark dataset (Warden, 2018)...
Dataset Splits No The paper mentions 'train' and 'test' for evaluation, and 'validation' in the algorithm phases, but it does not provide specific dataset split percentages or sample counts for training, validation, and test sets in the main text.
Hardware Specification No The paper does not provide specific hardware details such as GPU models, CPU types, or memory used for running the experiments. It only mentions 'Google Cloud and Amazon Web Services (AWS)' in the acknowledgements.
Software Dependencies No The paper does not specify any software dependencies with version numbers (e.g., Python 3.8, PyTorch 1.9), which are necessary for reproducibility.
Experiment Setup No The paper states, 'Due to limited space, many of the detailed settings are shown in Appendix...', and refers to Appendix B.1, B.2, B.3, B.4 for specific experiment settings. These details are not present in the main body of the paper.