Faster Projection-free Convex Optimization over the Spectrahedron
Authors: Dan Garber, Dan Garber
NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate our method, along with other conditional gradient variants, on the task of matrix completion [13]. Figure 1: Comparison between conditional gradient variants for solving the matrix completion problem on the MOVIELENS100K (left) and MOVIELENS1M (right) datasets. |
| Researcher Affiliation | Academia | Dan Garber Toyota Technological Institute at Chicago dgarber@ttic.edu |
| Pseudocode | Yes | Algorithm 1 Conditional Gradient; Algorithm 2 Randomized Rank one-regularized Conditional Gradient |
| Open Source Code | No | The paper does not provide any statements about releasing open-source code or links to a code repository for the described methodology. |
| Open Datasets | Yes | We have experimented with two well known datasets for the matrix completion task: the MOVIELENS100K dataset for which d1 = 943, d2 = 1682, n = 105, and the MOVIELENS1M dataset for which d1 = 6040, d2 = 3952, n 106. |
| Dataset Splits | No | The paper mentions the datasets used and their dimensions, but does not provide specific train/validation/test dataset splits, percentages, or methodology for data partitioning. |
| Hardware Specification | No | The paper does not provide any specific hardware details such as GPU models, CPU models, or cloud computing resources used for running the experiments. |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers (e.g., Python 3.8, PyTorch 1.9) needed to replicate the experiment. |
| Experiment Setup | Yes | We have set the parameter in Problem (12) to = 10000 for the ML100K dataset, and = 35000 for the ML1M dataset. First, on each iteration t, instead of picking an index it of a rank-one matrix... we choose it in a greedy way... Second, after computing the eigenvector vt using the step-size t = 1/t... we apply a line-search, as detailed in [13], in order to the determine the optimal step-size given the direction vtv>t. |