Nonlinear Behaviour of Critical Points for a Simple Neural Network
Authors: Gerrit Welper
TMLR 2024 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 6 Numerical Experiments. This section contains some preliminary numerical experiments. We train the target function f(x) = 1 |x|0.1, x [ 1, 1], which has a cusp at the origin... Figure 3 shows the evolution of the networks and breakpoints during gradient descent training with the following setup. |
| Researcher Affiliation | Academia | G. Welper EMAIL Department of Mathematics University of Central Florida Orlando, FL 32816, USA |
| Pseudocode | No | The paper describes mathematical derivations and proofs (e.g., in Sections 3, 4, and A) but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not contain any explicit statement about providing source code or a link to a code repository for the methodology described. |
| Open Datasets | No | The paper describes using a synthetic target function, f(x) = 1 |x|0.1, for its numerical experiments, rather than a publicly available dataset with concrete access information (link, DOI, or repository). |
| Dataset Splits | No | The paper mentions generating "1280 samples in 10 batches" for its numerical experiments but does not specify any training, validation, or test dataset splits. |
| Hardware Specification | No | The paper details the setup for numerical experiments, including network architecture, width, learning rate, samples, and epochs, but does not provide any specific hardware details such as GPU or CPU models. |
| Software Dependencies | No | The paper specifies training parameters like "Network architecture (9)", "Network width: 16", "Learning rate: 0.01", and "100000 epochs", but does not list any specific software libraries or their version numbers. |
| Experiment Setup | Yes | Network architecture (9). Network width: 16. Learning rate: 0.01. 1280 samples in 10 batches. 100000 epochs, with a plot of the network every 20000 epochs. |