Convex Geometry of ReLU-layers, Injectivity on the Ball and Local Reconstruction
Authors: Daniel Haider, Martin Ehler, Peter Balazs
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Section 5 demonstrates how the method can be used to analyze the injectivity behavior of a Re LU-layer in numerical experiments. and We train a neural network with one Re LU-layer and a soft-max output layer on the Iris data set (Fisher, 1936). |
| Researcher Affiliation | Collaboration | Daniel Haider 1 2 Martin Ehler 2 Peter Balazs 1 1Acoustics Research Institute, Vienna, Austria 2University of Vienna, Department of Mathematics, Austria. |
| Pseudocode | Yes | Algorithm 1 PBE for Br and Algorithm 2 Reconstruction via Facets |
| Open Source Code | Yes | Our implementations of the algorithms are publicly available under https://github.com/danedane-haider/Alpha-rectifyingframes. |
| Open Datasets | Yes | We train a neural network with one Re LU-layer and a soft-max output layer on the Iris data set (Fisher, 1936). |
| Dataset Splits | No | The paper mentions 'validation set' and reports metrics on it, but does not specify the exact split percentages or sample counts for training, validation, and test sets, nor does it cite a standard split for the Iris dataset. Cross entropy loss on the validation set. After normalization to zero mean and a variance of one, all data samples lie within the ball of radius r = 3.1. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., GPU model, CPU type) used for running the experiments. |
| Software Dependencies | No | The paper mentions 'Polymake' and 'prefer 'lrs';' for implementation, but does not provide specific version numbers for these or any other software dependencies, libraries, or frameworks used in the experiments. |
| Experiment Setup | Yes | optimization of a cross-entropy loss is done using stochastic gradient descent at a learning rate of 0.5 for 100 epochs. |