ECLIPSE: An Extreme-Scale Linear Program Solver for Web-Applications

Authors: Kinjal Basu, Amol Ghoting, Rahul Mazumder, Yao Pan

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on real-world data show that our proposed LP solver, ECLIPSE, can solve problems with 1012 decision variables well beyond the capabilities of current solvers.
Researcher Affiliation Collaboration Kinjal Basu * 1 Amol Ghoting * 1 Rahul Mazumder * 2 Yao Pan * 1 *Equal contribution 1Linked In Corporation 2MIT. Correspondence to: Kinjal Basu <kbasu@linkedin.com>, Rahul Mazumder <rahulmaz@mit.edu>.
Pseudocode Yes Algorithm 1 ECLIPSE: Extreme-Scale LP Solver
Open Source Code No We will soon open-source the solver.
Open Datasets No We randomly simulate the vectors c, p and b, and solve the problem: ... Note that, since we could not find open-source example datasets that exactly fit the problem structure, we excluded comparisons with open-source LP instances.
Dataset Splits No The paper mentions 'convergence of our method as iterations increase' and 'relative duality gap as the stopping criteria' but does not explicitly provide training, validation, or test dataset splits.
Hardware Specification No The experiment was running in a development cluster in Spark 2.3 with up to 800 executors. The paper does not provide specific hardware details such as CPU/GPU models, memory, or processor types.
Software Dependencies Yes We built the system on top of Apache Spark (Zaharia et al., 2016) ... The experiment was running in a development cluster in Spark 2.3 with up to 800 executors.
Experiment Setup No The paper discusses concepts like the regularization parameter 'γ' and 'step-size' and mentions 'a constant step-size in our exposition,' but it does not provide concrete numerical values for these or other hyperparameters (e.g., learning rates, batch sizes, epochs) used in their experiments.