Will Bilevel Optimizers Benefit from Loops
Authors: Kaiyi Ji, Mingrui Liu, Yingbin Liang, Lei Ying
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 6 Empirical Verification Experiments on hyperparameter optimization on MNIST. We first conduct experiments to verify our theoretical results in Corollaries 1, 2, 3 and 4 on AID-Bi O with different implementations. We consider the following hyperparameter optimization problem. Figure 1: Training & test losses v.s. time (seconds) by AID-Bi O on MNIST with different Q and N. |
| Researcher Affiliation | Academia | Kaiyi Ji Department of CSE University at Buffalo kaiyiji@buffalo.edu Mingrui Liu Department of CS George Mason University mingruil@gmu.edu Yingbin Liang Department of ECE The Ohio State University liang.889@osu.edu Lei Ying Department of EECS University of Michigan leiying@umich.edu |
| Pseudocode | Yes | Algorithm 1 AID-based bilevel optimization (AID-Bi O) with double warm starts... Algorithm 2 ITD-based bilevel optimization algorithm (ITD-Bi O) with warm start |
| Open Source Code | Yes | Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes] |
| Open Datasets | Yes | Experiments on hyperparameter optimization on MNIST... The datasets we use are public. See Appendix E for details. |
| Dataset Splits | No | The paper mentions 'Dval' for validation dataset and 'Dtr' for training dataset in Section 6, but does not provide specific split percentages or sample counts for these datasets in the main text. |
| Hardware Specification | No | The main text of the paper does not provide specific details about the hardware used for running experiments (e.g., GPU models, CPU types, or memory). |
| Software Dependencies | No | The main text of the paper does not specify software dependencies with version numbers (e.g., specific libraries or frameworks like Python, PyTorch, or TensorFlow versions). |
| Experiment Setup | Yes | In Figure 2, ... we choose loop sizes Q and N from {1, 50}. In Figure 3, ... with different choices of N from {1, 20}. For ITD-Bi O, we choose N = 20 for N-N-loop ITD and N = 1 for No-loop ITD. The results are reported with the best-tuned hyperparameters. |