On the Inherent Regularization Effects of Noise Injection During Training

Authors: Oussama Dhifallah, Yue Lu

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical results corroborate our asymptotic predictions, showing that they are accurate even in moderate problem dimensions. Our theoretical predictions are based on a new correlated Gaussian equivalence conjecture that generalizes recent results in the study of random feature models.
Researcher Affiliation Academia Oussama Dhifallah 1 Yue M. Lu 1 1O. Dhifallah and Y. M. Lu are with the John A. Paulson School of Engineering and Applied Sciences, Harvard University, Cambridge, MA 02138, USA..
Pseudocode No No pseudocode or algorithm blocks were found in the paper.
Open Source Code No The paper does not provide any explicit statement or link indicating that the source code for the described methodology is publicly available.
Open Datasets Yes In Figure 6(a), we consider the Semeion Handwritten Digit Data Set downloaded from the Machine Learning Repository .
Dataset Splits No The paper does not explicitly provide training/test/validation dataset splits, specific percentages, or detailed splitting methodologies, although it discusses training and generalization errors.
Hardware Specification No The paper mentions numerical simulations and experiments but does not provide any specific details about the hardware (e.g., GPU/CPU models, memory) used to run them.
Software Dependencies No The paper does not provide specific software dependencies (e.g., library names with version numbers) needed to replicate the experiments.
Experiment Setup Yes Figure 1. Solid line: Theoretical predictions. Circle: Numerical simulations for (3). Black cross: Numerical simulations for (6). ϕ( ) is the sign function with probability θ of flipping the sign. bϕ( ) and σ( ) are the sign function. We set p = 500, = 0.5, α = n/p = 2, θ = 0.1, λ = 10 5. F has independent Gaussian components with zero mean and variance 1/p.