Convergence guarantees for a class of non-convex and non-smooth optimization problems
Authors: Koulik Khamaru, Martin Wainwright
ICML 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We illustrate our methods and theory via application to the problems of best subset selection, robust estimation and shape from shading reconstruction. ... In this section, we compare the performance of Algorithm 2 (prox-type method for short) with the convex-concave Procedure (CCCP). on best-subset selection problem described in problem 18. ... The performance comparison between the two algorithms is documented in the figure 1. |
| Researcher Affiliation | Academia | 1Department of Statistics, UC Berkeley, Berkeley, USA 2Department of EECS, UC Berkeley, Berkeley, USA. |
| Pseudocode | Yes | Algorithm 1 Subgradient type method ... Algorithm 2 Proximal type algorithm |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described. There are no links or explicit statements about code release. |
| Open Datasets | No | Synthetic data generation: We generated the rows of the n d matrix B from a d-dimensional Gaussian distribution with zero mean and an equicovariance matrix Σ satisfying Σii = 1 for all i, and Σij = 0.7 for all i = j. The regression vector x Rd (true value) was chosen to be a binary vector with sparsity s (s d). The location of the nonzero entries of the true regression vector x was chosen uniformly form {1, . . . , d}. |
| Dataset Splits | No | The paper describes synthetic data generation and performance comparison metrics, but it does not specify explicit training/validation/test dataset splits. The data is generated for each replication, not partitioned from a fixed dataset. |
| Hardware Specification | No | The paper discusses simulation performance but does not provide any specific hardware details such as CPU/GPU models or memory specifications used for running the experiments. |
| Software Dependencies | No | The paper mentions using 'proximal methods' for the inner CCCP optimization but does not provide specific software names with version numbers for any ancillary software dependencies. |
| Experiment Setup | Yes | For both algorithms, the tolerance level η was set to η = 10 8, maximum number of iterations was to be 1000, and both had same initializations. |