Global Convergence of Stochastic Gradient Descent for Some Non-convex Matrix Problems
Authors: Christopher De Sa, Christopher Re, Kunle Olukotun
ICML 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We also show experiments to illustrate the runtime and convergence of the algorithm. ... 5. Experiments |
| Researcher Affiliation | Academia | Christopher De Sa CDESA@STANFORD.EDU Department of Electrical Engineering, Stanford University, Stanford, CA 94309, Kunle Olukotun KUNLE@STANFORD.EDU Department of Electrical Engineering, Stanford University, Stanford, CA 94309, Christopher R e CHRISMRE@STANFORD.EDU Department of Computer Science, Stanford University, Stanford, CA 94309 |
| Pseudocode | Yes | Algorithm 1 Alecton: Solve stochastic matrix problem ... Algorithm 2 Alecton One-at-a-time |
| Open Source Code | No | The paper does not provide concrete access to source code. There are no explicit statements about code release, repository links, or mentions of code being available in supplementary materials. |
| Open Datasets | Yes | Figure 2(f) demonstrates convergence results on real data from the Netflix Prize problem (Funk, 2006). ... Funk, Simon. Netflix Update: Try this at Home. 2006. |
| Dataset Splits | No | The paper states that for the Netflix Prize problem, it uses 'a training dataset containing 110,198,805 revealed entries.' However, it does not provide specific details on how this dataset was further split into training, validation, and test sets for their own experiments (e.g., percentages, sample counts, or specific predefined split citations). |
| Hardware Specification | Yes | Experiments ran on a single twelve-core machine (Intel Xeon E5-2697, 2.70GHz) with 256 GB of shared memory. |
| Software Dependencies | No | The paper does not provide specific ancillary software details with version numbers (e.g., libraries, frameworks, or programming languages with their exact versions) that would be needed to replicate the experiment. |
| Experiment Setup | Yes | Figure 2(a) illustrates the convergence of Alecton with p = q = 1 using three sampling distributions on datasets with n = 104. (a) Angular convergence of three distributions on a synthetic dataset with η = 10 5. ... Figure 2(b) ... as the step size parameter η is varied... (figure shows η = 1e-8, 3e-8, 1e-7) ... using 107 iterations for each run of Alecton. |