Towards Learning Convolutions from Scratch

Authors: Behnam Neyshabur

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

Reproducibility Variable Result LLM Response
Research Type Experimental We present the following empirical findings for β-LASSO (Section 4): β-LASSO achieves state-of-the-art results on training fully connected networks on CIFAR10, CIFAR-100 and SVHN tasks.
Researcher Affiliation Industry Behnam Neyshabur Google neyshabur@google.com
Pseudocode Yes Algorithm 1 β-LASSO
Open Source Code No No explicit statement about open-sourcing the code for the described methodology or a link to a repository is provided.
Open Datasets Yes β-LASSO achieves state-of-the-art results on training fully connected networks on CIFAR10, CIFAR-100 and SVHN tasks.
Dataset Splits No The paper mentions training on CIFAR-10, CIFAR-100, and SVHN datasets and reports 'test accuracy', but it does not specify the exact percentages or counts for training, validation, and test splits. It refers to an 'Appendix for details of the training procedure', but these details are not present in the main text.
Hardware Specification No Figure 1's caption states: 'The dotted black line is the maximum model size for training using 16-bits on a V100 GPU.' This mentions a specific GPU model in the context of a theoretical constraint on model size, not as an explicit statement of the hardware actually used for their experiments.
Software Dependencies No The paper mentions various algorithms and optimizers (e.g., SGD, β-LASSO, Adam/RMSProp) but does not list specific software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions) used for their implementation.
Experiment Setup Yes For each experiment, the results for training 400 and 4000 epochs are reported. We fix β in the experiments but tune the regularization parameter λ for each datasets.