Asymptotics of Ridge Regression in Convolutional Models

Authors: Mojtaba Sahraee-Ardakan, Tung Mai, Anup Rao, Ryan A. Rossi, Sundeep Rangan, Alyson K Fletcher

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

Reproducibility Variable Result LLM Response
Research Type Experimental In this section we validate our theoretical results on simulated data. We generate data using a ground truth convolutional model of the form (3). We use i.i.d. complex normal convolution kernel and noise with different variances. For the data matrix X, we consider two different models: i) i.i.d. complex normal data; and ii) a non-Gaussian autoregressive process of order 1 (an AR(1) process). In both cases we take T = 256, ny = 500 and use different values of nx to create plots of estimation error with respect to δ = ny/nx.
Researcher Affiliation Collaboration 1Department of Electrical and Computer Engineering, University of California, Los Angeles 2Department of Statistics, University of California, Los Angeles 3Adobe Research 4Department of Electrical and Computer Engineering, New York University.
Pseudocode No The paper does not contain any pseudocode or clearly labeled algorithm blocks.
Open Source Code No The paper does not provide any statement or link indicating that the source code for the methodology is openly available.
Open Datasets No The paper states:
Dataset Splits No The paper validates its theoretical results through simulations. It does not mention using standard training/validation/test splits of a publicly available dataset, nor does it define such splits for its simulated data. It describes parameters for generating data for comparison with theoretical predictions.
Hardware Specification No The paper does not provide any specific details about the hardware used for running the experiments.
Software Dependencies No The paper does not provide specific version numbers for any software dependencies used in the experiments.
Experiment Setup Yes We generate data using a ground truth convolutional model of the form (3). We use i.i.d. complex normal convolution kernel and noise with different variances. For the data matrix X, we consider two different models: i) i.i.d. complex normal data; and ii) a non-Gaussian autoregressive process of order 1 (an AR(1) process). In both cases we take T = 256, ny = 500 and use different values of nx to create plots of estimation error with respect to δ = ny/nx. ... The parameter a controls how fast the process is mixing. ... To show this, we use both a Gaussian AR(1) process with var(ξ2 t ) = 0.1 as well as ξt unif({ s, s}) with s = 0.1 to match the variances. In both cases we take a = 0.9 and measurement noise variance σ2 = 0.1. ... In this case the variance of signal and noise are 0.004 and 1 respectively. Figure 1 shows the log of normalized estimation error with respect to δ = ny/nx for three different values of the regularization parameter λ.