Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Neural Collapse for Cross-entropy Class-Imbalanced Learning with Unconstrained ReLU Features Model
Authors: Hien Dang, Tho Tran Huu, Tan Minh Nguyen, Nhat Ho
ICML 2024 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We empirically validate our results through experiments on practical architectures and dataset. |
| Researcher Affiliation | Collaboration | 1Department of Statistics and Data Sciences, University of Texas at Austin, USA 2FPT Software AI Center, Vietnam 3Department of Mathematics, National University of Singapore, Singapore. |
| Pseudocode | No | The paper describes the methodology in mathematical and textual form, but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not contain any explicit statement about releasing source code for the methodology described, nor does it provide a link to a code repository. |
| Open Datasets | Yes | We train these models on the imbalanced subsets of 4 datasets: MNIST, Fashion MNIST, CIFAR10, and CIFAR100. |
| Dataset Splits | No | For this experiment, a subset of the CIFAR10 dataset with {1000, 1000, 2000, 2000, 3000, 3000, 4000, 4000, 5000, 5000} random samples per class is utilized as training data. The paper specifies training data sizes but does not provide details on how the dataset was split into training, validation, and test sets. |
| Hardware Specification | No | The paper mentions using specific model architectures (MLP, VGG11, ResNet18) but does not provide any details regarding the hardware specifications (e.g., GPU/CPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper mentions the use of Adam optimizer and ReLU activation, but it does not list any specific software or library versions (e.g., Python, PyTorch, TensorFlow, CUDA) that would be needed for reproducibility. |
| Experiment Setup | Yes | We train each backbone model with Adam optimizer with batch size 256, the weight decay is λW = 1e-4. Feature decay λH is set to 1e-5 for MLP and VGG11, and to 1e-4 for ResNet18. |