Almost surely constrained convex optimization
Authors: Olivier Fercoq, Ahmet Alacaoglu, Ion Necoara, Volkan Cevher
ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct numerical experiments on basis pursuit, hard margin support vector machines and portfolio optimization problems and show that our algorithm achieves state-of-the-art practical performance. |
| Researcher Affiliation | Academia | 1LTCI, Télécom Paris Tech, Université Paris-Saclay 2Laboratory for Information and Inference Systems, École Polytechnique Fédérale de Lausanne 3Department of Automatic Control and Systems Engineering, University Politehnica Bucharest. |
| Pseudocode | Yes | Algorithm 1 SASC |
| Open Source Code | No | The paper does not provide an explicit statement about releasing source code for the described methodology or a direct link to a code repository. |
| Open Datasets | Yes | We present numerical experiments on a basis pursuit problem on synthetic data, a hard margin SVM problem on the kdd2010, rcv1, news20 datasets from (Chang and Lin, 2011) and a portfolio optimization problem on NYSE, DJIA, SP500, TSE datasets from (Borodin et al., 2004). |
| Dataset Splits | Yes | kdd2010 raw version (bridge to algebra) with 19, 264, 997 training examples, 748, 401 testing examples and 1, 163, 024 features, rcv1.binary with 20, 242 training examples, 677, 399 testing examples and 47, 236 features. For the last dataset, news20.binary , since there was not a dedicated testing dataset, we randomly split examples for training and testing with 17.996 training examples, 2, 000 testing examples and 1, 355, 191 features. |
| Hardware Specification | No | The paper does not provide any specific hardware details such as GPU/CPU models or processor types used for running its experiments. |
| Software Dependencies | No | The paper mentions using 'cvx (Grant et al., 2008)' and 'libsvm database (Chang and Lin, 2011)' but does not specify version numbers for these software components. |
| Experiment Setup | Yes | We used the parameters µ = 10 5 for SPP, m0 = 2, ω = 2, α0 = 10 2 a1b1 , where a1 is the first measurement and b1 is the corresponding result. We take n = 105 and make two passes over the data. We run SASC with the parameters α0 = 1, ω = 1.2, m0 = 2 and Case 1 in Algorithm 1. For SASC, we use α0 = 1/2, ω = 2 in all experiments and use the parameter choices in Case 2 in Algorithm 1 due to strong convexity in the objective. |