Abstraction Mechanisms Predict Generalization in Deep Neural Networks

Authors: Alex Gain, Hava Siegelmann

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

Reproducibility Variable Result LLM Response
Research Type Experimental The CNA is highly predictive of generalization ability, outperforming norm-and-sharpness-based generalization metrics on an extensive evaluation of close to 200 network instances comprising a breadth of dataset-architecture combinations, especially in cases where additive noise is present and/or training labels are corrupted.
Researcher Affiliation Academia 1Department of Computer Science, The Johns Hopkins University, Baltimore, MD 21218, USA 2School of Computer and Information Sciences, University of Massachusetts Amherst, Amherst, MA 01003, USA.
Pseudocode No The paper provides a structured definition for the CNA in Figure 2, but it is presented as a definition overview rather than a formally labeled pseudocode or algorithm block.
Open Source Code Yes 1An implementation of this paper can be found on Git Hub: https://github.com/alexgain/cna-icml2020
Open Datasets Yes The datasets include Image Net-32, CIFAR-10, CIFAR-100, MNIST, Fashion-MNIST, SVHN, corrupted labels counterparts (i.e. the same datasets with varyling levels training labels shuffled), and a random noise dataset.
Dataset Splits No The paper mentions 'validation datapoints' in the context of defining the margin metric, but it does not specify concrete train/validation/test splits (percentages, counts, or explicit standard split usage) for its experiments needed for reproduction.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments, such as GPU or CPU models, memory specifications, or cloud instance types.
Software Dependencies No The paper does not list any specific software dependencies or their version numbers (e.g., Python, PyTorch, TensorFlow versions) that would be needed for replication.
Experiment Setup No The paper mentions that 'All training details are included in the supplement,' implying that specific hyperparameters, optimizers, or other detailed setup configurations are not present in the main text.