Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1]
Project and Forget: Solving Large-Scale Metric Constrained Problems
Authors: Rishi Sonthalia, Anna C. Gilbert
JMLR 2022 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate that using our method, we can solve large problem instances of general weighted correlation clustering, metric nearness, information theoretic metric learning and quadratically regularized optimal transport; in each case, out-performing the state of the art methods with respect to CPU times and problem sizes. Keywords: Large Scale Convex Optimization, Metric Constrained Optimization, Metric Learning, Correlation Clustering |
| Researcher Affiliation | Academia | Rishi Sonthalia EMAIL Department of Mathematics University of Michigan Ann Arbor, Michigan, 48104 Anna C. Gilbert EMAIL Departments of Mathematics and Statistics & Data Science Yale University New Haven, Connecticut, 06510 |
| Pseudocode | Yes | Algorithm 1 General Algorithm. 1: function Project and Forget(f convex function) ... Algorithm 2 Finding Metric Violations. 1: function Metric Violations(d) ... Algorithm 3 Project and Forget algorithms. 1: function Project(x, z, L) |
| Open Source Code | Yes | All implementations and experiments can be found at https://github.com/rsonthal/ Project And Forget. |
| Open Datasets | Yes | For dense graphs, we use four graphs from the Stanford sparse network repository. Then, following Veldt et al. (2019), we use the method from Wang et al. (2013) to convert these graphs into instances of weighted correlation clustering on the complete graph. |
| Dataset Splits | Yes | For each data set, we uniformly at random choose 80% of the data points to be the training set and the remaining to be the test set. |
| Hardware Specification | Yes | All experiments were run on a Computer with 52 GB of memory. All computations were done on a machine with 4 physical cores, each with 13 GB of RAM. |
| Software Dependencies | Yes | The solvers CPLEX, Mosek, OSQP, SCS, COSMO, Ipopt, Prox SDP, ECOS, and MASS all were accessed via Julia s Ju MP library (Dunning et al., 2017). SLSQP was inferfaced with using Scipy. |
| Experiment Setup | Yes | The hyper-parameters were set as follows: γ = 1, u = 1, l = 10. |