Double Trouble in Double Descent: Bias and Variance(s) in the Lazy Regime

Authors: Stéphane D’Ascoli, Maria Refinetti, Giulio Biroli, Florent Krzakala

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we present numerical experiments on a standard deep learning setup to show that our results are relevant to the lazy regime of deep neural networks.
Researcher Affiliation Academia 1Laboratoire de Physique de l Ecole normale sup erieure, ENS, Universit e PSL, CNRS, Sorbonne Universit e, Universit e de Paris, F-75005 Paris, France.
Pseudocode No The paper does not contain any pseudocode or algorithm blocks.
Open Source Code Yes The codes necessary to reproduce the results presented in this paper and obtain new ones are given at: https://github.com/mariaref/Random_Features.git
Open Datasets Yes We train a 5-layer fully-connected network on the CIFAR-10 dataset.
Dataset Splits No The paper mentions using the CIFAR-10 dataset and evaluates 'test error' but does not explicitly specify the training/validation/test dataset splits, such as percentages or sample counts for each split.
Hardware Specification No The paper does not specify any particular hardware (e.g., GPU model, CPU, memory) used for running the experiments. It only states that experiments were performed.
Software Dependencies No The paper mentions using
Experiment Setup Yes We train a 5-layer fully-connected network on the CIFAR-10 dataset. We keep only the first ten PCA components of the images, and divide the images in two classes according to the parity of the labels. We perform 105 steps of full-batch gradient descent with the Adam optimizer and a learning rate of 0.1, and scale the weights as prescribed in (Jacot et al., 2018).