A Universal Primal-Dual Convex Optimization Framework

Authors: Alp Yurtsever, Quoc Tran Dinh, Volkan Cevher

NeurIPS 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental This section illustrates the scalability and the flexibility of our primal-dual framework using some applications in the quantum tomography (QT) and the matrix completion (MC). Figure 1: The convergence behavior of algorithms for the q 14 qubits QT problem. Figure 2: The performance of the algorithms for the MC problems.
Researcher Affiliation Academia Alp Yurtsever: Quoc Tran-Dinh; Volkan Cevher: Laboratory for Information and Inference Systems, EPFL, Switzerland {alp.yurtsever, volkan.cevher}@epfl.ch ; Department of Statistics and Operations Research, UNC, USA quoctd@email.unc.edu
Pseudocode Yes Algorithm 1 (Universal Primal-Dual Gradient Method p Uni PDGradq) and Algorithm 2 (Accelerated Universal Primal-Dual Gradient Method p Acc Uni PDGradq)
Open Source Code No The paper mentions using existing functions like MATLAB's eigs and PROPACK's lansvd but does not state that the authors' own implementation code for their framework is open-source or available.
Open Datasets Yes We apply our algorithms to (20) and (21) using the Movie Lens 100K dataset.
Dataset Splits No The paper mentions "test and training data partition" for MovieLens but does not specify exact percentages or sample counts for these splits, nor does it mention a validation split.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions "MATLAB’s eigs function" and "lansvd function (MATLAB version) from PROPACK [21]" but does not specify version numbers for MATLAB or the functions.
Experiment Setup Yes We set ϵ 2 ˆ 10 4 for our methods and have a wall-time 2ˆ104s in order to stop the algorithms. we set the target accuracy ϵ 10 3, and we choose the tuning parameter κ 9975{2 as in [20]. We use lansvd function (MATLAB version) from PROPACK [21] to compute the top singular vectors, and a simple implementation of the power method to find the top singular value in the line-search, both with 10 5 relative error tolerance.