The Implicit Regularization of Stochastic Gradient Flow for Least Squares

Authors: Alnur Ali, Edgar Dobriban, Ryan Tibshirani

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

Reproducibility Variable Result LLM Response
Research Type Experimental Section 5 gives numerical examples supporting our theory. We generated the data matrix according to X = Σ1/2W, where the entries of W were i.i.d. following a normal distribution. ... Figure 4 plots the risk of ridge regression, discrete-time SGD (2), and Theorem 2. ... averaged over 30 draws of y (the underlying coefficients were drawn from a normal distribution, and scaled so the signalto-noise ratio was roughly 1).
Researcher Affiliation Academia Alnur Ali 1 Edgar Dobriban 2 Ryan J. Tibshirani 3 1Stanford University 2University of Pennsylvania 3Carnegie Mellon University.
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide concrete access to source code for the methodology described in this paper.
Open Datasets No We generated the data matrix according to X = Σ1/2W, where the entries of W were i.i.d. following a normal distribution. We allow for correlations between the features, setting the diagonal entries of the predictor covariance Σ to 1, and the off-diagonals to 0.5.
Dataset Splits No The paper does not provide specific dataset split information needed to reproduce the data partitioning. It only mentions the total data size (n=100, p=500).
Hardware Specification No The paper does not provide specific hardware details used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details needed to replicate the experiment.
Experiment Setup Yes We set ϵ = 2.2548e-4, following Lemma 5. ... Below, we present results for n = 100, p = 500, and m = 20.