Applied Online Algorithms with Heterogeneous Predictors

Authors: Jessica Maghakian, Russell Lee, Mohammad Hajiesmaili, Jian Li, Ramesh Sitaraman, Zhenhua Liu

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Theoretical results are supplemented by large-scale empirical evaluations with production data demonstrating the success of our methods on optimization problems occurring in large distributed computing systems.
Researcher Affiliation Collaboration 1Department of Applied Mathematics & Statistics, Stony Brook University, Stony Brook, NY, USA 2Manning College of Information and Computer Sciences, UMass Amherst, Amherst, MA, USA 3Department of Electrical and Computer Engineering, SUNY Binghamton, Binghamton, NY, USA 4Akamai Tech, Cambridge, MA, USA.
Pseudocode Yes Algorithm 1 par SBO, Algorithm 2 in DBO, Algorithm 3 Ro BO, Algorithm 4 Greedy Subroutine, Algorithm 5 DBO
Open Source Code No The paper provides a link to a dataset and related experiments, but not a general statement or link for the open-source code of the described methodologies (par SBO, in DBO, Ro BO).
Open Datasets Yes Dataset 2: The Impact of COVID-19 Lockdowns... We digitized publicly available records of public peering traffic at Milan s internet exchange to recover demand data... Dataset 2 and related experiments are available at https://github.com/jmaghakian/covid_exp
Dataset Splits Yes The data spans one month at five minute resolution and we use two weeks for training predictors and two weeks for testing.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts, or detailed computer specifications) used for running experiments.
Software Dependencies No The paper mentions 'sktime' and 'Auto ARIMA forecaster' but does not provide specific version numbers for these software dependencies or other key libraries.
Experiment Setup Yes To generate the parameter predictions, we use linear regression with price setting, metro-area location and size as the dependent variables. For the input predictions, we predicted day-ahead windows using the Auto ARIMA forecaster available through sktime. The classifier was trained using the same features as the parameter predictions, with the best choice of DBO, par SBO or in DBO on historical data as labels for the training dataset.