Stochastic Proximal AUC Maximization
Authors: Yunwen Lei, Yiming Ying
JMLR 2021 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, we perform extensive experiments over various benchmark data sets from real-world application domains which show the superior performance of our algorithm over the existing AUC maximization algorithms. |
| Researcher Affiliation | Academia | Yunwen Lei EMAIL School of Computer Science University of Birmingham Birmingham B15 2TT, United Kingdom Yiming Ying EMAIL Department of Mathematics and Statistics State University of New York at Albany Albany, USA |
| Pseudocode | Yes | Algorithm 1: Stochastic Proximal AUC Maximization (SPAUC) |
| Open Source Code | No | The paper does not contain any explicit statement or link indicating that the source code for the described methodology is publicly available. |
| Open Datasets | Yes | The remaining UCI datasets can be downloaded from the LIBSVM webpage (Chang and Lin, 2011). |
| Dataset Splits | Yes | For each dataset, we use 80% of data for training and the remaining 20% for testing. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for running the experiments. |
| Software Dependencies | No | The paper does not list specific software dependencies with version numbers used for the experiments. |
| Experiment Setup | Yes | For SPAUC, SPAM and SOLAM, we consider step sizes of the form ηt = 2/(µt + 1) and validate the parameter µ over the interval 10{ 7, 6.5,..., 2.5}. Both SPAM and SPAUC with the ℓ1/ℓ2 regularizer require another regularization parameter to tune, which is validated over the interval 10{ 5, 4,...,0}. SOLAM involves the constraint on w, i.e. w belonging to ℓ2-ball with radius R in Rd, for which we tune over the interval 10{ 1,0,...,5}. For OAM gra, we need to tune a parameter to weight the comparison between released examples and bulk, which is validated over the interval 10{ 3, 2.5,...,1.5}. As recommended in Zhao et al. (2011), we fix the buffer size to 100. For OPAUC, we need to tune both the constant step size and the regularization parameter λ, which are validated over the interval 10{ 3.5, 3,...,1} and 10{ 5, 4,...,0}, respectively. The multi-stage scheme in FSAUC specifies how the step size decreases along the implementation of the algorithm and leave the initial step size as a free parameter to tune, which we validate over the interval 10{ 2.5, 2,...,2}. Furthermore, each iteration of FSAUC requires a projection onto an ℓ1-ball of radius of R, which we tune over the interval 10{ 1,0,...,5}. |