Data Representations’ Study of Latent Image Manifolds
Authors: Ilya Kaufman, Omri Azencot
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experiments show consistent results over a variety of deep learning architectures and multiple data sets. Our code is publicly available at https: //github.com/azencot-group/CRLM |
| Researcher Affiliation | Academia | Ilya Kaufman 1 Omri Azencot 1 1Department of Computer Science, Ben-Gurion University of the Negev, Beer-Sheva, Israel. |
| Pseudocode | No | The paper describes the CAML algorithm from a cited work (Li, 2018) and details the estimation of the Hessian matrix using equations in Appendix E, but it does not present any pseudocode or algorithm blocks labeled "Algorithm" or "Pseudocode". |
| Open Source Code | Yes | Our code is publicly available at https: //github.com/azencot-group/CRLM |
| Open Datasets | Yes | Our evaluation focuses on convolutional neural network (CNN) architectures such as VGG (Simonyan & Zisserman, 2015) and Res Net (He et al., 2016) and on image classification benchmark datasets such as CIFAR-10 and CIFAR-100 (Krizhevsky et al., 2009). To verify the generality of our results we analyzed the behavior of curvature along the layers of Resnet models trained on Tiny Imagenet (Le & Yang, 2015) and Imagenet (Deng et al., 2009). |
| Dataset Splits | No | The paper mentions training on 'train and test sets' (Sec. 3) and evaluates on 'train sets' for NMAPC gap (Sec. 4.2), but it does not specify the use of a distinct 'validation set' or provide details about its split or size for reproducibility. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for the experiments, such as GPU models, CPU types, or memory specifications. |
| Software Dependencies | No | The paper describes the methods and algorithms used but does not explicitly list specific software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions, or specific libraries). |
| Experiment Setup | No | The paper mentions using a ResNet18 network trained with CIFAR-10 and regularization methods like 'weight decay of 5e 4 and cosine annealing learning rate scheduling' and 'mixup', but it does not provide a comprehensive set of hyperparameters (e.g., specific learning rates, batch sizes, number of epochs) or a dedicated section detailing the experimental setup for full reproducibility. |