Feature learning in deep classifiers through Intermediate Neural Collapse

Authors: Akshay Rangamani, Marius Lindegaard, Tomer Galanti, Tomaso A Poggio

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this paper, we conduct an empirical study of the feature learning process in deep classifiers.
Researcher Affiliation Academia 1Center for Brains, Minds, and Machines, Massachusetts Institute of Technology. Correspondence to: Akshay Rangamani <arangam@mit.edu>.
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes The code for running our experiments is available at https://github.com/ mariuslindegaard/Intermediate_Neural_ Collapse/tree/ICML2023
Open Datasets Yes We consider the MNIST (Le Cun et al., 1998), Fashion MNIST (Xiao et al., 2017), CIFAR10 (Krizhevsky & Hinton, 2009), and SVHN (Netzer et al., 2011) datasets.
Dataset Splits No The paper mentions 'batch-size of 128' and '350 training epochs' but does not specify explicit training, validation, or test dataset splits.
Hardware Specification Yes All experiments were conducted on a cluster with NVIDIA Tesla V100, Ge Force GTX 1080 TI, and A100 GPUs.
Software Dependencies No The paper does not list specific software dependencies with version numbers (e.g., Python 3.8, PyTorch 1.9, CUDA 11.1) to ensure reproducibility.
Experiment Setup Yes This includes a logarithmic learning rate sweep between 0.0001 and 0.25 decayed twice by a factor of 0.1,a momentum of 0.9, a weight decay of 5e-4, for 350 training epochs, and batch-size of 128.