Acceleration through spectral density estimation

Authors: Fabian Pedregosa, Damien Scieur

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

Reproducibility Variable Result LLM Response
Research Type Experimental Through empirical benchmarks on quadratic and logistic regression problems, we identify regimes in which the the proposed methods improve over classical (worst-case) accelerated methods. 7. Experiments We compare the proposed methods and classical accelerated methods on settings with varying degrees of mismatch with our assumptions. We first compare them on quadratics generated from a synthetic dataset, where the empirical spectral density is (approximately) a Marchenko Pastur distribution. We then compare these methods on another quadratic problem, generated using two non-synthetic datasets, where the MP assumption breaks down. Finally, we compare some the applicable methods in a logistic regression problem.
Researcher Affiliation Industry 1Google Research 2Samsung SAIT AI Lab, Montreal.
Pseudocode Yes Input: Initial guess x0, λ0 > 0 Algorithm: Iterate over t = 1... xt = xt 1+t 1 t + 1(xt 1 xt 2) λ0 t + 1 f(xt 1) Decaying Exponential Acceleration
Open Source Code No The paper does not contain any statement about releasing source code or provide links to a code repository.
Open Datasets Yes We use two UCI datasets: digits2 (n = 1797, d = 64), and breast cancer3 (n = 569, d = 32). 2https://archive.ics.uci.edu/ml/datasets/Optical+Recognition+ of+Handwritten+Digits 3https://archive.ics.uci.edu/ml/datasets/Breast+Cancer+ Wisconsin+%28Diagnostic%29
Dataset Splits No The paper mentions using datasets but does not provide specific details on training, validation, or test dataset splits, such as percentages, sample counts, or references to predefined splits.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, processor types, memory amounts) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers) needed to replicate the experiment.
Experiment Setup No The paper describes algorithms and their inputs but does not provide specific experimental setup details such as hyperparameters (learning rate, batch size, number of epochs) or specific training configurations used in the empirical evaluations.