On the Spectral Bias of Convolutional Neural Tangent and Gaussian Process Kernels
Authors: Amnon Geifman, Meirav Galun, David Jacobs, Basri Ronen
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Real networks: We further tested our predictions on two real networks, a CNN and a fully connected architecture. We trained the networks to fit various SH-products in which the higher frequencies (k = 1, ..., 4) are positioned at either two neighboring or two distant pixels and measured the number of iterations to convergence. Figure 4(right) shows that, consistent with our predictions, the CNN learned each function faster than the FC network. |
| Researcher Affiliation | Academia | 1Department of Computer Science, Weizmann Institute of Science, Rehovot, Israel 2Department of Computer Science, University of Maryland, College Park, MD |
| Pseudocode | No | The paper provides mathematical formulas for the network architecture (Figure 1) but does not include structured pseudocode or algorithm blocks. |
| Open Source Code | No | Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [N/A] |
| Open Datasets | No | The paper describes properties of inputs like 'multi-channel input' or 'inputs drawn from the uniform distribution on either the sphere or the hypercube' for theoretical analysis and numerical evaluation, but does not provide details or access information for a publicly available or open dataset used for training real networks. |
| Dataset Splits | No | The ethics review checklist explicitly states 'Did you specify all the training details (e.g., data splits, hyperparameters, how they were chosen)? [N/A]'. |
| Hardware Specification | No | The paper does not provide specific details on the hardware used for experiments, such as GPU or CPU models. The ethics review checklist states '[N/A]' for 'total amount of compute and the type of resources used'. |
| Software Dependencies | No | The paper does not specify any software dependencies with version numbers used for the experiments. |
| Experiment Setup | No | The paper states 'We trained the networks to fit various SH-products' in the 'Real networks' section, but does not provide specific experimental setup details such as hyperparameters, optimizer settings, or training schedules. The ethics review checklist confirms '[N/A]' for 'all the training details (e.g., data splits, hyperparameters, how they were chosen)'. |