Fast Projection Onto Convex Smooth Constraints

Authors: Ilnura Usmanova, Maryam Kamgarpour, Andreas Krause, Kfir Levy

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 4. EXPERIMENTAL EVALUATION We first demonstrate the performance of our approach on synthetic problems of projection onto a randomly generated quadratic set and onto their intersection. ... We compare our approach with the Interior Point Method (IPM) from the MOSEK solver, as well as with SLSQP from the scipy.optimize.minimize package.
Researcher Affiliation Academia Ilnura Usmanova 1 Maryam Kamgarpour 2 Andreas Krause 3 Kfir Yehuda Levy 4 5 1Automatic Control Laboratory, D-ITET, ETH Z urich, Switzerland 2Department of Electrical and Computer Engineering, University of British Columbia, Canada 3Department of Computer Science, ETH Z urich, Switzerland 4A Viterby fellow 5Department of Electrical & Computer Engineering, Technion Israel Institute of Technology.
Pseudocode Yes Algorithm 1 Cutting Plane Method with Approximate Oracles; Algorithm 2 Oapproximate gradient/value oracles for d(); Algorithm 3 Accelerated Gradient Descent (AGD) (Nesterov, 1998); Algorithm 4 Fast Projection Method
Open Source Code No The paper does not provide any concrete access information (specific repository link, explicit code release statement, or code in supplementary materials) for the methodology described.
Open Datasets Yes We use the doc-rna dataset (Uzilov et al., 2006) from LIBSVM with n = 4000, 10000, 11000 data points and compare the results and the running time with the IPM.
Dataset Splits No The paper does not provide specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology) needed to reproduce the data partitioning.
Hardware Specification Yes The experiments were run on a machine with Intel Core i77700K, 64Gb RAM.
Software Dependencies No The paper mentions "MOSEK solver", "SLSQP from the scipy.optimize.minimize package", and "LIBSVM" but does not specify version numbers for these or other software dependencies.
Experiment Setup Yes For Algorithm 2, to solve the primal subproblems we use the AGD method as described before. We select the smoothness parameter L is based on the norms of the matrices Ai, and tune the parameter R empirically using the doubling trick.The run-times are shown in Table 4.1. The accuracy is fixed to 10 4.