Topological Convolutional Layers for Deep Learning
Authors: Ephy R. Love, Benjamin Filippenko, Vasileios Maroulas, Gunnar Carlsson
JMLR 2023 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We perform experiments on image and video data. The results show significant improvement in TCNNs compared to conventional CNNs with respect to various metrics. |
| Researcher Affiliation | Academia | Ephy R. Love EMAIL Bredesen Center DSE University of Tennessee Knoxville, TN 37996, USA; Benjamin Filippenko EMAIL Department of Mathematics Stanford University Stanford, CA 94305, USA; Vasileios Maroulas EMAIL Department of Mathematics University of Tennessee Knoxville, TN 37996, USA; Gunnar Carlsson EMAIL Department of Mathematics Stanford University Stanford, CA 94305, USA |
| Pseudocode | Yes | Appendix B. Pseudocode |
| Open Source Code | No | The paper contains pseudocode (Algorithms 1 and 2), but there is no explicit statement about releasing source code for the described methodology, nor are there any links to a code repository. |
| Open Datasets | Yes | We perform digit classification on 3 datasets: MNIST Le Cun et al. (1998), SVHN Netzer et al. (2011) and USPS Hull (1994). ... We also use two collections of labeled images of cats and dogs: Cats vs Dogs Kaggle (2013) (which we call Kaggle), and the cat and dog labeled images from CIFAR-10, see Krizhevsky (2012). ... We use three datasets of videos: UCF-101 Soomro et al. (November, 2012), KTH Schuldt et al. (2004), and Weizmann Gorelick et al. (2007). URLs are provided for all datasets. |
| Dataset Splits | Yes | For type (1) [fixed dataset], the train/test splits are as follows: Dataset Train Test MNIST 85% 15% SVHN 80% 20% USPS 80% 20%. For type (2) [generalization], we train and test on the entire datasets. ... The UCF-101 split is as follows: Dataset Train Test UCF-101 67% 33%. |
| Hardware Specification | Yes | Computing was performed on an AMD 2990WX CPU with 128GB RAM and an NVIDIA RTX 2080TI GPU. |
| Software Dependencies | No | The paper mentions using ReLU and softmax activation functions, and the Adam optimizer. It also implicitly refers to Pytorch (via Unbox AI). However, no specific version numbers for any software dependencies (e.g., programming language, libraries, or frameworks) are provided. |
| Experiment Setup | Yes | We choose metaparameters for each experiment... The following table lists various metaparameters by Figure number: Figure Conv-layers Conv-slices Kernel size LR Batch size Epochs... All experiments use 2 fully connected layers following a flattening layer. Our video experiments differ from image experiments in that we use kernels which are not square (cubic). To indicate the size of a 3-dimensional kernel, we use the notation (t, y, x)... |